What do teachers want from AI? How can AI make the lives of students and teachers better? What new tools actually work?
AI scholars, educators, school administrators, and startup founders grappled with these questions at the second AI+Education Summit, hosted by Stanford HAI and the Stanford Accelerator for Learning on Feb. 1. The event highlighted the latest advances in AI technologies for students and teachers, offered insights on the future of teaching strategies and student assessment, and surfaced ethical and safety issues.
“We think education is one of the key areas where AI is going to make an impact on the world,” said HAI Vice Director James Landay to a standing-room-only crowd. “The changes that AI and foundation models are bringing to education is almost going to force educational reform.”
“We knew AI was coming,” added Daniel Schwartz, dean of Stanford Graduate School of Education (GSE) and faculty director of the Stanford Accelerator for Learning. “We didn’t know it would come this fast and this big. This could completely disrupt education – check your assumptions of what you should learn and how you should think.”
The day included panel discussions, keynotes, lightning round conversations, and a poster session featuring new research. Here we pulled together insights from speakers on what issues teachers currently face and exciting new research that could make a difference.
Improving AI Literacy
Many teachers are understandably skeptical about AI and its effects on student learning, said Amanda Bickerstaff, co-founder and CEO of AI for Education and a former high school teacher. She suggested starting with comprehensive AI literacy training. Teachers and school leaders need to know what AI is, its capabilities and limitations. “That’s No. 1,” she said. “They don’t need to be AI experts. Teachers should be expert at teaching, and they should be augmented by technology to help them teach better.”
Tools That Solve Real Problems
What could help free teachers’ time for more learning and less paperwork? “We don’t need a thousand tools,” Bickerstaff said. “We need ones that are really helpful.” Candace Thille, associate professor (teaching) at Stanford GSE and faculty director of the Adult and Workforce Learning initiative of the Stanford Accelerator for Learning, identified several paths for AI assistance: grading, lesson planning, creating new and interesting questions and problems. Once teachers identify a real need, developers must solve that problem with the work processes and tools already in use, incorporating teacher input along the way. “That’s the only way to do it. You can’t add one more thing onto a teacher’s role,” she said.
One toolkit now available to teachers is CRAFT, which offers AI literacy resources for high school teachers in any subject, said Victor Lee, associate professor at Stanford GSE and faculty lead for AI + Education at the Stanford Accelerator for Learning. “Our goal was to be highly responsive to teachers. We co-designed with teachers to build the resources that are going to help them address AI in their disciplines.”
Safe and Smart Rollout
Teachers and school systems want guidance on responsible use and more help on which tools are safe and reliable for students. As panelists pointed out, U.S. senators just this week took social media executives to task for algorithms that harm young people’s mental health. How can we learn lessons from social media to prevent similar harm to students using AI tools? Pat Yongpradit, the chief academic officer of Code.org, says his organization has developed AI guidance toolkits that cover both AI in instruction and in policy. Keith R. Krueger, CEO of the Consortium for School Networking (CoSN), said his association developed a “readiness” assessment for school districts to determine how prepared they are to roll out AI, covering operations, data, security, legal risk, and more. “There’s a lot of confusion about where to get started,” he said.
Efficiency at What Cost?
While many AI education tools target efficiency, some panelists challenged that goal. “If efficiency means teachers will now have to serve 150 students instead of 30, do they want that?” asked Dora Demszky, assistant professor of education at Stanford GSE. “Efficiency doesn’t always mean it’s going to lead to better life and work conditions.” Additionally, she noted, certain aspects of teaching should never be optimized for speed. “Being able to learn new skills efficiently is super important, but you can’t think about relationship building, for example, as a function of efficiency.” Added Ge Wang, associate professor of music in the Stanford School of Humanities and Sciences, we should critically evaluate our desires in the first place and the means to achieve them. “What are we doing with AI, what should we be doing with AI?” he asked the audience.
Latest Stanford Research
The summit included new research from leading Stanford AI and education scholars. Emma Brunskill, associate professor of computer science in the Stanford School of Engineering, explores how reinforcement learning improves AI tutors and assistants. In one project, a chatbot “monster” helps students master math problems and proved particularly useful for struggling students. Another project – a learning game called DreamGrader – offers dense feedback on what parts of the game students struggled with. Brunskill said DreamGrader reduced grading time by 44% and improved accuracy by 6%.
Demszky’s work includes Tutor CoPilot, which helps novice tutors effectively remediate student math mistakes in real time; StaffGen, which assists teachers in individualizing lesson plans for students with different needs; and Teach M-Powered, which encourages teachers to use more growth-mindset language in feedback to students (read more about these projects). Demszky involves teachers in the process of developing all her tools, and to ensure that new models are accurate and effective for marginalized learners, rolls out projects in a variety of school settings including Title 1 schools.
While saving teachers’ time is laudable, Demszky questioned whether we could build more revolutionary tools: “Can AI someday give students agency, motivate them, ensure they feel like they belong in the classroom, improve how they learn?”
Judith Fan, assistant professor of psychology in the Stanford School of Humanities and Sciences, offered a note of caution. Her lab, working to improve STEM education through AI, recently explored learning data visualization skills to find that humans still greatly outperform models: “I’m really excited for the next generation of AI teaching assistants and tutors, and I think they can have an enormous transformative impact on education. But we really need rigorous and thorough testing in place to ensure robustness, reliability, and responsible innovation.”