On April 30, the MIT Schwarzman College of Computing’s Social and Ethical Responsibility of Computing (SERC) initiative hosted an all-day research symposium examining how artificial intelligence is shaping the world and its implications for society.
The symposium included research talks by SERC’s latest seed grant recipients on topics such as air pollution forecasting and responsible computer vision deployment, panels on AI alignment and AI in education, and a keynote address by John Kleinberg PhD ’96, Tisch University Professor of Computer Science and Informatics at Cornell University. The event also featured a poster session, where student researchers showcased the projects they worked on throughout the year as SERC Scholars.
Brian Hayden, SERC co-associate dean and professor of philosophy, who shares a position with the Department of Electrical Engineering and Computer Science (EECS) in the MIT Schwarzman College of Computing, said, “There is so much amazing research being done at MIT on how AI and computing can be a force to benefit humanity. It was inspiring to see the community interest in all this cutting-edge work.”
“As computing and AI become increasingly embedded in nearly every dimension of society, SERC’s mission is to help ensure that ethical reflection and technological progress move forward together,” said Nikos Trichakis, SERC co-associate dean and JC Penney Professor of Management. “This year’s symposium highlights the extraordinary work going on at MIT, and creates a platform for our community to deeply connect with the responsibilities that come with shaping the future of computing.”
Aligning AI with human values – and what values those might be
The challenges with AI alignment and the ethical trap lie in the ethical questions of establishing “human values” on a very powerful and rapidly changing technology. Who decides what values and rationalities are included in the moral framework? How can one take distortion into account when translating these values from user to machine?
These questions, among others, were asked by EECS Associate Professor Dylan Hadfield-Meynell during a panel he moderated, which brought together an interdisciplinary group of speakers.
Iason Gabriel, a philosopher and research scientist at Google DeepMind, used the example of a judge to illustrate his point. “You want a judge to have good character, but still interpret the rules. A reasonable person, although not necessarily the best person ever. When it comes to AI, it’s not fair to present it as an ideal. AI should do what we tell it to do, while using its character to interpret according to our ethical values.”
Bailey Flanigan, assistant professor of political science in a shared appointment with the MIT Schwarzman College of Computing at EECS, took it a step further. For him, the most important problem for AI alignment is “resolving fundamental questions about who is entitled to control different types of AI systems in the first place.”
Flannigan was joined on the panel by Bernardo Zaca, associate professor of political science. Given the pace of AI and complex institutional designs, Jacka expressed, “One of the most urgent problems is understanding the knowledge implicit in the systems we are replacing, and why they function the way they do.”
As deployment pressures increase, it often feels like people are building the plane while flying it, though overall the panelists seemed optimistic about the trajectory of AI alignment, emphasizing how important the human component is to shaping these systems.
descent vs rise
As students at all levels of education begin to use AI, questions arise about whether there is a way to ethically incorporate AI tools while maintaining academic accuracy and rigor. In a panel on AI and education, MIT faculty and Gemini for Education director Marta McAllister explored how AI is already being used in their classrooms and discussed ways it can aid learning while remaining aligned with instructional and curricular goals.
Professors Eric Klopfer and Samuel Madden, co-chairs of MIT’s Ad Hoc Committee on the Use of AI in Teaching, Learning, and Research Training, considered the central dilemma of whether AI is being used to burden work rather than to help formulate the concepts being taught.
Madden, faculty chair of computer science at EECS and distinguished professor in the MIT College of Computing, described the process of cognitive conflict, whereby learning is done through a series of trials and failures. “Students now, when they hit that wall, their first instinct is to ask the AI. They don’t see it as excellence in the process, and they haven’t really acquired the skill that you’re evaluating,” he said. The question is how trainers maintain the process of cognitive conflict so that it provides enough challenge to overcome the desire to use AI.
Klopfer, who serves as director of the Scheler Teacher Education Program and Education Arcade at MIT, echoed similar sentiments, saying that critical thinking is no longer becoming an important step in work output. As for where to start in keeping the material challenging enough, Klopfer suggested examining the curriculum as a whole. “Some key content has to go. We keep adding rather than analyzing or pruning,” he said.
Moderator Justin Reich, director of the Teaching Systems Lab and an associate professor in the Comparative Media Studies Program/Writing, said that although teens know AI is bad, that doesn’t necessarily stop their AI use. However, by inviting them into discussions of how AI is implemented and engaging in more reflective exchanges with instructors, students may be more equipped to choose how and why they use these tools.
Regardless, AI tools and their implementation should not be treated as a one-size-fits-all policy. “AI is not just one thing. It can and should be designed differently to foster things like creativity and critical thinking. What we measure and how we measure it shouldn’t be about getting the right answer. We should be thinking about how these “What would it really mean for a student to learn this day?”
Is imitating human logic as good as the real thing?
With a slide deck that included chess grandmasters and film references, Kleinberg’s keynote speech, titled “AI’s Models of the World, and Ours”, evaluated examples where AI systems have inadvertently failed us due to a mismatch between the system’s models of the world and ours.
To illustrate this point, Kleinberg used chess, where modern chess engines can compete at superhuman levels, but when paired with human partners, their strategies are not understandable or predictable to their human counterparts. These human tricks will then create confusion. Kleinberg gave the example of “The Fellowship of the Ring”, where Gandalf, a powerful wizard, entrusts a ragtag group of adventurers with a highly dangerous and important quest. For those familiar with the story, the group is unexpectedly left without Gandalf’s guidance, sending them into a temporary period of very serious turmoil.
When the chess engine assigns a turn to its human partner, the human struggles to adopt the predictable move pattern that the engine has been following up to this point. “The danger of human-algorithm teams is that when the human takes over, the algorithm knows what it wants to do next, but the human doesn’t,” explained Kleinberg.
These analogies illustrate differences in the ways in which AI understands a world – through predictive simulations, pattern recognition, and constraints – to mimic human reasoning versus the innate, embodied knowledge that comes with human experience, and whether these systems actually understand the worlds in which they are operating. But the question is, does it matter if the outcome of the game is still checkmate?
