AI+Education Summit: Generative AI for Education

Rob Rish, a faculty member teaching political philosophy and holding an appointment at the institute for human-centered AI, introduces the panel on AI and education. He reflects on his background as a sixth-grade teacher and his PhD from The Graduate School of Education, expressing keen interest in the integration of AI in education. Rish highlights the fragmented nature of American education and its challenges to adopting new innovations, mentioning Larry Cuban's view on the persistence of the blackboard as the last adopted innovation. The panel aims to explore the potential and perils of AI in revolutionizing education, with a focus on navigating the complex policy and governance system, including higher education.

Percy Tang, the director of the center for research on Foundation models, discusses the concept of foundation models, using Chat GPT as an example. These models are trained on internet-scale data and can perform a variety of tasks, such as answering questions and generating content. Tang explores the potential of these models in education, including their ability to assist students with question-answering and providing teachers with tools for simulation, problem generation, and assessment. He acknowledges the models' unreliability and the need for improvement in their pedagogical application, emphasizing the importance of managing expectations and understanding the models' limitations in the educational context.

The speaker discusses the transformative potential of foundation models on education, suggesting that these AI tools can enhance teaching and learning by generating problems, simulating students for training purposes, and providing feedback. There's an emphasis on the importance of teaching fundamentals and moving beyond to higher levels of skills, such as ideation and oversight. The rapid progression of AI models from being able to generate fluent text to multi-purpose applications is highlighted, with a call for considering the trajectory of AI development in educational discussions. The talk concludes with an optimistic view of AI's role in human learning and the need for continuous improvement in the technology's reliability and pedagogical application.

The speaker, who is humbled to be on the panel, discusses the potential of generative AI to empower teachers, an aspect often overlooked in discussions about AI in education. The focus is on the importance of teacher-student discourse and how AI can support teachers in fostering a culture of curiosity and growth mindsets. The speaker's research involves developing applications that provide feedback to teachers using foundation models, with evidence suggesting that such interventions can improve instruction and student outcomes. There's also an exploration of how generative AI could offer suggestions to teachers and an emphasis on the ethical use of AI, ensuring it does not propagate existing inequities in education.

The speaker shares examples of how generative AI, such as GPT, can be used to enhance teaching by providing innovative suggestions for classroom activities, offering more constructive feedback to students, and supporting students' emotional needs. The potential of AI in providing real-time and post-teaching feedback is discussed, along with the importance of involving expert teachers in the development process. The speaker emphasizes the need for data representative of the target population and the ethical considerations of privacy and oversight when using AI in education.

Noah Goodman, starting with a personal anecdote about homeschooling during the pandemic, delves into the effectiveness of one-on-one tutoring and the potential of AI to enhance this method. He discusses the 'Bloom's two Sigma Paradox' and the importance of good tutors who follow specific principles, such as the Inspire model. Goodman then shares examples of AI tutoring conversations, highlighting the challenges of ensuring AI follows instructions and understands the motivational aspects of tutoring. He concludes by emphasizing the need for a collaborative approach between educators and AI developers to create effective AI tutors.

The speaker discusses the challenges and potential solutions in teaching AI systems, like Alfred, mathematical problem-solving skills. The Piano system is introduced as a method to formalize math curriculum and train AI agents using logical consistency and expert iteration. The Piano system's success in learning algebra sections from the Khan Academy curriculum is highlighted, along with the interpretability of the learned abstractions. The speaker raises the question of combining the strengths of open-domain language models with logically constrained solvers for effective AI tutoring.

The conversation explores the potential of generative AI in classrooms, with a focus on the transition from traditional teaching methods to AI-assisted learning. The panelists discuss the importance of maintaining human involvement in the educational process, even as AI augments teaching and learning. There's a debate on whether AI could replace human teachers, with opinions leaning towards AI serving as a tool for augmentation rather than displacement. The discussion also touches on the need for policies regarding the use of generative AI in educational settings and the importance of critical thinking in learning.

The final panel discussion centers on the role of writing in education and the potential impact of generative AI on this skill. Panelists consider whether the ability to write will continue to be valuable or if familiarity with AI-generated writing will become more important. There's an emphasis on disentangling writing from analytical thinking and exploring new ways to teach analytical skills, with AI potentially assisting in this process. The conversation concludes with the idea that AI should be used to enhance the educational experience, focusing on the 'what' of learning rather than just the 'how.'