engineering problem solving in the age of ai

Forget problem-solving. In the age of AI, it's problem-finding that counts

In the age of AI, the most successful people will be those who can identify the problems that AI is best placed to solve. Image:  Shutterstock/Baranq

.chakra .wef-1c7l3mo{-webkit-transition:all 0.15s ease-out;transition:all 0.15s ease-out;cursor:pointer;-webkit-text-decoration:none;text-decoration:none;outline:none;color:inherit;}.chakra .wef-1c7l3mo:hover,.chakra .wef-1c7l3mo[data-hover]{-webkit-text-decoration:underline;text-decoration:underline;}.chakra .wef-1c7l3mo:focus,.chakra .wef-1c7l3mo[data-focus]{box-shadow:0 0 0 3px rgba(168,203,251,0.5);} Ravi Kumar S.

.chakra .wef-9dduvl{margin-top:16px;margin-bottom:16px;line-height:1.388;font-size:1.25rem;}@media screen and (min-width:56.5rem){.chakra .wef-9dduvl{font-size:1.125rem;}} Explore and monitor how .chakra .wef-15eoq1r{margin-top:16px;margin-bottom:16px;line-height:1.388;font-size:1.25rem;color:#F7DB5E;}@media screen and (min-width:56.5rem){.chakra .wef-15eoq1r{font-size:1.125rem;}} Artificial Intelligence is affecting economies, industries and global issues

.chakra .wef-1nk5u5d{margin-top:16px;margin-bottom:16px;line-height:1.388;color:#2846F8;font-size:1.25rem;}@media screen and (min-width:56.5rem){.chakra .wef-1nk5u5d{font-size:1.125rem;}} Get involved with our crowdsourced digital platform to deliver impact at scale

Stay up to date:, society and equity.

Listen to the article

Have you read?

Navigating ai: what are the top concerns of chief information security officers, can we embrace the ai revolution and journey towards a brighter future, don't miss any update on this topic.

Create a free account and access your personalized content collection with our latest publications and analyses.

License and Republishing

World Economic Forum articles may be republished in accordance with the Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International Public License, and in accordance with our Terms of Use.

The views expressed in this article are those of the author alone and not the World Economic Forum.

Related topics:

The agenda .chakra .wef-n7bacu{margin-top:16px;margin-bottom:16px;line-height:1.388;font-weight:400;} weekly.

A weekly update of the most important issues driving the global agenda

.chakra .wef-1dtnjt5{display:-webkit-box;display:-webkit-flex;display:-ms-flexbox;display:flex;-webkit-align-items:center;-webkit-box-align:center;-ms-flex-align:center;align-items:center;-webkit-flex-wrap:wrap;-ms-flex-wrap:wrap;flex-wrap:wrap;} More on Society and Equity .chakra .wef-nr1rr4{display:-webkit-inline-box;display:-webkit-inline-flex;display:-ms-inline-flexbox;display:inline-flex;white-space:normal;vertical-align:middle;text-transform:uppercase;font-size:0.75rem;border-radius:0.25rem;font-weight:700;-webkit-align-items:center;-webkit-box-align:center;-ms-flex-align:center;align-items:center;line-height:1.2;-webkit-letter-spacing:1.25px;-moz-letter-spacing:1.25px;-ms-letter-spacing:1.25px;letter-spacing:1.25px;background:none;padding:0px;color:#B3B3B3;-webkit-box-decoration-break:clone;box-decoration-break:clone;-webkit-box-decoration-break:clone;}@media screen and (min-width:37.5rem){.chakra .wef-nr1rr4{font-size:0.875rem;}}@media screen and (min-width:56.5rem){.chakra .wef-nr1rr4{font-size:1rem;}} See all

Countries with more women in government are more prosperous

Silvana Koch-Mehrin and Dominik Weh

November 15, 2023

Governing Smart Cities: Use Cases for Urban Transformation

Generative AI holds great potential for those with disabilities - but it needs policy to shape it

Yonah Welker

November 3, 2023

The finance gap for women entrepreneurs is $1.7 trillion. Here’s how to close it

Mark Muckerheide

October 26, 2023

How poor oral health impacts health equity 

Maria Carvajal

October 23, 2023

12 top leaders share how they are redefining leadership to embrace inclusivity

Christa Odinga-Svanteson and Kulé Galma

• Data Science
• Engineering
• Entrepreneurship
• Technology Insider
• Manufacturing
• MIT Bootcamps
• MIT Open Learning
• MITx MicroMasters Programs
• Online Education
• Professional Development
• Quantum Computing

  View All Posts

By: MIT xPRO on November 3rd, 2021 4 Minute Read

Print/Save as PDF

Why Aren’t More Organizations Realizing the Potential of Machine Learning?

Machine Learning

As organizations increasingly invest in artificial intelligence (AI) initiatives, most are not achieving their goals. Gartner estimates that through 2023, only 50% of organizations will take their AI projects past proof of concept.

We asked MIT Professor Youssef Marzouk what factors might be holding companies back from faster and broader machine learning adoption and impact. Marzouk is co-director of the MIT Center for Computational Engineering and the director of MIT’s Aerospace Computational Design Laboratory. He is also one of the lead faculty for the MIT xPRO online program Machine Learning, Modeling, and Simulation: Engineering Problem-Solving in the Age of AI .

Along with cost and organizational inertia, Marzouk believes that another major hurdle is limited AI knowledge and skill. According to a 2021 Genpact and MIT Sloan CIO Symposium survey of more than 500 CIOs, 49% say they don’t have sufficient talent inside of their companies and are relying on external providers to help with hiring employees who have experience with AI and cloud systems. While hiring to fill the AI capabilities gap is important, Professor Marzouk sees a distinct advantage in also building AI knowledge and skills among your current staff.

“ AI is not fairy dust. .. to realize its potential, we have to help people understand what machine learning is and what it isn’t, and how they can best apply it to their business. It’s time to extend this knowledge beyond the data scientists and de-mystify machine learning for the whole organization.” -- Professor Youssef Marzouk, Director of MIT’s Aerospace Computational Design Laboratory, Lead Instructor of MIT xPRO's online program, Machine Learning, Modeling, and Simulation: Engineering Problem-Solving in the Age of AI

“AI is not fairy dust,” says Marzouk. “We’ve already seen that the thoughtful application of machine learning can lead to positive business outcomes. But to realize its potential, we have to help people understand what machine learning is and what it isn’t , and how they can best apply it to their business. It’s time to extend this knowledge beyond the data scientists and de-mystify machine learning for the whole organization.”

Here are three ways that Professor Marzouk recommends for organizations that want to better prepare your workforce for scaling and optimizing AI and machine learning initiatives.

Get Your Organization Fluent with AI and Machine Learning Vocabulary

Ideas for how to leverage AI tools can come from any function or level of the organization, if they understand what’s possible. Understanding of these tools shouldn’t live only in technical groups. Instead, everyone should have a baseline understanding of the possibilities and potential of AI.

The average non-technical professional might not be able talk in detail with a data scientist about the parameters of a machine learning pilot. Having professionals across the organizations learn a common language will open up the meaningful dialogue and collaboration required for successful, holistic, and sustainable outcomes.

Develop AI Capability in Your Domain Experts and Bring Data Scientists Closer to the Business

Applying AI for impact requires a level of sophistication around how to apply the tools for positive business impact. This is only possible when you marry industry and domain knowledge with AI understanding. Marzouk suggests that real progress occurs when people understand a bit of both.

Providing on-the-job training for AI skills to business or domain experts can result in stronger business cases for action and can also help teams flag issues early on so projects are more likely to succeed.

According to Marzouk, “When it comes to machine learning, you want your team to have an appreciation and a healthy skepticism for the technology. You can’t blindly trust every result. The conclusions reached need to be interpreted by someone that has domain expertise.”

Domain experts can raise questions such as:

• How trustworthy are the results of our analysis?
• Do the findings align with the technical knowledge and data we already have?
• What is it going to cost to get results?

Build Out Your Machine Learning Toolbox with Industry Best Practices

When initiating an AI or machine learning project, starting small can make an impact. Data collection, curation, communication and interpretation are key steps for organizations looking to move their AI projects past proof of concept.

“It’s hard to move the ship and challenge the status quo, so companies leading with AI are starting with small exercises and pilots,” says Marzouk. “Once small projects work, they are translated to other applications and move into broad production.”

Typically, getting the data set ready takes more time than the actual application of machine learning. For that reason, developing and maintaining best practices for cleaning the data and feeding the models is as important as the model itself. Moreover, teams who adopt AI and machine learning tools will also need crucial non-technical skills to interpret and communicate the results, to allow for quick cycle refinement and action.

“Developing more informed users and customers for AI tools across the organization is what creates a positive “push and pull” for this technology -- the back and forth dialogue between the business and the AI leaders required to make the best use of these tools,” says Marzouk.

Whether you’re an individual contributor or an executive, a domain expert or a technical professional, understanding the key concepts and applications of AI and machine learning through online workforce training is the first step toward pushing your projects past ideation and toward meaningful impact.

If you or your organization are ready to get started or refine your next AI and machine learning initiative, join Professor Marzouk for MIT xPRO’s upcoming online program, Machine Learning, Modeling, and Simulation: Engineering Problem-Solving in the Age of AI.

• More about MIT xPRO
• About this Site
• Terms of Service
• Privacy Policy

• From the Directors
• Mission & History
• Advisory Council
• Newsletters & Brochure
• Corporate Responsibility Initiative
• Corporations, Government and Public Policy
• Digital Assets Policy Project
• Education Policy Program
• Financial Sector Program
• GrowthPolicy
• Harvard Electricity Policy Group
• Harvard Environmental Economics Program
• Harvard Kennedy School Healthcare Policy Program
• Harvard Project on Climate Agreements
• Kansai Keizai Doyukai Program
• Regulatory Policy Program
• Rising Chinese Economic Power
• Sustainability Science Program
• Seminars & Conferences
• Annual Robert Glauber Lecture
• Working Papers and Reports
• Funding & Prizes
• Other Opportunities

Design in the Age of Artificial Intelligence

In this section.

• Marco Iansiti

2020, Paper: "Artificial Intelligence (AI) is affecting the scenario in which innovation takes place. What are the implications for our understanding of design? Is AI just another digital technology that, akin to many others, will not significantly question what we know about design? Or will it create transformations in design that our current frameworks cannot capture? To address these questions, we have investigated two pioneering cases at the frontier of AI, Netflix and AirBnB (complemented with analyses in Microsoft and Tesla), which offer a privileged window on the future evolution of design. We found that AI does not undermine the basic principles of Design Thinking (people-centered, abductive and iterative). Rather, it enables to overcome past limitations (in scale, scope and learning) of human intense design processes. In the context of AI factories solutions may even be more user-centered (to an extreme level of granularity, i.e. being designed for every single person), more creative, and continuously updated through learning iterations that span the entire life cycle of a product. Yet, we found that AI profoundly changes the practice of design. Problem solving tasks, traditionally carried on by designers, are now automated into learning loops that operate without limitations of volume and speed. These loops think in a radically different way than a designer: they address complex problems through very simple tasks, iterated exponentially. The article therefore proposes a new framework for understanding design practice in the age of AI. We also discuss the implications for design and innovation theory. Specifically, we observe that, as creative problem solving is significantly conducted by algorithms, human design increasingly becomes an activity of sense making, i.e. to understand which problems make sense to be addressed. This shift in focus calls for new theories and brings design closer to leadership, which is, inherently, an activity of sense making." 

Non-HKS Author Website -  Marco Iansiti  

Related Publications

Tech & regulation study group: how do we oversee the digital world, techlash: who makes the rules in the digital gilded age, ten ways ai will change democracy.

Machine Learning, Modeling, and Simulation Principles

Description.

Course 1 of 2 in the program Machine Learning, Modeling, and Simulation: Engineering Problem-Solving in the Age of AI

Start Date(s)

Certificate

Course Certificate of Completion

Program for this course

• Machine Learning, Modeling, and Simulation: Engineering Problem-Solving in the Age of AI

Add this chart to Favorites

• Join the Revolution
• Job Listings

Machine Learning, Modeling, and Simulation: Engineering Problem-Solving in the Age of AI

• Certifications
• Certified AI Transformation Leader (CAITL ™ )
• Certified Artificial Intelligence Scientist (CAIS ™ )
• Certified Artificial Intelligence Consultant (CAIC ™ )
• Certified Artificial Intelligence Engineer (CAIE ™ )
• Certified Artificial Intelligence Prefect (CAIP ™ )
• Certified Artificial Intelligence Prefect – Advanced ( ™ CAIPa)
• MIT xPRO Programs
• Speedster Program
• Maintain Certification
• Flexi Payments
• Examination Policy
• About USAII ®
• Why USAII ®
• Board of Advisors
• USAII ® Partners
• Key Insights
• Terms of use
• Privacy Policy
• Report Misconduct
• Email : [email protected]

• Create USAII ® Account
• myControlPanel

Machine Learning for new age AI

This two-course program from MIT helps in demystifying Machine Learning through computational engineering principles and applications.

What you will learn

Understand and Learn how the computational tools used in engineering problem-solving are put into practice in this 2-course program.

Simulate physical processes using numerical discretization methods.

Assess cost-accuracy trade-offs in numerical simulation., learn powerful optimization techniques and understand their fundamental role in machine learning., describe canonical machine learning problems from a statistical perspective., practice real-world forecasting and risk assessment problems using monte carlo simulation., understand why and how machine learning methods may improve engineering problem-solving., learn how researchers make better predictions with missing or sparse data., transfer machine learning approaches developed in one industry to another industry., quantify risk and clarify salient features from data in complex systems., assess conditions when a machine learning approach may not be helpful or worth the extra effort., who should enroll.

• Industry professionals with at least a bachelor's degree in engineering (e.g., computer science, mechanical, civil, aerospace, chemical, materials, nuclear, biological, electrical, electronics etc.) or the physical sciences.
• Other technical professionals with a background in college-level mathematics including differential calculus, linear algebra, and statistics.
• Programming experience not necessary , but some experience with MATLAB (R) is very useful.

PROGRAM FEE: $2,375 / $2,255

(All-Inclusive)

Pay in Full or in Interest-Free Installments

How you will learn

LEARN BY DOING

Practice processes and methods through simulations, assessments, case studies, and tools.

LEARN FROM OTHERS

Connect with an international community of professionals while working on projects based on real-world examples.

LEARN ON DEMAND

Access all of the content online and watch videos on the go.

REFLECT AND APPLY

Bring your new skills to your organization, through examples from technical work environments and ample prompts for reflection.

DEMONSTRATE YOUR SUCCESS

Earn a Professional Certificate and 2.5 Continuing Education Units (CEUs) from MIT.

LEARN FROM THE BEST

Gain insights from leading MIT faculty and industry experts.

MIT Faculty & Industry Experts

Meet your instructors

Youssef M. Marzouk

Faculty Co-Director of MIT Center of Computational Engineering, Associate Professor of Aeronautics & Astronautics and Director of Aerospace Computational Design Laboratory, MIT.

George Barbastathis

Professor of Mechanical Engineering, MIT.

Heather Kulik

Associate Professor of Chemical Engineering, MIT

Aram Harrow

Associate Professor of Physics at MIT

John Williams

Professor Civil & Environmental Engineering, MIT.

Themistoklis Sapsis

Associate Professor of Mechanical & Ocean Engineering, MIT.

Markus Buehler

McAfee Professor of Engineering & Head, Department of Civil & Environmental Engineering, MIT.

Richard Braatz

Edwin R. Gilliland Professor of Chemical Engineering, MIT.

Justin Solomon

Associate Professor of Electrical Engineering and Computer Science, MIT.

Laurent Demanet

Professor of Applied Mathematics & Director of MIT's Earth Resources Laboratory.

MIT Center for Computational Science and Engineering

MIT CCSE Faculty

Courses In this Program

To earn a Professional Certificate and CEUs, you must complete the two courses in the program.

Machine Learning, AI-based Modeling, and Simulation Principles

Course 1 of 2 in the program is Machine Learning, AI-based Modeling, and Simulations which involve engineering problem-solving in the age of AI.

Applying Machine Learning to Business, Engineering, and Science

Course 2 of 2 in the program is Applying Machine Learning to Business, Engineering, and Science which includes use cases of machine learning in various industries.

Dual Certification Programs

Select any USAII ® program according to your eligibility with Machine Learning for new age AI and Save 10% on each program fee.

Certified Artificial Intelligence Scientist (CAIS™)

Certified artificial intelligence consultant (caic™).

@ $3,269 / $3,060 ONLY

@ $3,162 / $2,960 ONLY

Frequently Asked Questions

What will i gain with this program.

• This program will help you acquire deeper foundational knowledge of machine learning from technical and business perspectives.
• Learn from the best in the world, the distinguished MIT faculty, and earn your certificate from MIT.
• Connect with an international community of professionals
• Work on real-world projects

Is this just one program?

Yes, however, there are two courses under this program:

• Course 1: Machine Learning, AI-based Modeling, and Simulations This course has a duration of 5 weeks and only 4-6 hours per day of learning is required to complete the course.
• Course 2: Applying Machine Learning to Business, Engineering, and Science This course has a duration of another 5 weeks and only 4-6 hours per day of learning is required to complete this course.

How to apply for the program?

You can enroll for this program through USAII ® , since they are the marketing and delivery partner of MIT xPRO for this program.

• Click on the Enroll Now button
• Fill in your details to apply for the program
• Once applied, your candidature will be reviewed for approval
• Meanwhile, you’ll receive access to the USAII ® dashboard
• After approval, you’ll receive an email from MIT on access to your MIT dashboard and the program orientation.

What is the refund policy?

• On behalf of MIT xPRO , USAII ® will enroll you in the program and collect the program fees. If you cancel the program seven (7) days before the start of the program, USAII ® will refund the program fee after deducting the processing fee of $249 .
• Any requests for withdrawal within seven (7) days of the start of the program will not be refunded.
• If you applied for the dual certification program from MIT xPRO and USAII ® and wish to cancel the dual program within 24 hours of payment, USAII ® will refund the program fee after deducting the processing fee of $249 .
• If you wish to cancel only the USAII ® program within 24 hours of payment, USAII ® will refund the ‘USAII ® program fee’ after deducting the processing fee of $147 .. For withdrawal requests beyond 24 hours, the fee paid will not be refunded.
• For any cancellation requests for the MIT xPRO program, please refer to the above-mentioned refund policy ‘For MIT xPRO Programs’.

Can I apply for Speedster discounts?

Speedster discounts are only for the USAII ® certification programs and not for MIT xPRO programs. However, if USAII ® is running any offers/campaigns you can receive the voucher codes that can be utilized while registering for the MIT xPRO programs.

Can I pay in parts or installments?

Yes, you can pay in two-part installments.

• The total program fee is divided into two (2) equal parts.
• This is an interest-free payment option.
• The first payment must be paid during your registration, which is 50% of the program fee.
• The second or final installment must be paid 7 days before the start of the program or earlier.
• In case of non-payment of the final installment, your candidature will be canceled and your first installment will be refunded.
• For refunds, please refer to the refund policy.

The best companies connect with the best minds at MIT

Deepen your team’s career knowledge and expand their abilities with MIT xPRO’s online courses for professionals. Develop customized learning for your team with bespoke courses and programs on your schedule. Set a standard of knowledge and skills, leading to effective communication among employees and consistency across the enterprise.

Find out what MIT xPRO can do for your team.

Other Courses From MIT

Get skilled with technical and leadership programs from the best in the world! 

Quantum Computing Fundamentals

Discover the business and technical implications of the new frontier in computing and how you can apply them to your organization with this two-course program from MIT.

Negotiating and Applying Influence and Power

A three-week online course for technical professionals that will empower you with the essential skills needed to solve problems, innovate, and drive change.

Understanding Organizational Strategy and Capabilities

A four-week online course for technical professionals that will empower you with the essential skills needed to solve problems, innovate, and drive change.

Managing Complex Projects and Organizations for Success

Gain an extensive understanding of all phases of project organization. Delivered in collaboration with Global Alumni..

Technical Innovation: From Strategy to Application

A three-week online course that will teach you to spark and drive innovation, the modern way.

For any queries, please write to us at

• Select Country* United States Afganistan Algeria Andorra Anguilla Antarctica Antigua and Barbuda Argentina Armenia Aruba Australia Austria Azerbaijan Bahamas Bahrain Bangladesh Barbados Belarus Belgium Belize Benin Bermuda Bhutan Bolivia Bosnia and Herzegovina Botswana Bouvet Island Brazil British Indian Ocean Territory British Virgin Islands Brunei Bulgaria Burkina Faso Burundi Cambodia Cameroon Canada Cape Verde Cayman Islands Central African Republic Chad Chile China Christmas Island Cocos (Keeling) Islands Colombia Comoros Congo Cook Islands Costa Rica Croatia Cuba Cyprus Czech Republic Democratic People's Rep. of Korea Democratic Republic of Congo Denmark Djibouti Dominica Dominican Republic East Timor Ecuador Egypt El Salvador Equatorial Guinea Eritrea Estonia Ethiopia Faroe Islands Federated States of Micronesia Fiji Finland France French Guyana French Polynesia French Southern Territories Gabon Gambia Georgia Germany Ghana Gibraltar Greece Greenland Grenada Guadeloupe Guam Guatemala Guinea Guinea-Bissau Guyana Haiti Heard Island and Mcdonald Islands Honduras Hong Kong Hungary Iceland India Indonesia Iran Iraq Iraq-Saudi Arabia Neutral Zone Ireland Israel Italy Ivory Coast Jamaica Japan Jordan Kazakhstan Kenya Kiribati Kuwait kyrgyzstan Laos Latvia Lebanon Lesotho Liberia Libya Liechtenstein Lithuania Luxembourg Macau Macedonia Madagascar Malawi Malaysia Maldives Mali Malta Marshall Islands Martinique Mauritania Mauritius Mayotte Mexico Moldova Monaco Mongolia Montserrat Morocco Mozambique Myanmar Namibia Nauru Nepal Netherlands Netherlands Antilles New Caledonia New Zealand Nicaragua Niger Nigeria Niue Norfolk Island North Korea Northern Mariana Islands Norway Oman Pakistan Palau Palestinian Occupied Territories Panama Papua New Guinea Paraguay Peru Philippines Pitcairn Islands Poland Portugal Puerto Rico Qatar Reunion Romania Russia Rwanda Saint Helena and Dependencies Saint Kitts and Nevis Saint Lucia Saint Pierre and Miquelon Saint Vincent and the Grenadines Samoa San Marino Sao Tome and Principe Saudi Arabia Senegal Serbia and Montenegro Seychelles Sierra Leone Singapore Slovakia Slovenia Solomon Islands Somalia South Africa South Georgia and South Sandwich Islands South Korea South Sudan Spain Spratly Islands Sri Lanka Sudan Suriname Svalbard and Jan Mayen Swaziland Sweden Switzerland Syria Taiwan Tajikistan Tanzania Thailand Togo Tokelau Tonga Trinidad and Tobago Tunisia Turkey Turkmenistan Turks And Caicos Islands Tuvalu Uganda Ukraine United Arab Emirates United Kingdom United Republic of Tanzania United States Minor Outlying Islands Uruguay US Virgin Islands Uzbekistan Vanuatu Vatican City Venezuela Vietnam Wallis and Futuna Western Sahara Yemen Zambia Zimbabwe
• Captcha validation is required.

Note: USAII ® respects your privacy, and your data is 100% secured with us. We are not sharing your data with any third party vendor.

This website uses cookies to enhance website functionalities and improve your online experience. By clicking Accept or continue browsing this website, you agree to our use of cookies as outlined in our privacy policy .

Suggestions or feedback?

MIT News | Massachusetts Institute of Technology

• Machine learning
• Social justice
• Black holes
• Classes and programs

Departments

• Aeronautics and Astronautics
• Brain and Cognitive Sciences
• Architecture
• Political Science
• Mechanical Engineering

Centers, Labs, & Programs

• Abdul Latif Jameel Poverty Action Lab (J-PAL)
• Picower Institute for Learning and Memory
• Lincoln Laboratory
• School of Architecture + Planning
• School of Engineering
• School of Humanities, Arts, and Social Sciences
• Sloan School of Management
• School of Science
• MIT Schwarzman College of Computing

Technique enables AI on edge devices to keep learning over time

Press contact :, media download.

*Terms of Use:

Images for download on the MIT News office website are made available to non-commercial entities, press and the general public under a Creative Commons Attribution Non-Commercial No Derivatives license . You may not alter the images provided, other than to crop them to size. A credit line must be used when reproducing images; if one is not provided below, credit the images to "MIT."

Previous image Next image

Personalized deep-learning models can enable artificial intelligence chatbots that adapt to understand a user’s accent or smart keyboards that continuously update to better predict the next word based on someone’s typing history. This customization requires constant fine-tuning of a machine-learning model with new data.

Because smartphones and other edge devices lack the memory and computational power necessary for this fine-tuning process, user data are typically uploaded to cloud servers where the model is updated. But data transmission uses a great deal of energy, and sending sensitive user data to a cloud server poses a security risk.  

Researchers from MIT, the MIT-IBM Watson AI Lab, and elsewhere developed a technique that enables deep-learning models to efficiently adapt to new sensor data directly on an edge device.

Their on-device training method, called PockEngine , determines which parts of a huge machine-learning model need to be updated to improve accuracy, and only stores and computes with those specific pieces. It performs the bulk of these computations while the model is being prepared, before runtime, which minimizes computational overhead and boosts the speed of the fine-tuning process.    

When compared to other methods, PockEngine significantly sped up on-device training, performing up to 15 times faster on some hardware platforms. Moreover, PockEngine didn’t cause models to have any dip in accuracy. The researchers also found that their fine-tuning method enabled a popular AI chatbot to answer complex questions more accurately.

“On-device fine-tuning can enable better privacy, lower costs, customization ability, and also lifelong learning, but it is not easy. Everything has to happen with a limited number of resources. We want to be able to run not only inference but also training on an edge device. With PockEngine, now we can,” says Song Han, an associate professor in the Department of Electrical Engineering and Computer Science (EECS), a member of the MIT-IBM Watson AI Lab, a distinguished scientist at NVIDIA, and senior author of an open-access paper describing PockEngine .

Han is joined on the paper by lead author Ligeng Zhu, an EECS graduate student, as well as others at MIT, the MIT-IBM Watson AI Lab, and the University of California San Diego. The paper was recently presented at the IEEE/ACM International Symposium on Microarchitecture.

Layer by layer

Deep-learning models are based on neural networks , which comprise many interconnected layers of nodes, or “neurons,” that process data to make a prediction. When the model is run, a process called inference, a data input (such as an image) is passed from layer to layer until the prediction (perhaps the image label) is output at the end. During inference, each layer no longer needs to be stored after it processes the input.

But during training and fine-tuning, the model undergoes a process known as backpropagation. In backpropagation, the output is compared to the correct answer, and then the model is run in reverse. Each layer is updated as the model’s output gets closer to the correct answer.

Because each layer may need to be updated, the entire model and intermediate results must be stored, making fine-tuning more memory demanding than inference

However, not all layers in the neural network are important for improving accuracy. And even for layers that are important, the entire layer may not need to be updated. Those layers, and pieces of layers, don’t need to be stored. Furthermore, one may not need to go all the way back to the first layer to improve accuracy — the process could be stopped somewhere in the middle.

PockEngine takes advantage of these factors to speed up the fine-tuning process and cut down on the amount of computation and memory required.

The system first fine-tunes each layer, one at a time, on a certain task and measures the accuracy improvement after each individual layer. In this way, PockEngine identifies the contribution of each layer, as well as trade-offs between accuracy and fine-tuning cost, and automatically determines the percentage of each layer that needs to be fine-tuned.

“This method matches the accuracy very well compared to full back propagation on different tasks and different neural networks,” Han adds.

A pared-down model

Conventionally, the backpropagation graph is generated during runtime, which involves a great deal of computation. Instead, PockEngine does this during compile time, while the model is being prepared for deployment.

PockEngine deletes bits of code to remove unnecessary layers or pieces of layers, creating a pared-down graph of the model to be used during runtime. It then performs other optimizations on this graph to further improve efficiency.

Since all this only needs to be done once, it saves on computational overhead for runtime.

“It is like before setting out on a hiking trip. At home, you would do careful planning — which trails are you going to go on, which trails are you going to ignore. So then at execution time, when you are actually hiking, you already have a very careful plan to follow,” Han explains.

When they applied PockEngine to deep-learning models on different edge devices, including Apple M1 Chips and the digital signal processors common in many smartphones and Raspberry Pi computers, it performed on-device training up to 15 times faster, without any drop in accuracy. PockEngine also significantly slashed the amount of memory required for fine-tuning.

The team also applied the technique to the large language model Llama-V2. With large language models, the fine-tuning process involves providing many examples, and it’s crucial for the model to learn how to interact with users, Han says. The process is also important for models tasked with solving complex problems or reasoning about solutions.

For instance, Llama-V2 models that were fine-tuned using PockEngine answered the question “What was Michael Jackson’s last album?” correctly, while models that weren’t fine-tuned failed. PockEngine cut the time it took for each iteration of the fine-tuning process from about seven seconds to less than one second on a NVIDIA Jetson Orin, an edge GPU platform.

In the future, the researchers want to use PockEngine to fine-tune even larger models designed to process text and images together.

“This work addresses growing efficiency challenges posed by the adoption of large AI models such as LLMs across diverse applications in many different industries. It not only holds promise for edge applications that incorporate larger models, but also for lowering the cost of maintaining and updating large AI models in the cloud,” says Ehry MacRostie, a senior manager in Amazon’s Artificial General Intelligence division who was not involved in this study but works with MIT on related AI research through the MIT-Amazon Science Hub.

This work was supported, in part, by the MIT-IBM Watson AI Lab, the MIT AI Hardware Program, the MIT-Amazon Science Hub, the National Science Foundation (NSF), and the Qualcomm Innovation Fellowship.

Share this news article on:

Related links.

• MIT-IBM Watson AI Lab
• Department of Electrical Engineering and Computer Science

Related Topics

• Computer science and technology
• Artificial intelligence
• Mobile devices
• Internet of things
• Electrical Engineering & Computer Science (eecs)
• National Science Foundation (NSF)

Related Articles

AI model speeds up high-resolution computer vision

Learning on the edge

Tiny machine learning design alleviates a bottleneck in memory usage on internet-of-things devices

System brings deep learning to “internet of things” devices

Previous item Next item

More MIT News

Celebrating diversity and cultural connections

Read full story →

Working to beat the clock on climate change

Mark Bear wins Society for Neuroscience Julius Axelrod Prize

Foreign policy scholars examine the China-Russia relationship

Ingestible electronic device detects breathing depression in patients

Rewarding excellence in open data

• More news on MIT News homepage →

Massachusetts Institute of Technology 77 Massachusetts Avenue, Cambridge, MA, USA

• Map (opens in new window)
• Events (opens in new window)
• People (opens in new window)
• Careers (opens in new window)
• Accessibility
• Social Media Hub
• MIT on Facebook
• MIT on YouTube
• MIT on Instagram

K-12 Education in the Age of AI: A Call to Action for K-12 AI Literacy

• OPINION/COMMENTARY
• Open access
• Published: 20 June 2023
• volume  33 ,  pages 228–232 ( 2023 )

You have full access to this open access article

• Ning Wang 1 &
• James Lester 2  

2664 Accesses

1 Altmetric

Explore all metrics

Cite this article

The emergence of increasingly powerful AI technologies calls for the design and development of K-12 AI literacy curricula that can support students who will be entering a profoundly changed labor market. However, developing, implementing, and scaling AI literacy curricula poses significant challenges. It will be essential to develop a robust, evidence-based AI education research foundation that can inform AI literacy curriculum development. Unlike K-12 science and mathematics education, there is not currently a research foundation for K-12 AI education. In this article we provide a component-based definition of AI literacy, present the need for implementing AI literacy education across all grade bands, and argue for the creation of research programs across four areas of AI education: (1) K-12 AI Learning & Technology; (2) K-12 AI Education Integration into STEM, Language Arts, and Social Science Education; (3) K-12 AI Professional Development for Teachers and Administrators; and (4) K-12 AI Assessment.

Avoid common mistakes on your manuscript.

Artificial intelligence (AI) is a profoundly transformative technology. Recent years have seen the emergence of powerful advances in AI that are quickly finding their way into every sector of industry and government and bringing about extraordinary developments in science, business, law, agriculture, transportation, security, and medicine. It has become clear that virtually no aspect of society will be untouched by AI. Further, the magnitude of these impacts is increasing, and the rate of these changes is accelerating. In short, we are rapidly entering a new era, the “Age of AI,” which has world-changing implications for K-12 education.

When today’s students enter the workforce, they will join a labor market that will have undergone a major shift from previous generations and be radically reshaped by advances in AI. AI holds not merely the potential to disrupt the labor market, it is clear that it will radically reshape what jobs are available, how they will be performed, and how workers should be prepared for them. Workers will share their jobs with AI, with most jobs featuring workers’ sharing responsibilities with AI systems. Industry will place a premium on “AI teaming” competencies that enable workers and AI to collaboratively solve problems with each playing roles that build on their respective strengths. Human-AI teaming will reward those workers who develop the ability to understand what tasks can best be performed by AI, what an AI system’s capabilities are, and how “AI job sharing” can best be accomplished.

We must re-envision K-12 education to prepare for the reality that AI will be prominently featured in every aspect of students’ lives, including their work lives. K-12 education in the Age of AI must cultivate students’ AI literacy , which we define as follows:

AI literacy is the ability to readily engage with AI by leveraging AI tools, systems, and frameworks to effectively and ethically solve problems in a wide range of sociocultural contexts.

Our framing of AI literacy has three key components:

Understanding AI capabilities : Workers will be required to understand AI capabilities at varying levels of technical expertise, which will range from essentially no technical expertise at all to deep knowledge of the theoretical underpinnings of state-of-the-art AI. The level of technical expertise required will be dependent on the tasks associated with a job: most tasks and jobs will require modest technical expertise or no technical expertise at all; AI engineering tasks will require substantial technical expertise; and AI research will require the highest levels of technical expertise. Different jobs and tasks will also require varying levels of AI teaming expertise.

Utilizing AI for problem solving : Workers will be required to apply AI tools, systems, and frameworks to solve a broad range of problems. Applying AI to solve problems will span cognitive, perceptual, psychomotor, and communicative tasks. Workers will be required to utilize AI to solve problems effectively, efficiently, and ethically. They will need to develop solutions to problems that are correct, address necessary time constraints, and consider the myriad ethical issues bearing on their tasks.

Applying AI in sociocultural contexts : Workers will need to be able to readily work with AI in a variety of contexts. The broader social and cultural dimensions in which AI is applied will significantly affect its utilization. For example, workers will need to be able to engage in AI teaming while also attending to sociocultural issues of communication and interaction where appropriate problem solving is contingent on the traditions and norms of a particular workplace and the broader society in which it is situated.

To cultivate AI literacy in students, K-12 education should introduce all K-12 students to the fundamentals of AI as well as provide an on-ramp to advanced AI education through a series of learning progressions for students who wish to pursue technical careers in AI. Beginning in elementary grades, students should begin developing AI literacy through examples of how AI manifests in widely used software and being introduced to social issues in AI. In middle school, AI literacy should focus on age-appropriate AI technical concepts that provide a foundation for future studies, and it should encompass social aspects of AI that enable students to understand the broader contexts in which AI is deployed. In high school, AI education should continue to be universal, i.e., required of all students, and advanced AI coursework should be available for students who may elect to pursue AI engineering and AI research careers. It is important to note that AI literacy does not entail requiring all students to engage with advanced concepts or mathematical foundations of AI. AI literacy, however, should require all students to develop a solid understanding of AI capabilities, their limits, their application, and the ethical considerations bearing on their use.

Developing, implementing, and scaling an AI literacy curriculum poses significant challenges. It will be essential to develop an evidence-based AI education research base that can inform AI literacy curriculum development. We do not currently have a research foundation for how AI is learned, how AI learning progressions should proceed, how AI education should be integrated into STEM, language arts, and social science education, how AI should be taught, and how AI learning should be assessed. Funding agencies have begun to recognize this shortcoming, and we are seeing the very early emergence of AI education research, but it is still in its nascent form. While experimental AI curricula have sprung up internationally, these are not informed by evidence-based pedagogies. A significant concern is that the enormous and growing demand for K-12 AI education is beginning to result in the adoption and implementation of curricula that are not grounded in education research.

Building the educational research foundation for K-12 AI literacy

Given the global economic and societal importance of AI, we must develop a robust foundation for K-12 AI education. Following the lead of the science and mathematics education research communities, which have produced firm foundations for K-12 science and math education, and leveraging advances in computer science education research, research programs should be developed for four areas of K-12 AI education (Fig.  1 ):

K-12 AI Learning & Teaching : We need to develop an understanding of how to design effective frameworks and methods for learning and teaching AI in K-12 schools. Design principles for K-12 AI education should be theoretically grounded in the learning sciences, STEM education research, and educational psychology. In addition to informing the design of curricula for AI conceptual knowledge and problem-solving practices, they should also inform the design of learning experiences that give primacy to AI ethics and the social impact of AI. Research should be conducted on AI learning progressions, and AI education should be investigated in both formal and informal learning contexts. Finally, the emergence of increasingly powerful AI technologies calls for the investigation of AI-driven learning technologies to support learning and teaching of AI (i.e., AI learning technologies supporting AI as a subject matter), with opportunities for leveraging advances in explainable AI being particularly intriguing.

K-12 AI Education Integration into STEM, Language Arts and Social Science Education : Because AI will play an increasingly important role in every discipline and in every sector of the economy, we must develop an understanding of how to develop an integrated model of K-12 AI education that infuses AI throughout the curriculum. For example, students must become adept at using AI in science and math, as well as in language arts and social science. We need to develop innovative approaches to integrative AI-science, AI-math, AI-language arts, and AI-social science education. Doing so will not only yield more effective approaches to, for example, “doing science,” it will also address the practicalities of school curricula that are already exceedingly full.

K-12 AI Professional Development for Teachers : We must develop an understanding of how to create the most effective AI professional development (PD) for K-12 teachers. Doing so will entail creating a research program in design principles for creating professional development around both AI content knowledge and AI pedagogical knowledge. K-12 AI PD research should investigate approaches that can effectively take into account the broad range of competencies in AI that teachers and administrators begin with. While exploring K-12 AI PD methods for in-service teachers is critical, it will be equally important to do so for pre-service teachers as well.

K-12 AI Assessment : We must develop frameworks for assessing students’ AI competencies. To best support the broader enterprise of K-12 AI education, we need to create innovative approaches to measuring students’ conceptual understanding of AI, their proficiency in using AI in problem solving, and their ability to address considerations of AI ethics. In addition to developing new instruments that use rigorous approaches for assessing AI competencies, we need to investigate new methods and technologies (including AI-driven assessment technologies) that can reliably measure AI conceptual knowledge, problem-solving practices, and ethical foundations.

As the demand for K-12 AI education continues to grow, and every sign points to its exceptionally rapid growth, education policy makers across the globe should promote the development, implementation, and scaling of a comprehensive K-12 AI literacy curriculum that is designed to support all students, not just those who will pursue R&D AI careers. Beginning in elementary school and continuing through secondary school, a robust AI literacy curriculum will enable today’s K-12 students to flourish in the Age of AI.

This research was supported by funding from the National Science Foundation (NSF) under grants DRL-1,938,758 and DRL-1,938,778. Any opinions, findings, and conclusions or recommendations expressed in this material are those of the author(s) and do not necessarily reflect the views of the National Science Foundation.

Open access funding provided by SCELC, Statewide California Electronic Library Consortium

Author information

Authors and affiliations.

Institute for Creative Technologies, University of Southern California, Los Angeles, CA, USA

Center for Educational Informatics, North Carolina State University, Raleigh, NC, USA

James Lester

You can also search for this author in PubMed   Google Scholar

Contributions

Both authors are co-editors of the special issue. Both authors read and approved the final manuscript.

Corresponding author

Correspondence to Ning Wang .

Ethics declarations

Conflicts of interest.

No potential conflicts of interest.

Additional information

Publisher’s note.

Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Rights and permissions

Springer Nature or its licensor (e.g. a society or other partner) holds exclusive rights to this article under a publishing agreement with the author(s) or other rightsholder(s); author self-archiving of the accepted manuscript version of this article is solely governed by the terms of such publishing agreement and applicable law.

Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/ .

Reprints and Permissions

About this article

Wang, N., Lester, J. K-12 Education in the Age of AI: A Call to Action for K-12 AI Literacy. Int J Artif Intell Educ 33 , 228–232 (2023). https://doi.org/10.1007/s40593-023-00358-x

Download citation

Published : 20 June 2023

Issue Date : June 2023

DOI : https://doi.org/10.1007/s40593-023-00358-x

Share this article

Anyone you share the following link with will be able to read this content:

Sorry, a shareable link is not currently available for this article.

Provided by the Springer Nature SharedIt content-sharing initiative

• AI education
• K-12 education
• K-12 AI education
• Computer science education

Advertisement

• Find a journal
• Publish with us

• Open access
• Published: 19 September 2022

The application of AI technologies in STEM education: a systematic review from 2011 to 2021

• Weiqi Xu 1 &
• Fan Ouyang   ORCID: orcid.org/0000-0002-4382-1381 1  

International Journal of STEM Education volume  9 , Article number:  59 ( 2022 ) Cite this article

32k Accesses

24 Citations

2 Altmetric

Metrics details

The application of artificial intelligence (AI) in STEM education (AI-STEM), as an emerging field, is confronted with a challenge of integrating diverse AI techniques and complex educational elements to meet instructional and learning needs. To gain a comprehensive understanding of AI applications in STEM education, this study conducted a systematic review to examine 63 empirical AI-STEM research from 2011 to 2021, grounded upon a general system theory (GST) framework.

The results examined the major elements in the AI-STEM system as well as the effects of AI in STEM education. Six categories of AI applications were summarized and the results further showed the distribution relationships of the AI categories with other elements (i.e., information, subject, medium, environment) in AI-STEM. Moreover, the review revealed the educational and technological effects of AI in STEM education.

Conclusions

The application of AI technology in STEM education is confronted with the challenge of integrating diverse AI techniques in the complex STEM educational system. Grounded upon a GST framework, this research reviewed the empirical AI-STEM studies from 2011 to 2021 and proposed educational, technological, and theoretical implications to apply AI techniques in STEM education. Overall, the potential of AI technology for enhancing STEM education is fertile ground to be further explored together with studies aimed at investigating the integration of technology and educational system.

Introduction

Artificial intelligence in education (AIEd) is an emerging interdisciplinary field that applies AI technologies in education to transform and promote the instructional and learning design, process and assessment (Chen et al., 2020 ; Holmes et al., 2019 ; Hwang et al., 2020 ). The application of AI in STEM education (referred to AI-STEM in this paper), as a sub-branch of AIEd, focuses on the design and implementation of AI applications to support STEM education. Automated AI technologies, e.g., intelligence tutoring, automated assessment, data mining and learning analytics, have been used in STEM education to enhance the instruction and learning quality (Chen et al., 2020 ; Hwang et al., 2020 ; McLaren et al., 2010 ). STEM education is a complex system, from a system perspective, consisting of interdependent elements, including subject, information, medium, and environment (Rapoport, 1986 ; Von Bertalanffy, 1968 ). The application of AI, as a critical technology element, should take careful consideration of these complex factors, to achieve a high-quality STEM education (Byrne & Callaghan, 2014 ; Krasovskiy, 2020 ; Xu & Ouyang, 2022 ). This systematic review aims to examine the different elements, including AI technology, subject, information, medium, environment in the AI-STEM system to gain a holistic understanding of the application and integration of AI technologies in the STEM education contexts. Specifically, we collected and reviewed empirical AI-STEM research from 2011 to 2021, summarized the AI techniques and applications, the characteristics of other system elements (i.e., information, subject, medium, environment), the distribution of AI in these elements, and the effects of AI in STEM education. Based on the results, this systematic review provided educational and technological implications for the practice and research in the AI-STEM education.

Literature review

With the development of computer science and computational technologies, automatic, adaptive, and efficient AI technologies have been widely applied in various academic fields. Artificial Intelligence in Education (AIEd), as an interdisciplinary field, emphasizes applying AI to assist instructor’s instructional process, empower student’s learning process, and promote the transformation of educational system (Chen et al., 2020 ; Holmes et al., 2019 ; Hwang et al., 2020 ; Ouyang & Jiao, 2021 ). First, AIEd has potential to enhance instructional design and pedagogical development in the teaching processes, such as accessing students’ performance automatically (Wang et al., 2011 ; Zampirolli et al., 2021 ), monitoring and tracking students’ learning (Berland et al., 2015 ; Ji & Han, 2019 ), and predicting at-risk students (Hellings & Haelermans, 2020 ; Lamb et al., 2021 ). Second, AIEd is beneficial for improving student-centered learning, such as providing adaptive tutoring (Kose & Arslan, 2017 ; Myneni et al., 2013 ), recommending personalized learning resources (Ledesma & García, 2017 ; Zhang et al., 2020 ), and diagnosing students’ learning gaps (Liu et al., 2017 ). Third, AIEd also brings opportunities to transform the educational system by highlighting the essential role of technology (Hwang et al., 2020 ), enriching the mediums of knowledge delivery (Holstein et al., 2019 ; Yannier et al., 2020 ), and changing the instructor–student relationship (Xu & Ouyang, 2022 ). Overall, different AI technologies (e.g., machine learning, deep learning) have been deployed in the field of education to enhance instructional and learning process.

The development of AIEd also brought transformations to the field of science, technology, engineering and mathematics (STEM) education, as a sub-branch of AIEd named AI-STEM. STEM education aims to improve students’ interdisciplinary knowledge inquiry and application, as well as their higher-order thinking, critical thinking and problem-solving ability (Bybee, 2013 ; Pimthong & Williams, 2018 ). The application of AI in STEM education has advantages to provide adaptive and personalized learning environments or resources, and aid instructors to understand students’ learning behavioral patterns, and automatically assess STEM learning performances (Alabdulhadi & Faisal, 2021 ; Walker et al., 2014 ). However, STEM education is a complex system, consisting of interdependent elements, including subject (e.g., instructor, student), information, medium, and environment (Rapoport, 1986 ; Von Bertalanffy, 1968 ). Achieving a high quality of STEM education requires a careful consideration of the complex social, pedagogical, environmental factors, rather than merely applying AI technologies in education (Krasovskiy, 2020 ; Xu & Ouyang, 2022 ). Therefore, a major challenge in AI-STEM is how to appropriately select and apply AI techniques to adapt to the multiple elements (e.g., subject, information, environment) in STEM education with a goal of high-quality instruction and learning (Castañeda & Selwyn, 2018 ; Selwyn, 2016 ). To gain a holistic understanding of the integration of AI technologies in the STEM education contexts, it is crucial to systematically review and examine the complex elements in AI-STEM from a system perspective.

During the past decade, the emerging field of AIEd has gained great attention (Chen et al., 2020 ; Holmes et al., 2019 ; Hwang et al., 2020 ; Ouyang et al., 2022 ). But existing literature review of AIEd has mainly focused on the trends, applications, and effects of AIEd from a technological perspective (Chen et al., 2020 ; Tang et al., 2021 ; Zawacki-Richter et al., 2019 ). Specifically, we located 18 literature review articles of AIEd published from 2011 to 2021 (see Fig.  1 ). These AIEd reviews focused on different educational levels, fields, and contexts, including higher education (Zawacki-Richter et al., 2019 ), e-learning (Tang et al., 2021 ), mathematics education (Hwang & Tu, 2021 ), language education (Liang et al., 2021 ), medical education (Khandelwal et al., 2019 ; Lee et al., 2021 ), programming education (Le et al., 2013 ), and special education (Drigas & Ioannidou, 2012 ). For example, Zawacki-Richter et. al. ( 2019 ) reviewed AIEd in the higher education context and four AI technical applications were classified, namely intelligent tutoring systems, adaptive systems and personalization, profiling and prediction, and assessment and evaluation. Liang et. al. ( 2021 ) focused on the application of AI in language education and investigated the roles and research foci (e.g., research methods, research sample groups) of AI techniques in language education. Drigas and Ioannidou ( 2012 ) explored AIEd in special education and summarized AI applications based on the student’s disorders, including reading, writing and spelling difficulties, dyslexia, autistic spectrum disorder, etc.

Existing literature review of AIEd articles, ranging from 2011 to 2021

Although various reviews were conducted to understand the field of AIEd, few of them focused on STEM education. Among these 18 literature review articles, we only located two works exploring the application of AI in STEM education. Le et. al. ( 2013 ) reviewed the AI-supported tutoring approaches in computer programming education and found that AI techniques were mainly applied to support feedback-based programming tutoring during the student’s individual learning. Hwang and Tu ( 2021 ) conducted a bibliometric mapping analysis to systematically review the roles of AI in mathematics education. The results clarified the role of AI in mathematics education into three main types, including intelligent tutoring systems, profiling and prediction, and adaptive systems and personalization. Although some review examined AI in computer science and mathematics education, there is a lack of literature review to investigate the application of AI in general STEM education context. More importantly, due to the complexity of AI-STEM, it is essential to systematically review multiple elements in AI-STEM as well as the effects of AI in the STEM education system.

To fill this gap, this systematic review aims to gain a comprehensive understanding of the integration of AI technologies in the STEM education contexts. Specifically, this review examined and summarized the applications and categories of AI element in the AI-STEM system, the characteristics of other system elements in AI-STEM except AI, the distribution of AI in these elements, and the effects of AI in STEM education. Three research questions (RQs) were proposed:

RQ1: What are the categories of the AI element in the AI-STEM system?

Rq2: what are the characteristics of other system elements (i.e., information, subject, medium, environment element) as well as the distribution of ai in these elements, rq3: what are the effects of ai in stem education.

In order to map the state-of-art of the application of AI techniques in STEM education, we conducted a systematic review from 2011 to 2021, following the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) principles (Moher et al., 2009 ).

Database search

To locate the empirical studies of AI application in STEM education, the following major publisher databases were selected: Web of Science, Science Direct, Scopus, IEEE, EBSCO, ACM, Taylor & Francis, and Wiley (Guan et al., 2020 ). Filters were used to the empirical research and peer-reviewed articles in the field of education and educational research from January 2011 to December 2021. After the preliminary screening of articles, snowballing was conducted (Wohlin, 2014 ) to find the articles that were not extracted using the search strings.

Identification of search terms

Based on the specific requirements of bibliographic databases, we proposed the searching strategies. In terms of the research questions, three types of keywords were used as the search terms. First, keywords related to AIEd and specific AI applications were added (i.e., “artificial intelligence” OR “AI” OR “AIED” OR “machine learning” OR “intelligent tutoring system” OR “expert system” OR “recommended system” OR “recommendation system” OR “feedback system” OR “personalized learning” OR “adaptive learning” OR “prediction system” OR “student model” OR “learner model” OR “data mining” OR “learning analytics” OR “prediction model” OR “automated evaluation” OR “automated assessment” OR “robot” OR “virtual agent” OR “algorithm”). Second, keywords related to STEM were added (i.e., “STEM” OR “science” OR “technology” OR “math” OR “physics” OR “chemistry” OR “biology” OR “geography” OR “engineering” OR “programming” OR “lab”). Third, keywords related to education were added (i.e., “education” OR “learning” OR “course” OR “class” OR “teaching”).

Searching criteria

The search criteria were designed to locate the articles that focused on the applications of AI in STEM education. According to the research objectives, inclusion and exclusion criteria were adopted (see Table 1 ).

The screening process

The screening process involved the following procedures: (1) removing the duplicated articles; (2) reading the titles and abstracts and removing the articles according to the inclusion and exclusion criteria; (3) reading the full texts and removing the articles according to the inclusion and exclusion criteria; (4) using the snowballing to further locate the articles in Google Scholar, and (5) extracting data from the final filtered articles (see Fig.  2 ). All articles were imported into Mendeley software for screening.

The selection flowchart used based on PRISMA (Moher et al., 2009 )

3373 articles were located as the result of the first round of searching. Among these records, 777 duplicates were removed and then 1879 records were excluded because they were not classified under Education & Educational research or journal article. By reviewing the titles and abstracts, the number of articles was reduced to 717 based on the criteria (see Table 1 ). The selected articles were examined by the first author to determine whether they were suitable for the purpose of this systematic review. The second author independently reviewed approximately 30% of the articles to confirm the reliability. The inter-rater agreement was 92%. Then, the full-text of articles were reviewed by the first author to verify that the articles met all the criteria for inclusion in the review. Finally, a total of 63 articles that met the criteria were identified for the systematic review.

Theoretical framework and analysis procedure

General system theory (GST) is a theoretical framework, arguing that the world is composed of different organic systems, which contain dynamically interacting elements and mutual relationships between them (Rapoport, 1986 ; Von Bertalanffy, 1950 ). The main principle of GST is that a system is not simply equal to the sum of its elements, but greater than the sum of its parts (Drack & Pouvreau, 2015 ; Von Bertalanffy, 1968 ). To deeply understand the complex nature and general rules of systems, GST highlights the system’s holistic principle to identify the internal elements, functional relationships of them, as well as the external influences upon a system (Crawford, 1974 ). The theoretical framework of GST has been widely applied in various fields to analyze different types of systems, such as physical, biological, social and educational systems (Drack & Pouvreau, 2015 ; Kitto, 2014 ). For example, Chen and Stroup ( 1993 ) suggested applying GST as an underpinned theoretical framework to guide the reform of science education and highlighted the integration of science curriculum to avoid the compartmentalized learning of physics, biology, and chemistry. Following this philosophy, we argue that GST can provide a new, holistic perspective to understand the integration of AI technologies and STEM education.

From the perspective of GST, AI-STEM can be viewed as an organic system, which mainly contains five basic elements, namely subject , information , medium , environment , and technology (Von Bertalanffy, 1968 ) (see Fig.  3 ). First, subject is defined as people in an educational system and different subjects of people (e.g., instructor, student) can take agency to interact with each other constantly and adaptively. Second, information refers to knowledge spread and constructed between subjects in an educational system, such as learning contents, course materials, knowledge artifacts, etc. Third, medium is the way or carrier to convey information and connect subjects in the system. Fourth, environment serves as an underlying context in an educational system, which influences the function of the whole educational system. Fifth, technology (e.g., AI techniques) is usually appeared as an external element to impact the functions of the educational system. Grounded upon GST, the integration of AI, as an external technology element, in an educational system (such as STEM) is a complex process, that has influences on other system elements (i.e., subject, information, medium, environment) and on the relationships between them. In summary, the framework of GST (see Fig.  3 ) highlights the multiple elements as well as their mutual relationships in AI-STEM system, which provides us a holistic view for applying AI technologies in STEM education.

The integration of technology in an educational system from the GST perspective

We used content analysis method (Cohen et al., 2005 ; Zupic & Čater, 2015 ) to classify 63 AI-STEM articles in order to answer the research questions. Based on GST, a coding scheme of educational system elements was developed to systematically examine AI-STEM articles (see Table 2 ). This coding scheme included the subject of instructor (including instructor involvement and instructor strategy), the subject of learner (including educational level, sample size, and learning outcome), information (i.e., learning content), medium (i.e., educational medium), environment (i.e., educational context), and technology (i.e., AI technique).

63 articles were coded by two raters. The same article can be coded more than one code in one dimension. First, 20% of articles were coded by two coders independently in order to calculate coding reliability. Krippendorff’s ( 2004 ) alpha reliability was 0.91 among two raters at this phase. The remaining articles were coded independently by two raters after the reliability was ensured. Consensus was reached by two raters on conflicting coding results. We provided details and examples below to demonstrate how the coding results represented the review data (Graneheim & Lundman, 2004 ).

To answer the research questions, the results section presents the following three main topics: (1) the categories of the AI element in the AI-STEM system; (2) the characteristics of other system elements (i.e., information, subject, medium, and environment) as well as the distribution of AI in these elements, and (3) the effects and findings of the application of AI in STEM education.

Figure  4 demonstrates the trends of empirical studies by year. According to the distribution, the number of publication generally increased along the years. In addition, a majority of reviewed articles ( N  = 42) were published in the last 4 years from 2018 to 2021. Only 9 of 63 reviewed articles were published in the first 4 years from 2011 to 2014.

Distribution of the articles by year ( N  = 63)

Regarding the element of AI technology in AI-STEM, six types of AI applications were identified, namely learning prediction ( N  = 18, percentage = 29%), intelligent tutoring system ( N  = 16, percentage = 25%), student behavior detection ( N  = 13, percentage = 21%), automation ( N  = 8, percentage = 13%), educational robots ( N  = 6, percentage = 9%), and others ( N  = 2, percentage = 3%) (see Fig.  5 and Table 3 ).

The application categories of AI techniques in STEM education ( N  = 63)

Learning prediction

The first category of the AI applications in STEM education was learning prediction , illustrating should be something like systems which predict student learning performance or status in advance through AI algorithms and modeling approaches (Agrawal & Mavani, 2015 ; Lee et al., 2017 ). 18 of 63 reviewed articles (29%) focused on learning prediction in STEM education (see Fig.  5 ). Two sub-categories were summarized under learning prediction: learning performance prediction ( N  = 14) and at-risk student prediction ( N  = 4) (see Table 3 ). First, in the sub-category of learning performance prediction, AI algorithms and modeling techniques were employed in STEM education to help instructors adjust the instructional processes by predicting students’ learning performance (Deo et al., 2020 ; Hellings & Haelermans, 2020 ). For example, Buenaño-Fernández et. al. ( 2019 ) applied educational data mining and machine learning technique (i.e., decision tree) in computer engineering courses to predict students’ final performance based on their historical grades. Zabriskie et. al. ( 2019 ) utilized the random forest model and logistic regression model to predict the physics course outcomes. Another sub-category was predicting at-risk students and dropout factors in STEM education to help instructors intervene in student learning (Vyas et al., 2021 ; Yang et al., 2020 ). For example, Lacave et. al. ( 2018 ) used Bayesian networks techniques to investigate dropout factors of computer science students in higher education. Yang et. al. ( 2020 ) utilized random forest classification to create and examine the prediction models of identifying at-risk students in introductory physics classes. In summary, AI algorithms had been used in STEM education to help instructors or researchers predict students’ final academic performances and learning risks.

Intelligent tutoring system

The second category of the AI applications in STEM education was the intelligent tutoring system (ITS), defined as an AI-enabled system that was designed to provide customized instruction or feedback to students and promote personalized, adaptive learning (Chen et al., 2020 ; Hooshyar et al., 2015 ; Murray, 2003 ). Among the 63 reviewed articles, 16 articles (25%) focused on the applications of ITSs in promoting instruction and learning in STEM education (see Fig.  4 ). Three sub-categories were identified: instructional content delivery ( N  = 9), recommendation of personalized learning path ( N  = 4), and resource recommendation ( N  = 3) (see Table 3 ). The first sub-category was using ITSs to deliver instructional content in STEM education. For example, Myneni et. al. ( 2013 ) introduced an interactive and intelligent learning system in physics education, where a virtual agent delivered physics concepts to students and decision algorithms were utilized to determine the support level of the virtual agent. Hooshyar et. al. ( 2018 ) proposed a novel flowchart-based ITS based on Bayesian networks techniques, which imitated a human instructor to conduct one-to-one instruction with students. The second sub-category of ITS was recommendations of personalized learning path based on student’s profile in STEM education. For example, De-Marcos et. al. ( 2015 ) combined genetic algorithm and parliamentary optimization algorithm to create personalized courseware sequencing paths in online STEM learning. Saito and Watanobe ( 2020 ) proposed an approach of recommending learning paths that applied recurrent neural network and sequential prediction model to create students’ ability charts and learning paths in programming learning based on their submission history. The third sub-category of ITS was recommending learning resource according to student’s needs in STEM education. For example, Ledesma and García ( 2017 ) introduced an expert system as a support tool to tackle mathematical topics, by recommending appropriate mathematical problems in accordance with a student’s learning style. Lin and Chen ( 2020 ) proposed a deep learning recommendation based system in programming learning that recommended learning tasks, learning missions and materials according to students’ learning processes and levels. In summary, AI technologies were widely applied in ITSs to enhance personalized and adaptive learning in STEM education through providing one-to-one tutoring and recommending personalized learning paths and resources.

Student behavior detection

The third category of the AI applications in STEM education was student behavior detection , which referred to systems to exploit and track students’ learning behaviors, patterns, and characteristics with AI-enabled data mining and learning analytics in the instructional and learning processes (Chrysafiadi & Virvou, 2013 ; Ji & Han, 2019 ; Zheng et al., 2020 ). Among the 63 reviewed articles, 13 articles (21%) focused on the applications of AI techniques to detect student behaviors in STEM education (see Fig.  5 ). Two sub-categories were summarized under student behavior detection: student behavior analysis ( N  = 8) and student behavior monitoring ( N  = 5) (see Table 3 ). First, the sub-category of student behavior detection, was applied in STEM education to analyze and reveal students’ latent behaviors. For example, Hsiao et. al. ( 2020 ) collected students’ learning data from programming learning platform and examined their learning behaviors through hidden Markov model, and the results revealed the reviewing patterns and reflecting strategies of students in learning programming. Pereira et. al. ( 2020 ) used data mining techniques including k-means and association rule algorithm to understand students’ behavior in introductory programming, to help novice programmers promote their learning. Another sub-category of student behavior monitoring was applied to help instructors track students’ learning in STEM education. For example, Balakrishnan ( 2018 ) helped instructors to motivate engineering students’ learning through monitoring their learning behaviors such as preferred learning materials and self-directed learning performance. Yannier et. al. ( 2020 ) introduced a mixed-reality AI system supported with computer vision algorithms to track children’s active learning behaviors in science education. In summary, student behavior detection had great potential to aid instructors and researchers to analyze, understand, and monitor students’ behaviors in STEM education.

The fourth category of the AI applications in STEM education was automation , which utilized AI technologies to automatically assess students’ performances and generate questions or tasks for instructors (Aldabe & Maritxalar, 2014 ; Wang et al., 2011 ; Zampirolli et al., 2021 ). Among 63 reviewed articles, 8 articles (13%) focused on the AI-supported automated techniques in STEM education (see Fig.  5 ). Two sub-categories were summarized under automation: automated assessment ( N  = 7) and automated questions generation ( N  = 1) (see Table 3 ). The first sub-category of automated assessment provided instructors and students with convenient assistance in STEM education. For example, Wang et. al. ( 2011 ) developed an automated assessment system, AutoLEP, to help novice programmers gain programming skills by providing syntactic and structural checking and immediate feedback automatically. García-Gorrostieta et. al. ( 2018 ) introduced a system for automatic argument assessment of computer engineering students’ final reports, to help them improve the abilities of statement and justification in science argumentation. Another sub-category of automated questions generation had potential to reduce instructors’ instructional burdens in STEM education. For example, Aldabe and Maritxalar ( 2014 ) proposed an approach to help instructors automatically create multiple-choice tests in science courses through the use of corpora and natural language processing techniques. In summary, AI techniques were used in STEM education to aid instructors and students through automatically generating questions and assessing academic performances.

Educational robots

The fifth category of the AI applications in STEM education was educational robots , which was the adoption of robots in STEM education to facilitate students’ learning experience as well as allow them to acquire knowledge in interactive ways (Atman Uslu et al., 2022 ; Cao et al., 2021 ; Yang & Zhang, 2019 ). It is worth noting that robots are applications that contain various techniques (e.g., mechanical manufacturing, electronic sensors, AI); therefore, considering the research topic, only AI-supported robots were included in this review. Among the 63 reviewed articles, 6 articles (9%) focused on the application of educational robots in STEM education (see Fig.  5 ). Two sub-categories were identified under educational robots: programming robots ( N  = 3) and social robots ( N  = 3) (see Table 3 ). The first sub-category, programming robots, were specifically designed as learning tools that engaged students to design and operate them with programming languages (Atman Uslu et al., 2022 ). For example, Rodríguez Corral et. al. ( 2016 ) applied a specific ball-shaped robot with sensing, wireless communication and output capabilities in computer courses to teach students object-oriented programming languages. Cao et. al. ( 2021 ) introduced an artificial intelligence robot called LEGO MINDSTORMS EV3, to implement instructional tasks in information technology courses to promote students’ innovation and operational ability. Another sub-category of social robots was a kind of intelligent humanoid robots, which could serve as tutors, tutees or learning companions to students and allow students to interact with them orally and physically (Belpaeme et al., 2018 ; Xu & Ouyang, 2022 ). For example, Verner et. al. ( 2020 ) employed RoboThespian, a life-size humanoid robot, as a tutor to convey science knowledge and concepts to elementary school students. In summary, AI-based educational robots were used in STEM education as instructional tools or educational subjects (e.g., tutor, tutee, companion) to convey knowledge, promote students’ operational skills, and enhance their learning experience.

Among 63 reviewed articles, 2 articles (3%) focused on other applications of AI techniques in STEM education, including AI textbook and group formation. Tehlan et. al. ( 2020 ) utilized a genetic algorithm‐based approach to form student groups in collaborative learning based on their skills and personality traits in a programming course. Koć-Januchta et. al. ( 2020 ) introduced AI-enriched textbook in biology course to improve students’ engagement by encouraging them to ask questions and receive suggested questions.

To further understand how AI techniques have been integrated in STEM education, we examined the other system elements, including information, subject (i.e., instructor, student), medium, and environment in AI-STEM research. In addition, we explored the distribution of AI categories in these elements, to reveal the relationships between AI techniques and these elements.

Information in AI-STEM research

Information (referred to learning content in this study) was described as the subject knowledge and learning contents conveyed in AI-STEM system. In the reviewed 63 studies, all of them mentioned learning content in STEM education, including science, technology, engineering, mathematics, and cross-disciplinary (i.e., more than one discipline) (see Table 4 ). Among 63 articles, 24 studies focused on technology, followed by articles that focused on science ( N  = 22) and engineering ( N  = 7). Mathematics ( N  = 3) attracted the least attention. In addition, 7 studies contained interdisciplinary subjects, such as computer engineering (Buenaño-Fernández et al., 2019 ; Tehlan et al., 2020 ), engineering mathematics (Deo et al., 2020 ), and integrated STEM education (Suh et al., 2019 ; Wang, 2016 ).

Figure  6 shows the frequency of AI application categories in different learning contents. Among all the AI categories, student behavior detection was most frequently applied in the technology domain ( N  = 8), followed by learning prediction in science ( N  = 6), and learning prediction in engineering ( N  = 5) (see Fig.  6 ).

AI categories under learning content ( N  = 63)

Instructor in AI-STEM research

Instructor , as a component of subject element in AI-STEM system, played a critical role in conducting instruction, conveying knowledge, and utilizing technologies. In the reviewed 63 studies, 50 of them mentioned the instructor involvement and 50 of them mentioned the instructional strategies, including traditional lecture, problem-based learning, project-based learning, game-based learning, self-learning, and collaborative learning (see Table 5 ). Regarding the instructor involvement, a majority of instructors would engage in the instructional and learning processes to support students ( N  = 42), while some studies were conducted without instructors’ involvement and support ( N  = 8). Additionally, among 50 articles, the traditional lecturing strategy was most frequently used by instructor ( N  = 27), followed by problem-based learning ( N  = 10). Also, some studies were carried out through project-based learning ( N  = 5), self-learning ( N  = 5), game-based learning ( N  = 4), and collaborative learning ( N  = 4).

All AI application categories were mainly applied with the instructor’s support, in which the most frequently used AI category were ITS ( N  = 11) and learning prediction ( N  = 11) (see Fig.  7 a). In addition, automation was only applied in lecture ( N  = 8). Learning prediction was most frequently applied in lecture ( N  = 8) and educational robots were most frequently applied in problem-based learning ( N  = 4). Compared to other AI technologies, ITS and student behavior detection were integrated with more types of instructional strategies (see Fig.  7 b).

AI categories under instructor involvement and instructional strategies

Learner in AI-STEM research

Learner , as another component of subject element, could take agency to actively participate in the learning process as to influence the AI-STEM system. In the reviewed 63 studies, 59 of them mentioned the educational levels of learners, from kindergarten to higher education, and 55 of them mentioned sample sizes (see Table 6 ). Among all the educational levels, 43 focused on higher education ( N  = 43), followed by elementary school ( N  = 7), high school ( N  = 5), and middle school ( N  = 4). Only one study was conducted in kindergarten. In addition, the number of AI-STEM studies with the medium scale of learners ( N  = 24) and the large scale of learners ( N  = 21) were larger than the small-scale study ( N  = 10).

AI application categories except educational robots were frequently applied in higher education (learning prediction: N  = 13, ITS: N  = 12, student behavior detection: N  = 8, automation: N  = 7). The educational robots were frequently applied in elementary school ( N  = 3) (see Fig.  8 a). Moreover, regarding the sample size, learning prediction was most frequently used with a large scale ( N  = 11), followed by ITS with a medium scale ( N  = 9), and student behavior detection with a medium scale ( N  = 7). Additionally, the categories of educational robots and others were not applied in large scale; the category of student behavior detection was not applied in small scale (see Fig.  8 b).

AI categories under educational levels and sample sizes

Medium in AI-STEM research

Medium (referred to educational medium in this study) was viewed as the way to convey information and connect subjects AI-STEM system. In the reviewed 63 studies, 50 of them mentioned the educational medium, including paper resource, entity resource (i.e., the material object in reality), computer system resource, web open resource, mobile phone resource and E-book resource (see Table 7 ). Among all educational mediums, computer system was the most frequently used in AI-STEM studies ( N  = 28), followed by entity resource ( N  = 10) and web open resource ( N  = 9). Additionally, mobile phone resource ( N  = 3), traditional paper resource ( N  = 1), and E-book resource ( N  = 1) was the infrequent medium to convey knowledge.

Among all the AI categories, ITS was most frequently used through computer system resource ( N  = 15), followed by educational robots through entity resource ( N  = 6), automation through computer system resource ( N  = 5), and learning prediction through web open resource (see Fig.  9 ).

AI categories under educational medium ( N  = 50)

Environment in AI-STEM research

Environment (referred to educational context in this study) served as an underlying context to influence the whole AI-STEM system. In the reviewed 63 studies, 51 of them mentioned the educational environment, including face-to-face environment, experimental learning environment, informal learning environment, web-based environment and augmented/virtual reality (see Table 8 ). Among the 51 studies, 33 studies were implemented in face-to-face environment, followed by web-based environment ( N  = 11) and experimental environment ( N  = 6). Two studies conducted in informal learning environment (McLurkin et al., 2013 ; Verner et al., 2020 ) and only one study conducted in augmented reality (Lin & Chen, 2020 ).

All categories of AI techniques were commonly applied in face-to-face environment, in which the most frequently used AI technology category was learning prediction ( N  = 10), followed by automation ( N  = 7), ITS ( N  = 6), and educational robots ( N  = 5). Moreover, compared to other AI categories, ITS was the most frequently used technique in the web-based environment ( N  = 7) (see Fig.  10 ).

AI categories under educational contexts ( N  = 51)

This review summarized the educational and technological effects of AI applications in AI-STEM research.

Educational effects and findings

From the educational perspective, 42 of the 63 reviewed articles reported the educational effects and findings when applying AI techniques in STEM education. Specifically, 30 out of the 42 articles reported the instruction and learning effects (e.g., learning performance, affective perception, higher-order thinking) of the application of AI techniques in STEM education. 12 articles out of 42 reported students’ learning behaviors and patterns by using AI-enabled data mining and learning analytics techniques.

The effect of learning performance

Among all the reviewed articles, 22 studies revealed the educational effects of AI technologies on students’ learning performance. Most of them showed significantly positive influence of AI techniques on the improvement of students’ learning performances ( N  = 20). For example, Wu et. al. ( 2013 ) investigated the effect of a context-aware ubiquitous learning system in a geosciences course and the results showed that context-aware ubiquitous learning system had significantly positive effects on the learning achievements of students. Thai et. al. ( 2021 ) conducted a cluster randomized study to examine the effect of My Math Academy, a digital game-based learning environment that provided personalized content on kindergarten students; the results revealed the significant improvement of learning gains, especially for the moderate-level students. Tehlan et. al. ( 2020 ) used a quasi-experiment approach to examine the effects of genetic algorithm-supported pair programming in a programming course; the results found that the students’ learning performances were significantly higher in pair programming than individual programming. Two articles reported insignificant results of the learning performance effects. Koć-Januchta et. al. ( 2020 ) used a quasi-experiment to compare the effect of AI-enabled E-book and common E-book in students’ biology learning, and the results showed that there was no significant difference of students’ learning gains between these two types of books. Also, Hellings and Haelermans ( 2020 ) conducted a randomized experiment to examine the effect of a learning analytics dashboard with predictive function in a computer programming course, but no significant improvement was found on student performance in the final exam.

The effect of affective perception

Among all the reviewed articles, a majority of studies revealed the educational effects of AI technologies on students’ affective perception, such as attitude, interest, and motivation ( N  = 17). On the one hand, students showed satisfaction and positive attitude towards the integration of AI technologies and STEM education. For example, Azcona et. al. ( 2019 ) used a questionnaire to find students’ positive feedbacks and attitudes towards the application of learning analytics in computer programming classes to detect and warn learning risks. Gavrilović et. al. ( 2018 ) evaluated student’s satisfaction of an AI-supported adaptive learning system in Java programming learning through the survey approach; the results revealed the positive feedbacks of students. On the other hand, the application of AI technologies also arouses students’ interests and motivation in STEM learning. For example, Balakrishnan ( 2018 ) used a mixed-method approach (i.e., questionnaire and interview) to examine the impact of a computer-based personalized learning environment (PLE) on engineering students’ motivation, and the results revealed the potential of PLE to engage students in learning with a strong sense of interest and motivation. Verner et. al. ( 2020 ) investigated students’ perceptions and attitudes towards an interactive robot tutor in science classes and found that the human–robot interaction fostered students’ active learning, maintain their attention and interest in the learning processes.

The effect of higher-order thinking

Among all the reviewed articles, some studies revealed the educational effects of AI technologies on students’ higher-order thinking ( N  = 7), such as problem-solving ability, computational thinking, and self-regulated learning skills. For example, Hooshyar et. al. ( 2015 ) employed a quasi-experimental design to examine the impact of a flowchart-based intelligent tutoring system (FITS) on students’ programming learning and found better improvement of problem-solving abilities in the FITS group than the control group. Lin and Chen ( 2020 ) found that students who used a deep learning-based AR system performed significantly better in computational thinking than those using an AR system without deep learning recommendation. Jones and Castellano ( 2018 ) utilized adaptive robotic tutors to promote students’ self-regulated learning skills and found that when a robotic tutor provided scaffoldings adaptively, more self-regulated learning behaviors were observed from students over the control condition without scaffoldings. García-Gorrostieta et. al. ( 2018 ) used experimental evaluation to test the effect of the automatic argument assessment on students’ computer engineering writing, and the results revealed that the argument assessment system helped students improve argumentation ability in their writing.

The effect of student learning pattern and behavior

Among 42 reviewed articles that mentioned educational effects and findings, 12 articles revealed students’ learning patterns and behaviors in STEM education by using AI-enabled data mining and learning analytics approaches. For example, Sapounidis et. al. ( 2019 ) detected 48 children’s preference profiles on tangible and graphical programming through latent class modeling; results found that the graphical programming was preferred by a majority of children, especially children in younger ages. Pereira et. al. ( 2020 ) used learning analytics (i.e., k-means, association rule algorithms) in the Amazonas to understand students’ behavior in introductory programming courses and found high heterogeneity among them. Three clusters of novice programmers were detected to explain how student behaviors during programming influenced the learning outcomes. Wang ( 2016 ) utilized data mining and learning analytics techniques (i.e., association rule, decision tree) to investigate college students’ course-taking patterns in STEM learning; the results found that the most viable course-taking trajectories is taking mathematics courses after initial exposure to subject courses in STEM.

Technological effects and findings

From the technological perspective, 24 of the 63 reviewed articles reported the technological effect and findings (e.g., efficiency of technology, accuracy of algorithm) when applying AI techniques in STEM education. For example, Çınar et. al. ( 2020 ) utilized multiple machine learning algorithms, including Support Vector Machines (SVM), Gini, k-Nearest Neighbors (KNN), Breiman’s Bagging, Freund and Schapire’s Adaboost.M1 algorithms, to automatically grade open-ended physics questions; the results reported that AdaBoost.M1 had the best performance with the highest accuracy of prediction models among all machine learning algorithms. Nehm et. al. ( 2012 ) used a corpus of biology evolutionary explanations written by 565 undergraduates to test the efficacy of an automated assessment program, Summarization Integrated Development Environment (SIDE); the results showed that, compared to human expert scoring, SIDE had better performance when scoring models were built and tested at the individual item level, and the performance degraded when suites of items or entire instruments were used to build and test scoring models. Bertolini et. al. ( 2021 ) employed five machine learning methods to quantify predictive efficacy of predictive modeling in undergraduate students’ outcome in biology. Results found that individual machine learning methods, especially logistic regression achieved a poor prediction performance while ensemble machine learning methods, in particular the generalized linear model with elastic net (GLMNET), achieved the high accuracy. Deo et. al. ( 2020 ) adopted a computationally efficient AI model called extreme learning machines (ELM) to predict weighted score and the examination score in engineering mathematics courses; the results showed that ELM outperformed in prediction with respect to random forest and Volterra.

Discussion and implications

Addressing research questions.

Although AIEd has attracted wide attention in educational research and practice, few research works have investigated the applications of AI in STEM education context. To gain a comprehensive understanding of the integration of AI in STEM education, this study conducted a systematic review of AI-STEM empirical research from 2011 to 2021. Grounded upon GST, we examined the AI technologies and applications in STEM education, the characteristics of other system elements (i.e., information, subject, medium, environment), the distribution of AI in these elements, and the effects of AI applications in STEM education. To answer the first question, we found a gradually increasing trend of AI applications in STEM education in the past decade. Furthermore, six categories of AI applications were located, namely learning prediction, ITS, student behavior detection, automation, educational robots, and others (i.e., AI text book, group formation). Regarding the characteristics of elements and the distribution of AI in these elements, first, we found all categories of AI techniques, especially student behavior detection, ITS, and learning prediction, were frequently applied in the learning contents of science and technology. Second, instructors usually involved in STEM education to support students and they used lecturing strategy the most frequently, followed by problem-based learning. Automation was only applied in the lecturing instruction mode and educational robots were most frequently applied in the problem-based learning mode. Third, a majority of AI techniques (except educational robots) were applied in higher education with medium and large scale of learners. The most frequently used AI in higher education were learning prediction and ITSs. Fourth, computer system resource was the most frequently used medium to convey knowledge, particularly when it was applied in ITSs and automation, while paper, mobile phone, and E-book resources were seldom used in AI-STEM research. Fifth, the face-to-face environment was mainly utilized to support all categories of AI applications, and web-based environment was most frequently used supported with ITSs.

Regarding the third question, this review summarized the educational and technological effects and findings of AI applications in STEM. From the educational perspective, the results showed that most of the AI applications had positive effects on students’ academic performance. However, insignificant improvements of learning outcomes were also found in two empirical studies (Hellings & Haelermans, 2020 ; Koć-Januchta et al., 2020 ). Moreover, most students held positive attitudes towards the use of AI technology in STEM education, and AI technologies aroused their interest and motivation as well. In other words, the AI applications are beneficial for fostering student’s active learning in STEM education. Moreover, the applications of AI techniques also contributed to the development of students’ higher-order thinking, e.g., computational thinking, problem-solving ability. In addition, AI techniques have great potential to assist instructors by detecting students’ learning patterns and behaviors in STEM education. From the technological perspective, the reviewed articles mainly reported a good efficiency and algorithm accuracy when applying AI in STEM education. Specifically, AI algorithms, especially ensemble machine learning methods, performed well in learning prediction, automation, and personalized recommendation. Overall, underpinned by the GST framework, this review presented an overview of recent trends of the field of AI-STEM, which guided the following educational, technological, and theoretical implications.

Educational implications

The emergence of AI indirectly influences the subject elements (e.g., instructor, learner) in STEM education, which in turn would eventually influence the educational practices and effects. First, AI has potential to transform the instructor–student relationships in STEM education from the instructor-directed to student-centered learning (Cviko et al., 2014 ). When AI is applied in STEM education, the role of instructor is expected to shift from a leader to a collaborator or a facilitator under the AI-empowered, learner-as-leader paradigm (Ouyang & Jiao, 2021 ). However, this review found that the instructor-centered lecturing mode was the most frequently used instructional strategy in AI-STEM studies, while other student-centered instructional strategies (e.g., the project-based learning, collaborative learning, game-based learning) appeared infrequently. One of the reasons centers on the complexity of integrating technology and pedagogy in STEM education (Castañeda & Selwyn, 2018 ; Jiao et al., 2022 ; Loveless, 2011 ). For example, ITS and automation techniques are usually designed based on behaviorism (Skinner, 1953 ) to support instructor’s knowledge delivery and exam evaluation, which may be challenging for instructors to use when integrating it in the student-centered instructional strategies. Recent research has started to balance pedagogical design and technological application in educational practices in order to achieve the goal of AI–instructor collaboration and student-centered learning when AI is integrated (Baker & Smith, 2019 ; Holmes et al., 2019 ; Roll & Wylie, 2016 ). Furthermore, another critical question is: would AI replace instructor responsibilities and roles in STEM education (Segal, 2019 )? In this review, we found that the role of instructor was still irreplaceable, because the instructor’s involvement existed in most of the AI-STEM research. Even though AI can free instructors from redundant tasks in STEM education, it still lacks the human ability to convey social emotion, solve critical problems, and implement creative activities (Collinson, 1996 ; Gary, 2019 ; Muhisn et al., 2019 ). Therefore, although AI techniques can bring opportunities to develop STEM education (Hwang et al., 2020 ), we cannot overstate the role of technology and overlook the essential role of instructor (Selwyn, 2016 ). Overall, instructors, as important subjects in the educational system, need to take agency to promote the pedagogical designs and strategies when applying AI technologies, in order to achieve a high quality of AI-STEM education (Cantú-Ortiz et al., 2020 ).

Technological implications

Although AI has the potential to enhance the instruction and learning in STEM education (Chen et al., 2020 ; Holmes et al., 2019 ), the development of AI-STEM requires a better fit between AI technologies and other system elements in STEM education. First, regarding the relationships between AI and information element, the results showed that most of the AI applications were used in science and technology learning contents, and educational robots and automation were not applied in engineering and mathematics learning contents. Since STEM education contains interdisciplinary knowledge and learning contents from different subjects, AI is usually restricted in specific learning contents or courses (Douce et al., 2005 ). Therefore, one of the future directions is to expand the commonality and accessibility of AI techniques in different STEM subjects and courses. Second, the range of AI applications was mainly located in higher education, while few of AI techniques were applied in other educational levels, especially in kindergarten. To some extent, due to the complex function and feedback mechanisms, most of the AI techniques (e.g., ITS, learning prediction) might be appropriate for adult learners. Hence, some interactive AI techniques, e.g., social robots, AI-enabled games, can be designed and developed to support young children’s STEM learning (Belpaeme et al., 2018 ; Zapata-Cáceres & Martin-Barroso, 2021 ). Therefore, the ease of use is also one of the important considerations in future development of AI technologies (Law, 2019 ; Xu & Ouyang, 2022 ). Third, most of the AI-STEM research was conducted through the traditional mediums (e.g., computer system resource) and contexts (e.g., face-to-face learning environment). A future direction is to create AI-empowered STEM learning environment through combining the advanced educational mediums (e.g., E-book) and contexts (AR/VR), in order to better represent and convey knowledge (Mystakidis et al., 2021 ).

Theoretical implications

Due to the complexity of AI-STEM system, this research used a theoretical framework based on GST to examine the multiple elements (i.e., AI technology, information, subject, medium, environment) in AI-STEM research. Compared to previous AIEd reviews that mainly focused on the technological perspective, GST provides a holistic view for us to consider the complex human, pedagogical, environment factors when applying AI in STEM education (Kitto, 2014 ; Von Bertalanffy, 1968 ). For example, we found that sometimes instructors did not engage in STEM education to support students, especially when applying ITSs and educational robots. It might reveal a new trend that some AI technologies (e.g., social robot, virtual agent) might have the potential to replace the original role of instructor and work as a new subject to individually convey knowledge (Xu & Ouyang, 2022 ). Additionally, the results showed the different characteristics of learner’s sample size when applying different AI techniques. Learning prediction was more likely to applied with a large scale of students, and educational robots were inclined to be applied with a small scale of students. The features of AI technologies might explain this phenomenon. For example, a data training process is necessary before learning prediction, which requires the support of algorithmic modeling techniques and large data sets (Agrawal & Mavani, 2015 ; Lee et al., 2017 ), while educational robots, as human–machine interaction technologies, seem more suitable and practicable for small-scale STEM learning (Atman Uslu et al., 2022 ; Belpaeme et al., 2018 ). Overall, the current study utilized the GST framework to examine the multiple elements in the complex AI-STEM system; it is suggested that different stakeholders, e.g., educators, technical developers, and researchers, can adopt the GST framework as a guide to comprehensively consider the complex elements when applying AI techniques in STEM education (Kitto, 2014 ; Von Bertalanffy, 1950 ).

Conclusion, limitation, and future direction

The application of AI technology in STEM education is an emerging trend, which is confronted with the challenge of integrating diverse AI techniques in the complex STEM educational system. Grounded upon a GST framework, this research reviewed the empirical AI-STEM studies from 2011 to 2021. Specifically, this systematic review examined (1) the categories of the AI element in the AI-STEM system; (2) the characteristics of other system elements (i.e., information, subject, medium, and environment) as well as the distribution of AI in these elements, and (3) the effects of AI in STEM education. Based on the results, the current work proposed educational, technological, and theoretical implications for future AI-STEM research, to better aid the educators, researchers and technical developers to integrate the AI techniques and STEM education.

There are three limitations in this systematic review, which lead to future research directions. First, although we searched the best-known scholar databases with the keywords relevant to AI-STEM, some biases might exist in the searching and screening process. Since AI-STEM is a highly technology-dependent field, some studies might only highlight the technology rather than the education context. Therefore, future studies can adjust the searching criteria to solve these problems. Second, from a system perspective, we used a GST framework to examine the multiple elements in the complex AI-STEM system, but we did not investigate the mutual relationships between elements. Therefore, the complex relationships between different elements (e.g., instructor–learner, learner–learner relationship) in AI-STEM system need to be further explored in order to gain a deep understanding of the application of AI in STEM education (e.g., Xu & Ouyang, 2022 ). Third, the current study only implemented a systematic review, a meta-analysis could be conducted in the future to report the effect sizes of recent empirical studies to gain a deeper understanding of the effects of the AI-STEM integration in an educational system. Overall, the potential of AI technology for enhancing STEM education is fertile ground to be further explored together with studies aimed at investigating the integration of technology and educational system.

Availability of data and materials

The data are available upon request from the corresponding author.

*Reviewed articles ( N  = 63)

Agrawal, H., & Mavani, H. (2015). Student performance prediction using machine learning. International Journal of Engineering Research and Technology, 4 (03), 111–113. https://doi.org/10.17577/IJERTV4IS030127

Article   Google Scholar  

Alabdulhadi, A., & Faisal, M. (2021). Systematic literature review of STEM self-study related ITSs. Education and Information Technologies, 26 , 1549–1588. https://doi.org/10.1007/s10639-020-10315-z

*Aldabe, I., & Maritxalar, M. (2014). Semantic similarity measures for the generation of science tests in Basque. IEEE Transactions on Learning Technologies, 7 (4), 375–387. https://doi.org/10.1109/TLT.2014.2355831

*Alemán, J. L. F. (2011). Automated assessment in a programming tools course. IEEE Transactions on Education, 54 (4), 576–581. https://doi.org/10.1109/TE.2010.2098442

Atman Uslu, N., Yavuz, G. Ö., & KoçakUsluel, Y. (2022). A systematic review study on educational robotics and robots. Interactive Learning Environments . https://doi.org/10.1080/10494820.2021.2023890

*Azcona, D., Hsiao, I. H., & Smeaton, A. F. (2019). Detecting students-at-risk in computer programming classes with learning analytics from students’ digital footprints. User Modeling and User-Adapted Interaction, 29 (4), 759–788. https://doi.org/10.1007/s11257-019-09234-7

Baker, T., & Smith, L. (2019). Educ-AI-tion rebooted? Exploring the future of artificial intelligence in schools and colleges. https://media.nesta.org.uk/documents/Future_of_AI_and_education_v5_WEB.Pdf

*Balakrishnan, B. (2018). Motivating engineering students learning via monitoring in personalized learning environment with tagging system. Computer Applications in Engineering Education, 26 (3), 700–710. https://doi.org/10.1002/cae.21924

Belpaeme, T., Kennedy, J., Ramachandran, A., Scassellati, B., & Tanaka, F. (2018). Social robots for education: A review. Science Robotics, 3 (21), eaat5954. https://doi.org/10.1126/scirobotics.aat5954

*Berland, M., Davis, D., & Smith, C. P. (2015). AMOEBA: Designing for collaboration in computer science classrooms through live learning analytics. International Journal of Computer-Supported Collaborative Learning, 10 (4), 425–447. https://doi.org/10.1007/s11412-015-9217-z

*Bertolini, R., Finch, S. J., & Nehm, R. H. (2021). Testing the impact of novel assessment sources and machine learning methods on predictive outcome modeling in undergraduate biology. Journal of Science Education and Technology, 30 (2), 193–209. https://doi.org/10.1007/s10956-020-09888-8

*Blikstein, P., Worsley, M., Piech, C., Sahami, M., Cooper, S., & Koller, D. (2014). Programming pluralism: Using learning analytics to detect patterns in the learning of computer programming. Journal of the Learning Sciences, 23 (4), 561–599. https://doi.org/10.1080/10508406.2014.954750

*Buenaño-Fernández, D., Gil, D., & Luján-Mora, S. (2019). Application of machine learning in predicting performance for computer engineering students: A case study. Sustainability, 11 (10), 1–18. https://doi.org/10.3390/su11102833

Bybee, R. W. (2013). The case for STEM education: Challenges and opportunities . NSTA Press.

Google Scholar  

Byrne, D., & Callaghan, G. (2014). Complexity theory and the social sciences . Routledge.

Cantú-Ortiz, F. J., Galeano Sánchez, N., Garrido, L., Terashima-Marin, H., & Brena, R. F. (2020). An artificial intelligence educational strategy for the digital transformation. International Journal on Interactive Design and Manufacturing, 14 (4), 1195–1209. https://doi.org/10.1007/s12008-020-00702-8

*Cao, X., Li, Z., & Zhang, R. (2021). Analysis on academic benchmark design and teaching method improvement under artificial intelligence robot technology. International Journal of Emerging Technologies in Learning, 16 (5), 58–72. https://doi.org/10.3991/ijet.v16i05.20295

Castañeda, L., & Selwyn, N. (2018). More than tools? Making sense of the ongoing digitizations of higher education. International Journal of Educational Technology in Higher Education, 15 (1), 1–10. https://doi.org/10.1186/s41239-018-0109-y

Chen, D., & Stroup, W. (1993). General system theory: Toward a conceptual framework for science and technology education for all. Journal of Science Education and Technology, 2 (3), 447–459. https://doi.org/10.1007/BF00694427

Chen, L., Chen, P., & Lin, Z. (2020). Artificial intelligence in education: A review. IEEE Access, 8 , 75264–75278. https://doi.org/10.1109/ACCESS.2020.2988510

*Chrysafiadi, K., & Virvou, M. (2013). PeRSIVA: An empirical evaluation method of a student model of an intelligent e-learning environment for computer programming. Computers & Education, 68 , 322–333. https://doi.org/10.1016/j.compedu.2013.05.020

*Çınar, A., Ince, E., Gezer, M., & Yılmaz, Ö. (2020). Machine learning algorithm for grading open-ended physics questions in Turkish. Education and Information Technologies, 25 , 3821–3844. https://doi.org/10.1007/s10639-020-10128-0

Cohen, L., Manion, L., & Morrison, K. (2005). Research methods in education . Routledge Falmer.

Collinson, V. (1996). Reaching students: Teachers ways of knowing . Corwin Press.

Crawford, J. L. (1974). A systems approach model for the application of general systems theory principles to education [Doctoral dissertation, University of Houston]. The University of Houston Institutional Repository. https://hdl.handle.net/10657/10661

Cviko, A., McKenney, S., & Voogt, J. (2014). Teacher roles in designing technology-rich learning activities for early literacy: A cross-case analysis. Computers & Education, 72 , 68–79. https://doi.org/10.1016/j.compedu.2013.10.014

*De-Marcos, L., Garcia-Cabot, A., Garcia-Lopez, E., & Medina, J. A. (2015). Parliamentary optimization to build personalized learning paths: Case study in web engineering curriculum. International Journal of Engineering Education, 31 (4), 1092–1105.

*Deo, R. C., Yaseen, Z. M., Al-Ansari, N., Nguyen-Huy, T., Langlands, T. A. M., & Galligan, L. (2020). Modern artificial intelligence model development for undergraduate student performance prediction: An investigation on engineering mathematics courses. IEEE Access, 8 , 136697–136724. https://doi.org/10.1109/ACCESS.2020.3010938

Douce, C., Livingstone, D., & Orwell, J. (2005). Automatic test-based assessment of programming: A review. Journal on Educational Resources in Computing, 5 (3), 4-es. https://doi.org/10.1145/1163405.1163409

Drack, M., & Pouvreau, D. (2015). On the history of Ludwig von Bertalanffy’s “general systemology”, and on its relationship to cybernetics—Part III: Convergences and divergences. International Journal of General Systems, 44 (5), 523–571. https://doi.org/10.1080/03081079.2014.1000642

Drigas, A. S., & Ioannidou, R. E. (2012). Artificial intelligence in special education: A decade review. International Journal of Engineering Education, 28 (6), 1366. http://imm.demokritos.gr/publications/AI_IJEE.pdf

*Ferrarelli, P., & Iocchi, L. (2021). Learning Newtonian physics through programming robot experiments. Technology, Knowledge and Learning, 26 , 789–824. https://doi.org/10.1007/s10758-021-09508-3

*Figueiredo, M., Esteves, L., Neves, J., & Vicente, H. (2016). A data mining approach to study the impact of the methodology followed in chemistry lab classes on the weight attributed by the students to the lab work on learning and motivation. Chemistry Education Research and Practice, 17 (1), 156–171. https://doi.org/10.1039/c5rp00144g

*García-Gorrostieta, J. M., López-López, A., & González-López, S. (2018). Automatic argument assessment of final project reports of computer engineering students. Computer Applications in Engineering Education, 26 (5), 1217–1226. https://doi.org/10.1002/cae.21996

Gary, K. (2019). Pragmatic standards versus saturated phenomenon: Cultivating a love of learning. Journal of Philosophy of Education, 53 (3), 477–490. https://doi.org/10.1111/1467-9752.12377

*Gavrilović, N., Arsić, A., Domazet, D., & Mishra, A. (2018). Algorithm for adaptive learning process and improving learners’ skills in Java programming language. Computer Applications in Engineering Education, 26 (5), 1362–1382. https://doi.org/10.1002/cae.22043

Graneheim, U. H., & Lundman, B. (2004). Qualitative content analysis in nursing research: Concepts, procedures and measures to achieve trustworthiness. Nurse Education Today, 24 (2), 105–112. https://doi.org/10.1016/j.nedt.2003.10.001

Guan, C., Mou, J., & Jiang, Z. (2020). Artificial intelligence innovation in education: A twenty-year data-driven historical analysis. International Journal of Innovation Studies, 4 (4), 134–147. https://doi.org/10.1016/j.ijis.2020.09.001

*Hellings, J., & Haelermans, C. (2020). The effect of providing learning analytics on student behaviour and performance in programming: A randomised controlled experiment. Higher Education, 83 (1), 1–18. https://doi.org/10.1007/s10734-020-00560-z

Holmes, W., Bialik, M., & Fadel, C. (2019). Artificial intelligence in education: Promises and implications for teaching and learning . Center for Curriculum Redesign.

Holstein, K., McLaren, B. M., & Aleven, V. (2019). Co-designing a real-time classroom orchestration tool to support teacher—AI complementarity. Journal of Learning Analytics, 6 (2), 27–52. https://doi.org/10.18608/jla.2019.62.3

*Hooshyar, D., Ahmad, R. B., Yousefi, M., Yusop, F. D., & Horng, S. (2015). A flowchart-based intelligent tutoring system for improving problem-solving skills of novice programmers. Journal of Computer Assisted Learning, 31 , 345–361. https://doi.org/10.1111/jcal.12099

*Hooshyar, D., Binti Ahmad, R., Wang, M., Yousefi, M., Fathi, M., & Lim, H. (2018). Development and evaluation of a game-based Bayesian intelligent tutoring system for teaching programming. Journal of Educational Computing Research, 56 (6), 775–801. https://doi.org/10.1177/0735633117731872

*Hsiao, I. H., Huang, P. K., & Murphy, H. (2020). Integrating programming learning analytics across physical and digital space. IEEE Transactions on Emerging Topics in Computing, 8 (1), 206–217. https://doi.org/10.1109/TETC.2017.2701201

Hwang, G. J., & Tu, Y. F. (2021). Roles and research trends of artificial intelligence in mathematics education: A bibliometric mapping analysis and systematic review. Mathematics, 9 (6), 584. https://doi.org/10.3390/math9060584

Hwang, G. J., Xie, H., Wah, B. W., & Gašević, D. (2020). Vision, challenges, roles and research issues of artificial intelligence in education. Computers and Education: Artificial Intelligence, 1 , 100001. https://doi.org/10.1016/j.caeai.2020.100001

*Ji, Y., & Han, Y. (2019). Monitoring indicators of the flipped classroom learning process based on data mining—Taking the course of “virtual reality technology” as an example. International Journal of Emerging Technologies in Learning, 14 (3), 166–176. https://doi.org/10.3991/ijet.v14i03.10105

Jiao, P., Ouyang, F., Zhang, Q., & Alavi, A. H. (2022). Artificial intelligence-enabled prediction model of student academic performance in online engineering education. Artificial Intelligence Review . https://doi.org/10.1007/s10462-022-10155-y

*Jones, A., Bull, S., & Castellano, G. (2018). “I know that now, I’ m going to learn this next” Promoting self-regulated learning with a robotic tutor. International Journal of Social Robotics, 10 (4), 439–454. https://doi.org/10.1007/s12369-017-0430-y

*Jones, A., & Castellano, G. (2018). Adaptive robotic tutors that support self-regulated learning: A longer-term investigation with primary school children. International Journal of Social Robotics, 10 (3), 357–370. https://doi.org/10.1007/s12369-017-0458-z

*Khan, I., Ahmad, A. R., Jabeur, N., & Mahdi, M. N. (2021). Machine learning prediction and recommendation framework to support introductory programming course. International Journal of Emerging Technologies in Learning, 16 (17), 42–59. https://doi.org/10.3991/ijet.v16i17.18995

Khandelwal, P., Srinivasan, K., & Roy, S. S. (2019). Surgical education using artificial intelligence, augmented reality and machine learning: A review. In A. Sengupta & P. Eng (Eds.), 2019 IEEE international conference on consumer electronics—Taiwan (pp. 1–2). IEEE.

*Kinnebrew, J. S., Killingsworth, S. S., Clark, D. B., Biswas, G., Sengupta, P., Minstrell, J., Martinez-garza, M., & Krinks, K. (2017). Contextual markup and mining in digital games for science learning: Connecting player behaviors to learning goals. IEEE Transactions on Learning Technologies, 10 (1), 93–103. https://doi.org/10.1109/TLT.2016.2521372

Kitto, K. (2014). A contextualised general systems theory. Systems, 2 (4), 541–565. https://doi.org/10.3390/systems2040541

*Koć-Januchta, M. M., Schönborn, K. J., Tibell, L. A. E., Chaudhri, V. K., & Heller, H. C. (2020). Engaging with biology by asking questions: Investigating students’ interaction and learning with an artificial intelligence-enriched textbook. Journal of Educational Computing Research, 58 (6), 1190–1224. https://doi.org/10.1177/0735633120921581

*Kose, U., & Arslan, A. (2017). Optimization of self-learning in computer engineering courses: An intelligent software system supported by artificial neural network and vortex optimization algorithm. Computer Applications in Engineering Education, 25 (1), 142–156. https://doi.org/10.1002/cae.21787

*Krämer, N. C., Karacora, B., Lucas, G., Dehghani, M., Rüther, G., & Gratch, J. (2016). Closing the gender gap in STEM with friendly male instructors? On the effects of rapport behavior and gender of a virtual agent in an instructional interaction. Computers & Education, 99 , 1–13. https://doi.org/10.1016/j.compedu.2016.04.002

Krasovskiy, D. (2020). The challenges and benefits of adopting AI in STEM education. https://upjourney.com/the-challenges-and-benefits-of-adopting-ai-in-stem-education

Krippendorff, K. (2004). Reliability in content analysis: Some common misconceptions and recommendations. Human Communication Research, 30 (3), 411–433. https://doi.org/10.1093/hcr/30.3.411

*Lacave, C., Molina, A. I., & Cruz-Lemus, J. A. (2018). Learning analytics to identify dropout factors of computer science studies through Bayesian networks. Behaviour and Information Technology, 37 (10–11), 993–1007. https://doi.org/10.1080/0144929X.2018.1485053

*Lamb, R., Hand, B., & Kavner, A. (2021). Computational modeling of the effects of the science writing heuristic on student critical thinking in science using machine learning. Journal of Science Education and Technology, 30 (2), 283–297. https://doi.org/10.1007/s10956-020-09871-3

Law, N. W. Y. (2019). Human development and augmented intelligence. In The 20th international conference on artificial intelligence in education (AIED 2019), Chicago, IL, USA.

Le, N. T., Strickroth, S., Gross, S., & Pinkwart, N. (2013). A review of AI-supported tutoring approaches for learning programming. In N. Nguyen, T. Van Do, & H. Le Thi (Eds.), Advanced computational methods for knowledge engineering (pp. 267–279). Springer. https://doi.org/10.1007/978-3-319-00293-4_20

Chapter   Google Scholar  

*Ledesma, E. F. R., & García, J. J. G. (2017). Selection of mathematical problems in accordance with student’s learning style. International Journal of Advanced Computer Science and Applications, 8 (3), 101–105. https://doi.org/10.14569/IJACSA.2017.080316

Lee, J., Jang, D., & Park, S. (2017). Deep learning-based corporate performance prediction model considering technical capability. Sustainability, 9 (6), 899. https://doi.org/10.3390/su9060899

Lee, J., Wu, A. S., Li, D., & Kulasegaram, K. M. (2021). Artificial intelligence in undergraduate medical education: A scoping review. Academic Medicine, 96 (11S), S62–S70. https://doi.org/10.1097/ACM.0000000000004291

Liang, J. C., Hwang, G. J., Chen, M. R. A., & Darmawansah, D. (2021). Roles and research foci of artificial intelligence in language education: An integrated bibliographic analysis and systematic review approach. Interactive Learning Environments . https://doi.org/10.1080/10494820.2021.1958348

*Lin, P. H., & Chen, S. Y. (2020). Design and evaluation of a deep learning recommendation based augmented reality system for teaching programming and computational thinking. IEEE Access, 8 , 45689–45699. https://doi.org/10.1109/ACCESS.2020.2977679

Liu, M., Li, Y., Xu, W., & Liu, L. (2017). Automated essay feedback generation and its impact on revision. IEEE Transactions on Learning Technologies, 10 (4), 502–513. https://doi.org/10.1109/TLT.2016.2612659

Loveless, A. (2011). Technology, pedagogy and education: Reflections on the accomplishment of what teachers know, do and believe in a digital age. Technology, Pedagogy and Education, 20 (3), 301–316. https://doi.org/10.1080/1475939X.2011.610931

*Maestrales, S., Zhai, X., Touitou, I., Baker, Q., Schneider, B., & Krajcik, J. (2021). Using machine learning to score multi-dimensional assessments of chemistry and physics. Journal of Science Education and Technology, 30 (2), 239–254. https://doi.org/10.1007/s10956-020-09895-9

*Mahboob, K., Ali, S. A., & Laila, U. E. (2020). Investigating learning outcomes in engineering education with data mining. Computer Applications in Engineering Education, 28 (6), 1652–1670. https://doi.org/10.1002/cae.22345

*Matthew, F. T., Adepoju, A. I., Ayodele, O., Olumide, O., Olatayo, O., Adebimpe, E., Bolaji, O., & Funmilola, E. (2018). Development of mobile-interfaced machine learning-based predictive models for improving students’ performance in programming courses. International Journal of Advanced Computer Science and Applications, 9 (5), 105–115. https://doi.org/10.14569/IJACSA.2018.090514

McLaren, B. M., Scheuer, O., & Mikšátko, J. (2010). Supporting collaborative learning and e-discussions using artificial intelligence techniques. International Journal of Artificial Intelligence in Education, 20 (1), 1–46. https://doi.org/10.3233/JAI-2010-0001

McLurkin, J., Rykowski, J., John, M., Kaseman, Q., & Lynch, A. J. (2013). Using multi-robot systems for engineering education: Teaching and outreach with large numbers of an advanced, low-cost robot. IEEE Transactions on Education, 56 (1), 24-33. https://doi.org/10.1109/TE.2012.2222646

Moher, D., Liberati, A., Tetzlaff, J., Altman, D. G., Prisma Group. (2009). Preferred reporting items for systematic reviews and meta-analyses: The PRISMA statement. PLoS Medicine, 6 (7), e1000097. https://doi.org/10.1371/journal.pmed.1000097.t001

Muhisn, Z. A. A., Ahmad, M., Omar, M., & Muhisn, S. A. (2019). The Impact of socialization on collaborative learning method in e-learning management system (eLMS). International Journal of Emerging Technologies in Learning, 14 (20), 137–148.

Murray, T. (2003). An overview of intelligent tutoring system authoring tools: Updated analysis of the state of the art. In T. Murray, S. B. Blessing, & S. Ainsworth (Eds.), Authoring tools for advanced technology learning environments (pp. 491–544). Springer. https://doi.org/10.1007/978-94-017-0819-7_17

*Myneni, L. S., Narayanan, N. H., & Rebello, S. (2013). An interactive and intelligent learning system for physics education. IEEE Transactions on Learning Technologies, 6 (3), 228–239. https://doi.org/10.1109/TLT.2013.26

Mystakidis, S., Christopoulos, A., & Pellas, N. (2021). A systematic mapping review of augmented reality applications to support STEM learning in higher education. Education and Information Technologies, 27 , 1883–1927. https://doi.org/10.1007/S10639-021-10682-1/FIGURES/10

*Nehm, R. H., Ha, M., & Mayfield, E. (2012). Transforming biology assessment with machine learning: Automated scoring of written evolutionary explanations. Journal of Science Education and Technology, 21 (1), 183–196. https://doi.org/10.1007/s10956-011-9300-9

Ouyang, F., & Jiao, P. (2021). Artificial intelligence in education: The three paradigms. Computers and Education: Artificial Intelligence, 2 , 100020. https://doi.org/10.1016/j.caeai.2021.100020

Ouyang, F., Zheng, L., & Jiao, P. (2022). Artificial intelligence in online higher education: A systematic review of empirical research from 2011 to 2020. Education and Information Technologies . https://doi.org/10.1007/s10639-022-10925-9

*Pereira, F. D., Oliveira, E. H. T., Oliveira, D. B. F., Cristea, A. I., Carvalho, L. S. G., Fonseca, S. C., Toda, A., & Isotani, S. (2020). Using learning analytics in the Amazonas: Understanding students’ behaviour in introductory programming. British Journal of Educational Technology, 51 (4), 955–972. https://doi.org/10.1111/bjet.12953

Pimthong, P., & Williams, J. (2018). Preservice teachers’ understanding of STEM education. Kasetsart Journal of Social Sciences, 41 (2), 1–7. https://doi.org/10.1016/j.kjss.2018.07.017

Rapoport, A. (1986). General system theory: Essential concepts & applications . CRC Press.

*Rodríguez Corral, J. M., Morgado-Estévez, A., Cabrera Molina, D., Pérez-Peña, F., Amaya Rodríguez, C. A., & CivitBalcells, A. (2016). Application of robot programming to the teaching of object-oriented computer languages. International Journal of Engineering Education, 32 (4), 1823–1832.

Roll, I., & Wylie, R. (2016). Evolution and revolution in artificial intelligence in education. International Journal of Artificial Intelligence in Education, 26 (2), 582–599. https://doi.org/10.1007/s40593-016-0110-3

*Saito, T., & Watanobe, Y. (2020). Learning path recommendation system for programming education based on neural networks. International Journal of Distance Education Technologies, 18 (1), 36–64. https://doi.org/10.4018/IJDET.2020010103

*Sapounidis, T., Stamovlasis, D., & Demetriadis, S. (2019). Latent class modeling of children’s preference profiles on tangible and graphical robot programming. IEEE Transactions on Education, 62 (2), 127–133. https://doi.org/10.1109/TE.2018.2876363

Segal, M. (2019). A more human approach to artificial intelligence. Nature, 571 (7766), S18–S18. https://doi.org/10.1038/d41586-019-02213-3

Selwyn, N. (2016). Is technology good for education? Polity Press.

Skinner, B. F. (1953). Science and human behavior . Macmillan.

*Spikol, D., Ruffaldi, E., Dabisias, G., & Cukurova, M. (2018). Supervised machine learning in multimodal learning analytics for estimating success in project-based learning. Journal of Computer Assisted Learning, 34 (4), 366–377. https://doi.org/10.1111/jcal.12263

*Suh, S. C., Anusha Upadhyaya, B. N., & Ashwin Nadig, N. V. (2019). Analyzing personality traits and external factors for stem education awareness using machine learning. International Journal of Advanced Computer Science and Applications, 10 (5), 1–4. https://doi.org/10.14569/ijacsa.2019.0100501

Tang, K. Y., Chang, C. Y., & Hwang, G. J. (2021). Trends in artificial intelligence supported e-learning: A systematic review and co-citation network analysis (1998–2019). Interactive Learning Environments . https://doi.org/10.1080/10494820.2021.1875001

*Tehlan, K., Chakraverty, S., Chakraborty, P., & Khapra, S. (2020). A genetic algorithm-based approach for making pairs and assigning exercises in a programming course. Computer Applications in Engineering Education, 28 (6), 1708–1721. https://doi.org/10.1002/cae.22349

*Thai, K. P., Bang, H. J., & Li, L. (2021). Accelerating early math learning with research-based personalized learning games: A cluster randomized controlled trial. Journal of Research on Educational Effectiveness, 15 (1), 28–51. https://doi.org/10.1080/19345747.2021.1969710

*Troussas, C., Krouska, A., & Sgouropoulou, C. (2021). A novel teaching strategy through adaptive learning activities for computer programming. IEEE Transactions on Education, 64 (2), 103–109. https://doi.org/10.1109/TE.2020.3012744

*Tüfekçi, A., & Köse, U. (2013). Development of an artificial intelligence based software system on teaching computer programming and evaluation of the system. Hacettepe Üniversitesi Eğitim Fakültesi Dergisi, 28 (2), 469–481.

*Verner, I. M., Cuperman, D., Gamer, S., & Polishuk, A. (2020). Exploring affordances of robot manipulators in an introductory engineering course. International Journal of Engineering Education, 36 (5), 1691–1707.

Von Bertalanffy, L. (1950). An outline of general system theory. British Journal for the Philosophy of Science, 1 , 134–165. https://doi.org/10.1093/bjps/I.2.134

Von Bertalanffy, L. (1968). General system theory: Foundations, development, applications . George Braziller.

*Vyas, V. S., Kemp, B., & Reid, S. A. (2021). Zeroing in on the best early-course metrics to identify at-risk students in general chemistry: An adaptive learning pre-assessment vs. traditional diagnostic exam. International Journal of Science Education, 43 (4), 552–569. https://doi.org/10.1080/09500693.2021.1874071

Walker, E., Rummel, N., & Koedinger, K. R. (2014). Adaptive intelligent support to improve peer tutoring in algebra. International Journal of Artificial Intelligence in Education, 24 (1), 33–61. https://doi.org/10.1007/s40593-013-0001-9

*Wang, T., Su, X., Ma, P., Wang, Y., & Wang, K. (2011). Ability-training-oriented automated assessment in introductory programming course. Computers & Education, 56 (1), 220–226. https://doi.org/10.1016/j.compedu.2010.08.003

*Wang, X. (2016). Course-taking patterns of community college students beginning in STEM: Using data mining techniques to reveal viable STEM transfer pathways. Research in Higher Education, 57 (5), 544–569. https://doi.org/10.1007/s11162-015-9397-4

Wohlin, C. (2014). Guidelines for snowballing in systematic literature studies and a replication in software engineering. In C. Wohlin (Ed.), Proceedings of the 18th international conference on evaluation and assessment in software engineering (pp. 1–10). ACM Press. https://doi.org/10.1145/2601248.2601268

*Wu, P., Hwang, G., & Tsai, W. (2013). An expert system-based context-aware ubiquitous learning approach for conducting science learning activities. Journal of Educational Technology & Society, 16 (4), 217–230.

*Xing, W., Pei, B., Li, S., Chen, G., & Xie, C. (2019). Using learning analytics to support students’ engineering design: The angle of prediction. Interactive Learning Environments . https://doi.org/10.1080/10494820.2019.1680391

Xu, W., & Ouyang, F. (2022). A systematic review of AI role in the educational system based on a proposed conceptual framework. Education and Information Technologies, 27 , 4195–4223. https://doi.org/10.1007/s10639-021-10774-y

*Yahya, A. A., & Osman, A. (2019). A data-mining-based approach to informed decision-making in engineering education. Computer Applications in Engineering Education, 27 (6), 1402–1418. https://doi.org/10.1002/cae.22158

*Yang, J., Devore, S., Hewagallage, D., Miller, P., Ryan, Q. X., & Stewart, J. (2020). Using machine learning to identify the most at-risk students in physics classes. Physical Review Physics Education Research, 16 (2), 20130. https://doi.org/10.1103/PhysRevPhysEducRes.16.020130

Yang, J., & Zhang, B. (2019). Artificial intelligence in intelligent tutoring robots: A systematic review and design guidelines. Applied Sciences, 9 (10), 2078. https://doi.org/10.3390/app9102078

*Yannier, N., Hudson, S. E., & Koedinger, K. R. (2020). Active learning is about more than hands-on: A mixed-reality AI system to support STEM education. International Journal of Artificial Intelligence in Education, 30 (1), 74–96. https://doi.org/10.1007/s40593-020-00194-3

*Yu, Y. (2017). Teaching with a dual-channel classroom feedback system in the digital classroom environment. IEEE Transactions on Learning Technologies, 10 (3), 391–402. https://doi.org/10.1109/TLT.2016.2598167

*Zabriskie, C., Yang, J., Devore, S., & Stewart, J. (2019). Using machine learning to predict physics course outcomes. Physical Review Physics Education Research, 15 (2), 20120. https://doi.org/10.1103/PhysRevPhysEducRes.15.020120

*Zampirolli, F. A., BorovinaJosko, J. M., Venero, M. L. F., Kobayashi, G., Fraga, F. J., Goya, D., & Savegnago, H. R. (2021). An experience of automated assessment in a large-scale introduction programming course. Computer Applications in Engineering Education, 29 (5), 1284–1299. https://doi.org/10.1002/cae.22385

*Zapata-Cáceres, M., & Martin-Barroso, E. (2021). Applying game learning analytics to a voluntary video game: Intrinsic motivation, persistence, and rewards in learning to program at an early age. IEEE Access, 9 , 123588–123602. https://doi.org/10.1109/ACCESS.2021.3110475

Zawacki-Richter, O., Marín, V. I., Bond, M., & Gouverneur, F. (2019). Systematic review of research on artificial intelligence applications in higher education—Where are the educators? International Journal of Educational Technology in Higher Education, 16 (39), 1–27. https://doi.org/10.1186/s41239-019-0171-0

*Zhang, Z., Liu, H., Shu, J., Nie, H., & Xiong, N. (2020). On automatic recommender algorithm with regularized convolutional neural network and IR technology in the self-regulated learning process. Infrared Physics and Technology, 105 , 103211. https://doi.org/10.1016/j.infrared.2020.103211

Zheng, R., Jiang, F., & Shen, R. (2020). Intelligent student behavior analysis system for real classrooms. In P. C. Center (Ed.), 2020 IEEE international conference on acoustics, speech and signal processing (pp. 9244–9248). IEEE.

*Zulfiani, Z., Suwarna, I. P., & Miranto, S. (2018). Science education adaptive learning system as a computer-based science learning with learning style variations. Journal of Baltic Science Education, 17 (4), 711–727. https://doi.org/10.33225/jbse/18.17.711

Zupic, I., & Čater, T. (2015). Bibliometric methods in management and organization. Organizational Research Methods, 18 (3), 429–472. https://doi.org/10.1177/1094428114562629

Download references

Acknowledgements

The authors would like to thank Luyi Zheng for her help on preliminary data analysis.

This work was supported by National Natural Science Foundation of China, No. 62177041.

Author information

Authors and affiliations.

College of Education, Zhejiang University, #866, Yuhangtang Rd., Hangzhou, 310058, Zhejiang, China

Weiqi Xu & Fan Ouyang

You can also search for this author in PubMed   Google Scholar

Contributions

WX conducted data collection and analysis as well as writing of the first draft of this manuscript. FO designed the research, designed and facilitated data analysis and revised the manuscript. Both authors read and approved the final manuscript.

Corresponding author

Correspondence to Fan Ouyang .

Ethics declarations

Competing interests.

The authors declare that they have no competing interests.

Additional information

Publisher's note.

Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Rights and permissions

Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article's Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/ .

Reprints and Permissions

About this article

Cite this article.

Xu, W., Ouyang, F. The application of AI technologies in STEM education: a systematic review from 2011 to 2021. IJ STEM Ed 9 , 59 (2022). https://doi.org/10.1186/s40594-022-00377-5

Download citation

Received : 11 June 2022

Accepted : 09 September 2022

Published : 19 September 2022

DOI : https://doi.org/10.1186/s40594-022-00377-5

Share this article

Anyone you share the following link with will be able to read this content:

Sorry, a shareable link is not currently available for this article.

Provided by the Springer Nature SharedIt content-sharing initiative

• Artificial intelligence
• Artificial intelligence in education
• STEM education
• General system theory
• Educational system

• Give to JMU

Can AI alleviate ‘wicked’ world problems?

Math professor illuminates the role of artificial intelligence in pandemics, poverty and gender disparity.

by Amy Crockett (’10)

SUMMARY: Hala Nelson, associate professor of mathematics, believes AI can serve the public good — if smart and effective design policies are developed.

As a child, Hala Nelson lost her hair in a missile explosion and sur vived the Lebanese Civil War while landmines lurked underfoot. Experiencing the dark side of humanity at a young age shaped her interest in human behavior and the nature of intelligence. Nelson also developed a passion for mathematics. Her father taught her math at home until she graduated from high school, practicing problems with her from a thick book written in French. “It was ingrained in me from my father that I have a ‘clean brain,’” she said. According to him, math was “the one clean science.”

From 2002-05, Nelson earned bachelor’s and master’s degrees in mathematics, but she felt this foundation had little to do with real life and her career goals. “I knew that if I stayed on the path of algebra and abstract ness, I would never be able to use my brain to solve worldwide problems,” said Nelson, now an associate professor of mathematics at JMU. The nature of conflict, and how humans use their resources, thoughts and emotions, still fascinated her.

Beirut photograph by the Associated Press; family photograph courtesy of Hala Nelson

Her soul-searching guided her pursuit of a doctorate in mathematics from New York University and postdoctoral teaching and research at the University of Michigan, Ann Arbor, but discontentment lingered. She yearned to use mathematical model ing to address a specific problem and make an immediate impact. Further exploration opened Nelson’s eyes to the worlds of data science, machine learning and AI — effec tively striking a balance between math and the humanities.

She soon felt called to build a bridge between the two realms, and to educate others on the benefits of AI. Before she could teach the curriculum to JMU students and assign the text, she needed to produce the subject matter. Nelson’s first book,  Essential Math for AI: Next-Level Mathematics for Efficient and Successful AI Systems ,  unifies and grounds AI in math.

This past winter, through the Center for Global Engagement, Nelson shared her expertise abroad at an international student conference in Bandung, Indonesia, leading a workshop on “A New World With COVID: Can AI Help?”

“The COVID-19 pandemic highlighted a lot of weaknesses in the global supply chain of very essential resources,” she said. “AI can assist with these logistics and in the field of operations research. ”  

AI can also help directly model the spread of disease, Nelson said, taking into account more factors and data than previous models and predictions. “Maybe the whole popula tion doesn’t have to sit at home and be iso lated,” she said. “The whole economy does not have to stop, and then the next 10 years we’re trying to recover from that because everybody had to be locked down.”

A global pandemic is an example of a wicked problem — a phrase coined by design theorists in the 1970s. “Wicked problems are phenomena that are so com plex that it’s hard to even define them and almost impossible to solve them,” said  Se á n McCarthy, professor of writing, rhetoric and  technical communication and director of the Cohen Center for the Humanities. “It’s a term that helps us think about problems beyond a simple cause-and-solution sort of scenario, where these problems are con stantly changing. It’s hard to get a handle on what they are; they’re slippery.” 

Nelson said COVID-19 illuminated another wicked problem — poverty. Lower income people were the most affected by the virus, in sickness, access to vaccines and job losses. AI has a useful role in combining locational  and human information to more precisely pinpoint vulnerable groups, she said.

To further support less advantaged popu lations, AI holds the potential to increase access to education when demand for teach ers is high. In agriculture, AI has the capa bility to examine crops and find new spe cies of plants that grow in different soils. To protect crops and avoid hunger, the technol ogy can also simulate weather patterns to forecast an incoming cold front. “If you give people access to better health, education and food, AI can facilitate all of this and coun teract the effects of poverty,” Nelson said.

When studying the wicked problem of gender inequality through the lens of AI, Nelson said the most important factor is the quality of the data. If the data entered into an algorithm is gendered, it can affect whether a person receives a loan or a job. “We have to be careful that we don’t reduce our whole humanity into walking, talking vectors that we feed into algorithms that make decisions based on numbers or scores documented by data companies,” she explained. “If our data is correctly represented and clean of biases, then AI will help remove gender disparities, because then decisions will be really based on qualifications.” In the male-dominated fields of math and science, Nelson is closing the gender gap. She credits her father for her aspirations and accomplishments. “I never thought that I could not do it, or I never felt like I needed to prove myself to anyone,” she said. “He did me the favor of confidence.” This semester, alongside political sci ence professor Bernie Kaussler, Nelson and McCarthy are co-teaching Hacking for Diplomacy. The innovative course challenges student teams to solve specific problems posed by U.S. government agen cies related to business, logistics, technol ogy or human resources. The class takes an interdisciplinary approach, mixing social science, communication and math.

“I’m lucky that I get to do this program with really smart professors who are really good teachers and are able to work on highly complex problems that the next generation of leaders who we are teaching will need to know how to solve,” McCarthy said. “Hala is amazing at inspiring students to take their theoretical and technical understanding of mathematical concepts and apply them to the real world.”

Nelson is working on her next book,  Foundations of AI and Data , which is set to release in 2025. In the preface, she writes, “The chal lenges facing humanity are the same as ever: sustainable energy, food supply, clean water, access to education and health care, security and defense, and climate change.”

She estimates that climate change is the biggest dilemma. However, the difference in analyzing climate change today com pared to past decades is that now more than 8 billion people live in an interconnected, global society. AI has the computational power to address and model a wicked prob lem of this scale.

“We’re not going to scare people and say we’re all going to die in 2035, right? We’re not gonna do that,” Nelson said. “Countries can sit down now and look at their data — this is exactly how much resources we have, and this is what we can do with them.”

Nelson maintains a bright outlook on har nessing AI to serve the public good — if smart and effective design policies are developed.

“Every jump in human society has hap pened when humans were able to automate a process,” she said. “The more advanced we become, the more problems we’re able to solve and the more we’re able to help people and elevate their standard of living. ”

Back to Top

Published: Monday, October 23, 2023

Last Updated: Friday, November 17, 2023

Related Articles

• Expenditures
• Accessibility
• Social Media

Machine Learning, Modeling, and Simulation: Engineering Problem-Solving in the Age of AI

Demystify machine learning through computational engineering principles and applications in this two-course program from MIT . Leveraging the rich experience of the faculty at the MIT Center for Computational Science and Engineering (CCSE), this program connects your science and engineering skills to the principles of machine learning and data science. With an emphasis on the application of these methods, you will put these new skills into practice in real time.

Alliances members receive a 15% discount on the course with promo code CSAIL15 .

Friday Roundup

The AI Health Friday Roundup highlights the week’s news and publications related to artificial intelligence, data science, public health, and clinical research.

November 17, 2023

In this week’s Duke AI Health Friday Roundup: testing GPT-4’s diagnostic chops; yeast with 50% synthetic genome survives, replicates; roles for AI in clinical trials; role of pets in zoonotic spillover; vaccine status, bias, and perceptions of risk; potential for bias in radiological deep learning models; what rats remember; developing standards for health-related chatbots; how publishing professionals perceive recent changes in social media; much more:

AI, STATISTICS & DATA SCIENCE

• “In this pilot assessment, we compared the diagnostic accuracy of GPT-4 in complex challenge cases to that of journal readers who answered the same questions on the Internet. GPT-4 performed surprisingly well in solving the complex case challenges and even better than the medical-journal readers. However, performance did appear to change between different versions of GPT-4…Although it demonstrated promising results in our study, GPT-4 missed almost every second diagnosis.” A study by Eriksen and colleagues , published in NEJM AI , reports findings of a comparison of the diagnostic acuity of GPT-4 versus a group of medical-journal readers.
• “…our study demonstrates that biases in the chest radiography foundation model related to race and biologic sex led to substantial performance disparities across protected subgroups. To minimize the risk of bias associated with use of foundation models in critical applications such as clinical decision-making, we argue that these models need to be fully accessible and transparent.” An article published this past September in Radiology: Artificial Intelligence by Blocker and colleagues examines the potential for bias in foundation models applied to chest radiography.
• “AI offers enormous potential, yet rigorous validation and regulatory oversight are essential to ensure that deployment is safe, effective, and ethical in the clinical trials ecosystem. Not only do model outputs need to provide accurate assessment of health state used for evaluating treatment benefit and risk, but the framework must also address risks related to data privacy, security, and bias.” An editorial published this week in JAMA by Hernandez and Lindsell examines the potential roles AI could play in the clinical trials enterprise.
• “To develop a reporting guideline for chatbot assessment studies, we have gathered a diverse group of international stakeholders including statisticians, research methodologists, reporting guideline developers, natural language processing researchers, journal editors, chatbot researchers and patient partners. The development of the chatbot assessment reporting tool (CHART) is registered with the EQUATOR (enhancing the quality and transparency of health research) international network. This guideline will generate a reporting checklist and flow diagram by adhering to robust methodology, as well as the evidence-based EQUATOR toolkit on developing reporting guidelines.” A brief report published in Nature Medicine by Huo and colleagues describes the creation of partnership for developing standards for reporting on evaluations of AI chatbots designed to provide health-related advice.

BASIC SCIENCE, CLINICAL RESEARCH & PUBLIC HEALTH

• “Rats also encode spatial information in the hippocampus. But it’s been impossible to establish whether they have a similar capacity for voluntary mental navigation because of the practical challenges of getting a rodent to think about a particular place on cue, says study author Chongxi Lai…. In their new study, Lai, along with Janelia neuroscientist Albert Lee and colleagues, found a way around this problem by developing a brain-machine interface that rewarded rats for navigating their surroundings using only their thoughts.” Science’s Catherine Offord reports on recent research that suggests rats are able to use their imaginations to navigate through environments they’ve previously visited.
• “Although the process of making the cells was time-consuming, what really slowed things down is debugging, Boeke says. Researchers first had to test whether each yeast cell with a new synthetic chromosome in it was viable — meaning it could survive and function normally — then fix any problems by tweaking the genetic code. When two or more synthetic chromosomes are inside the same cell, this can lead to new bugs that must be fixed, so the debugging problem becomes more complex as the process proceeds.” Nature’s Katherine Bourzac reports on a genetic engineering milestone: the successful survival and reproduction of a strain of yeast with a 50% synthetic genome.
• “This analysis finds that COVID-19 and the drug-overdose epidemic were major contributors to the widening gender gap in life expectancy in recent years. Men experienced higher COVID-19 death rates for likely multifactorial reasons, including higher burden of comorbidities and differences in health behaviors and socioeconomic factors, such as labor force participation, incarceration, and homelessness. Differentially worsening mortality from diabetes, heart disease, homicide, and suicide suggest that chronic metabolic disease and mental illness may also contribute.” A research letter published this week in JAMA-Internal Medicine by Yan and colleagues has more bad news about US life expectancy, as an existing gender-based gap continues to widen.
• “Because of their close proximity with people, companion animals and peri-domestic wildlife occupy a key position in the epidemiological networks of many zoonotic pathogens. Surveillance of those zoonotic pathogens should include companion animals and peri-domestic wildlife, and research should combine ecological approaches with molecular approaches to understand their roles as epidemiological reservoirs (e.g., dogs for rabies viruses) or bridges (e.g., horses for bat-borne paramyxoviruses).” A review article published in Science Translational Medicine by Gamble and colleagues examines the potential for pets and other domestic or semi-domestic animals to pose risks for zoonotic infections.

COMMUNICATION, Health Equity & Policy

• “One year into Elon Musk’s acquisition of X (formerly known as Twitter), engagement metrics are down across the board, with app downloads down 38%, web traffic down 7% globally and down 11.6% in the U.S. “What is UP with Twitter?” “What about Threads?” “Is Academic Twitter disappearing?” “Could BlueSky be the replacement?” Are your metrics changing? These are  questions we continue to hear in personal conversations, organizational meetings, industry articles, and in gatherings of the Society for Scholarly Publishing (SSP) MarComm committee.”  An article at Scholarly Kitchen plumbs the publishing community’s response to recent tumult in the social media landscape.
• “The authors’ analysis revealed that unvaccinated individuals who identified strongly with their unvaccinated status were more likely to remember their earlier estimation of the risk as lower than it actually was. Conversely, and more markedly, those who had been vaccinated overestimated their earlier perception of their risk of catching the disease.” An editorial published this week in Nature describes a recent investigation that finds an association between COVID vaccination status and bias in how people recall perceived risk of the disease.
• “Drawing from political science orthodoxy, deliberative democracy theory, and the concept of infrastructure, Wong shows how Big Tech platforms affect our rights, interests, and values in multiple and often hidden ways. With this framing, she convincingly argues that we should hold these corporations accountable to the ethos and full spectrum of human rights, not just to singular issues such as freedom of expression or privacy, and advocates for more transparency, participation, and oversight in their decision-making processes.” A book review published in Science by Duke’s Nita Farahany examines Wendy Wong’s We, the Data: Human Rights in a Digital Age.
• “We find that freelancers in highly affected occupations suffer from the introduction of generative AI, experiencing reductions in both employment and earnings. We find similar effects studying the release of other image-based, generative AI models. Exploring the heterogeneity by freelancers’ employment history, we do not find evidence that high-quality service, measured by their past performance and employment, moderates the adverse effects on employment. In fact, we find suggestive evidence that top freelancers are disproportionately affected by AI.” A paper by Hui and colleagues , available as a preprint from SSRN, examines associations between the emergence of publicly available generative AI tools and employment prospects for freelance writers and editors.

Webinar: MIT xPRO Machine Learning, Modeling, and Simulation: Engineering Problem-Solving in the Age of AI

The advent of big data, cloud computing, and machine learning is revolutionizing how many professionals approach their work. These technologies offer exciting new ways for engineers to tackle real-world challenges. But with little exposure to these new computational methods, engineers lacking data science or experience in modern computational methods might feel left behind.

In order to ensure that the engineering workforce has continual training and access to the latest knowledge, MIT xPRO launched the two-course certificate program: Machine Learning, Modeling, and Simulation: Engineering Problem-Solving in the Age of AI.

In this interactive and engaging webinar, MIT's Youssef Marzouk will discuss:

• Who this program is meant for
• Who is involved in this program 
• How the courses are delivered 
• What you will learn from the program 
• And how to enroll in the next offering 

By attending the live webinar, you will also be able to participate in the live Q&A following the presentation. Register for this special free virtual event by filling out the form. You'll receive a confirmation e-mail with further details about the webinar. is a professor in the Department of Aeronautics and Astronautics at MIT and co-director of the CCSE. He is also director of MIT’s Aerospace Computational Design Laboratory.

Register for webinar >>