9 Takeaways from the J-WEL AI Innovators Series

The six-part AI Innovators Series from the MIT Jameel World Education Lab (J-WEL) brought expert thinking from MIT to J-WEL members and the public on the evolving and potential role of AI in education. The discussions featured evaluations of generative Artificial Intelligence (AI) opportunities and issues from education thinkers, organizations creating new AI-based education tools, education startup companies leveraging AI for education, and teachers using AI in classrooms. A summary of key points from these discussions follows below, and those interested can call up and view each of the videos in the series via the links below.
There were many important points made by the presenters in the series.
Nine takeaways rose to the top:
1. Understand What AI Is: It is important to understand the current capabilities and limits of generative AI. It is not yet (at this stage) capable of true thinking or understanding; it is an algorithm-driven process where massive databases of writing are searched for the next logical word in a sequence of words. As a result, while AI can repeat existing knowledge, it is still limited, so far, in its ability to generate more creative and new ideas. It can be a useful tool for conveying existing information, but despite programs that allow it to apply less than the most used (and therefore more disruptive) word selections, it has limits in creating truly new knowledge. That doesn’t mean it’s without value – offering summaries of existing knowledge can be quite useful and provide a platform to build on for additional stages of thinking.
2. AI is an Education Tool: Generative AI can enable individualized and personalized learning, which can also be organized to improve learning retention. AI tools, through AI-aided analysis, can also identify student learning patterns and help motivate students by identifying their interests. AI can also use bot-like interfaces to improve interactive engagement, provide ‘tutoring’ and support inquiry learning. Finally, AI can address data silos with bot and other interfaces to improve data retrieval for building course materials.
3. Combine AI and Robotics for Social Learning: AI has a major limit: it doesn’t engage in personal, human-based communication, yet learning is in large part social and relational between people. How can this fundamental challenge to AI’s utility be overcome? Obviously, AI can be a tool used by teachers who provide the element of human interaction. An alternative is personalized robotics, programmed to mimic human emotions, voices and expressions, to deliver AI-derived content. After all, teachers can’t be everywhere. The robots are programmed to interact with the particular student, to act as peers in providing support and as teachers in conveying information. The robots have limits and cannot replace human teacher and peer exchanges but can be used to supplement them.
4. Teach Students to be AI Fluent: It is important to educate students about this significant new force entering education as well as many job fields, but it is not enough to simply tell them about it. They need to reach the stage of actually using generative AI to truly understand its value and potential harm, so this hands-on work needs to be part of the new AI education program. AI literacy is not enough. Free apps available online make this maker approach practical. AI fluency builds on constructivist education concepts, so students undertake learning by doing, moving past simple understanding to a creative phase for real working knowledge of the new tool.
5. Digital Tutors have Real Potential: AI has potential for a new kind of enhanced digital tutoring, far less constrained than pre-AI digital tutoring efforts. It aims to embody both adaptive and personalized learning, as well as problem-based learning-by-doing, each of which represents a noteworthy educational reform. One-on-one personal tutoring has long been understood to be much more effective compared to classroom instruction. While the relationship-building aspects of human teachers cannot be replaced, AI-enhanced digital tutoring with personalized and learning-by-doing features looks promising.
6. Apply AI to Train the Trainers: AI can record and observe both human teaching and combined AI and human teacher learning sessions with students. And AI can collect and analyze assessment results. Both can be compiled and put into a database, analyzed, and best practices for best teaching practices can be identified. For teachers who are having difficulty conveying material, AI can be applied to help identify for them their problems and offer possible solutions. AI thus can perform a role in training the trainers.
7. Manage the Database to Control for Hallucinations: AI can hallucinate (provide misinformation) and it may not be easy for students to identify correct versus incorrect content – this is a major problem for generative AI. However, creating information boundaries to help assure accurate content for the AI to draw on can help control for the problem. Having AI operate within carefully assembled databases where the information is known to be reliable, and putting the AI onto tasks which are encompassed within that database and are in effect rules-based – such as solving types of math problems – can help manage the AI misinformation problem.
8. Create Communities of Practice for Teachers: Educators learn much from each other, so creating communities for educators around AI issues is a key step. In these discussions, teachers can share their AI experiences, read and exchange best ideas, and learn what works and what doesn’t. These communities of educators can provide important input to and be a part of the policies that schools, school districts and state and federal governments develop for AI. In sum, creating communities of practice for AI within schools, with teachers ready to experiment together, who can share what’s working, can be important in managing AI introduction and use in schools.
9. Don’t Ban It: The widespread threat of cheating by students using AI is a real one. But banning AI is likely counterproductive. Bans will create disparities between those students and schools with bans and those that have learned to implement AI’s potential benefits. Bans risk leaving schools and students behind. Approaches are now being developed to manage the cheating threat. Much better is to use teacher communities of practice to seek thoughtful and ethical ways to introduce AI.
Dive into the full AI Innovators Series:
Want to learn more about one of the takeaways shared above? Read the summaries of each session below and watch the full videos from the AI Innovator Series.
Imagining Education with Generative AI
MIT Professor Justin Reich demystified AI in the first session of the series. He provided an introduction to what generative AI is – it is a wording predictor, it doesn’t understand or think. Using massive databases of human writing, it predicts the next word based on the most used words that typically follow a sequence of words. It does this over and over. The larger the database of writing, the better the generative AI, using algorithms behind large language models (LLMs) that assemble logical and intelligible sequences of words. With this understanding of how it works, we can start to understand some of the limits of generative AI as well as what it can do – deliver existing knowledge and understanding, but despite adding features to include less than perfect word fits, it is still a work in progress for it to offer more novel and creative ideas. Because its systems don’t yet possess a full concept of accuracy they can also deliver misinformation “hallucinations.”
Innovations in AI for Education
Dean for Digital Learning Cynthia Breazeal discussed a challenge in AI: it doesn’t engage in personalized, human-based communication, yet learning is in large part social and relational between people. How can this fundamental challenge to AI’s utility be overcome? Her research group uses personalized robotics to deliver AI-derived content, working one-on-one with young children on language acquisition. The robots are programmed to interact with students, to act as peers in providing support and as teachers in conveying information. This interaction leads to more personal learning experiences, enhancing learning. While teachers are best at such interaction, they are limited in their ability to engage in one-on-one instruction and robots can be a supplement. She also spoke about the need for AI fluency. It will be important to teach students about this new force entering education and many job fields, but it is not enough to just tell them about it. They need to actually make use AI to understand its value and limits, so creating AI needs to be part of this new AI education program. Available and free AI maker apps can be enablers. This approach builds on constructivist education concepts to undertake learning by doing.
The Innovations in AI You Imagine
At the end of Breazeal’s session, she closed with a five-year vision AI-empowered education. In this follow up session attendees reflected on that vision, shared their own, and explored how their institution can contribute to the future of education and AI. Joining from MIT were Hae Won Park and Sharifa Alghowinem, research scientists working with Prof. Breazeal’s research group. Park elaborated on the research results for the combination of robotics with AI working with early elementary school students. She said they found interaction with social robots – robotic companions – activates social thinking. This social aspect uses different parts of the brain than analytical thinking – and social thinking can help leverage analytical thinking for better learning outcomes. Using the robot social agent as a companion can lead to a socio-emotive relationship. Alghowinem has worked on providing students in middle and high schools training in AI fluency - having students not simply learn about but actually create AI in order to understand it. She said that the curriculum they have been building on AI allowed students to play with AI and build models of AI. The students can pursue project-based learning that integrates policy and ethics into the AI they are making which helps them understand the potential harms as well as benefits of AI. The curriculum that has been developed includes using AI so students can design digital citizenship projects.
Emerging AI Innovations: New Tools from the MIT Ecosystem
Cerine Hamida is the founder of a new company, Empower 1for1, and Salome Aguilar and Bernardo García are founders of a new NGO, JANN, both applying AI for education in the Global South. Hamida spoke of her new company’s efforts to bring an AI-based digital tutor to math and science education for secondary students in Tunisia. It offers a personalized learning experience that adapts to each student's proficiency level, offers interactive problem-solving, and provides 24/7 advice-giving support from a generative AI conversational professor. This AI-driven tutor provides real-time assistance, ensuring students receive context-specific guidance and support as they progress from basic to complex topics. Aguillar and Garcia’s NGO, JANN, combines online technology with personal, small group teaching to minimize implementation costs while broadening impact. It recruits university students as volunteer math tutors, providing free online math tutoring for groups of 5 children in primary and middle schools in Mexico, and has now reached over 23,000 students. JANN is now employing AI that reviews every tutoring session with students and then identifies both best practices being used by tutors and problems in connecting with students. This AI-based analytical approach is then applied in both training tutors and improving their performance. The AI enhancement is improving program quality.
How EdTech Startups Are Leveraging AI for Education
Jean Hammond, Deepak Verna and Joy Dasgupta are founders and CEOs of entrepreneurial EdTech companies introducing AI for education. Hammond discussed LearnLaunch, her venture capital firm that aims to promote better outcomes in education by building AI-based EdTech companies as well as helping to scale them. LearnLaunch is now supporting eight companies focused on AI for education. The firms are finding that AI tools can discover student learning patterns and help motivate them by identifying their interests; AI can also use bot interfaces to improve interactive engagement. Verna discussed his firm English Helper, an EdTech company with a mission to provide innovative and affordable learning for English literacy. It reaches 139,000 schools and 26 million students in 9 countries, led by India. Its “Read-To-Me” platform provides multisensory exposure to words and enables word recognition, also helping with development of pronunciation and oral skills. To supplement this platform, it launched a new AI-powered digital tutoring system for listening, speaking and writing – the core language skills. Desgupta discussed his company, GyanAI, which has developed a toolset to bring AI to higher education, teachers, and individual learners. It applies its AI language model for research and content development and works with educational institutions and companies to apply AI that its users have identified that is both available and contains reliable, accurate information. To avoid hallucinations and misinformation and to generate trust it avoids Large Language Models. Instead, the student or instruction designer selects the information sources she or he wants to use.
Applying AI to Education
In the final session, MIT Professor Eric Klopfer and Research Scientist Kate Moore discussed applications of AI for schools and teachers. Klopfer noted at the outset that augmenting people with technologies opens new opportunities and abilities; the tasks that humans and machines can do together are much larger than those either can do alone. He argued that people won’t be replaced by AI, they will be replaced by people who know how to harness AI wisely to create value. He discussed a roadmap for creating guiding principles for how we use AI in schools, including ideas of how to move beyond AI-driven tutors focused on short-term retention and quick recall to deeper learning. We require learning that is more durable and comprehensive, and we need to think about how to obtain the value added that people plus AI can produce. The key issue, then, for AI education is how we get to this combined model of teachers, peer-to-peer, and AI. Moore discussed a project called “Everyday AI,” which is a middle school and high school curriculum for AI literacy and fluency. The project has also been working on a teacher professional development program for AI that connects teachers, who are often isolated from each other, through book club-like reading and discussion sessions, along with an opportunity to practice during summers teaching this AI curriculum. Both presenters said their most important finding so far is the importance of creating communities of practice within schools, both for teachers to experiment together and to help manage AI introduction and use.