On April 30, the Social and Ethical Responsibility of Computing (SERC) Initiative at the MIT Schwarzman School of Computing hosted a day-long research symposium examining how artificial intelligence is shaping the world and its impact on society.
The symposium included research talks by SERC’s latest seed grant recipients on topics such as air pollution forecasting and responsible computer vision implementation, a panel discussion on AI collaboration and AI in education, and a keynote address by Dr. John Kleinberg ’96, Tisch Professor of Computer Science and Information Science at Cornell University. A poster session was also held at the event, where student researchers presented the projects they worked on throughout the year as SERC scholars.
“There is so much amazing research being done at MIT about how AI and computing can be a force for the benefit of humanity, and it was inspiring to see the community’s interest in all of this cutting-edge research,” said Brian Hedden, SERC co-associate dean and professor of philosophy. He holds a joint position with the Department of Electrical Engineering and Computer Science (EECS) in the MIT Schwarzman College of Computing.
“As computing and AI become increasingly integrated into nearly every aspect of society, SERC’s mission is to help ensure ethical reflection and technological progress,” said Nikos Trichakis, SERC Co-Assistant Dean and J.C. Penney Professor of Management. “This year’s symposium highlights the incredible range of work underway across MIT and creates a forum for our community to engage deeply with the responsibilities that come with shaping the future of computing.”
Aligning AI with human values – and what those values will be
The challenge for AI coordination and moral mesh lies in the ethical question of how to instill “human values” into highly powerful and rapidly changing technologies. Who decides what values and rationalities are included in an ethical framework? How are distortions taken into account when translating these values from user to machine?
These questions, among others, were posed by EECS Associate Professor Dylan Hadfield Mennell during a panel moderated by an interdisciplinary group of speakers.
Google DeepMind philosopher and researcher Iason Gabriel used the example of a judge to illustrate his point. “Judges are expected to be able to interpret the rules while also having great character. They are rational human beings, but not necessarily the best human beings who have ever lived. When it comes to AI, it is not appropriate to model it as perfect. AI should do what we tell it to do, using its character to interpret according to human moral values.”
Bailey Flannigan, an assistant professor of political science at EECS who holds a joint appointment with the MIT Schwarzman College of Computing, took this a step further. For her, the most important issue in AI collaboration is “resolving the fundamental question of who is qualified to manage different types of AI systems in the first place.”
Joining Flannigan on the panel was Bernado Zacca, an associate professor of political science. Given the momentum of AI and the complexity of institutional design, Zacca said, “One of the most pressing challenges is understanding the wisdom contained in the systems we are replacing and why they work the way they do.”
As deployment pressures mount, it can sometimes feel like people are building planes while they’re flying them, but panelists seemed overall optimistic about the trajectory of AI coordination, emphasizing how important the human component will be in shaping these systems.
off-road and exhilaration
As students at all educational levels begin to use AI, the question arises: Is there a way to ethically incorporate AI tools while maintaining academic accuracy and rigor? In a panel discussion on AI and education, MIT faculty and Gemini for Education Director Marta McAllister examined how AI is already being used in the classroom and discussed how AI can support learning while remaining aligned with instructional and curriculum goals.
Professors Eric Klopfer and Samuel Madden, co-chairs of MIT’s Task Force on the Use of AI in Teaching, Learning, and Research Training, focused on the central dilemma of whether AI is being used to lighten the workload rather than to scaffold the concepts being taught.
Madden, chair of the EECS Department of Computer Science and Distinguished Professor in the MIT College of Computing, described the process of cognitive struggle, in which learning occurs through a series of trials and failures. “When students hit that wall, their instinct is to question the AI,” he said. “They don’t think the AI is good at this process, and they’re not actually learning the skills you’re assessing.” The question then becomes how instructors maintain the cognitive struggle process so that it is sufficiently challenging to fight the urge to use the AI.
Klopfer, director of the Scheler Teacher Education Program and Educational Arcade at MIT, echoed similar sentiments, noting that critical thinking is no longer an important step in research output. As for where to start to keep the material challenging enough, Klopfer suggested looking at the entire curriculum. “Some core content has to be removed. We keep adding rather than parsing and removing,” he said.
Moderator Justin Reich, director of the Educational Systems Lab and associate professor in the Comparative Media Studies Program/Writing, noted that while teens know AI is bad, that doesn’t necessarily mean they’ll stop using it. However, by inviting students into discussions about how to implement AI and incorporating more thoughtful exchanges with instructors, students may become even more capable of choosing how and why they use these tools.
In any case, AI tools and their implementation should not be treated as a one-size-fits-all policy. Pat Pataranutaporn, career development professor of media arts and sciences at Asahi Broadcasting Corporation and head of the cyborg psychology research group at MIT Media Lab, said, “AI is not just one thing. “We can and should design differently. What we measure and how we measure it should not be about getting the right answers. We should think about what it really means for today’s students to learn.”
Is imitating human reasoning as good as the real thing?
Kleinberg’s keynote, titled “AI’s Models of the World and Our Models,” used a slide deck that included references from chess grandmasters and movies to evaluate instances in which AI systems mistakenly cause us to fail due to mismatches between the systems’ models of the world and our models.
To illustrate this point, Kleinberg used chess as an example. Modern chess engines can compete at superhuman levels, but when paired with a human partner, their strategies are incomprehensible and unreasonable to the human opponent. This human handover can lead to confusion. Kleinberg used the example of The Fellowship of the Ring, where Gandalf, a powerful wizard, entrusts a ragtag group of adventurers with a very dangerous and important quest. For those familiar with the story, the group is unexpectedly left without Gandalf’s guidance and temporarily finds themselves in very serious disarray.
When a chess engine passes a turn to a human partner, the human has a hard time understanding the predictive pattern of moves that the engine has been tracking up until that point. “The danger with human-algorithm teams is that when the human takes over, the algorithm knows what it wants to do next, and the human doesn’t,” Kleinberg explained.
These analogies illustrate the difference between how AI understands the world through predictive simulation, pattern recognition, and constraints to mimic human reasoning, and the innate, embodied knowledge that comes with human experience, and whether these systems truly understand the world in which they operate. But the question remains, does it matter even if the game still results in checkmate?
