On April 30, the MIT Schwarzman College of Computing’s Social and Ethical Responsibilities of Computing (SERC) initiative hosted a full-day research series examining how artificial intelligence is shaping the world and its implications for society.
The series included research discussions by recent SERC seed grant recipients on topics such as air pollution forecasting and the deployment of computer vision in the right way, panels on the alignment of AI and AI in education, and a keynote speech by Jon Kleinberg PhD ’96, Tisch University Professor of Computer Science and Information Science 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.
“There is amazing research being done at MIT about how AI and computing can have the power to do good for the benefit of humanity. It was encouraging to see so much public interest in all this cutting-edge work,” said Brian Hedden, SERC associate dean and professor of philosophy, who holds the MIT Schwarzman College of Computing College of Computing position shared with the Department of Computer Science (EECS).
“As computing and AI grow in focus in almost every part of society, SERC’s mission is to help ensure that ethical reflection and technological progress move forward together,” said Nikos Trichakis, SERC associate and JC Penney Professor of Management. “This year’s series highlights the extraordinary range of work underway across MIT, and creates a platform for our community to deeply engage with the responsibilities that come with shaping the future of computing.”
Aligning AI with human values - and what those values might be
The challenges of AI alignment and moral meshing lie in the moral questions of how to instill “humane values” in such a powerful and rapidly changing technology. Who makes the decision about what principles and ideas are included in the ethical framework? How does one deal with distortions when translating these values from user to machine?
These questions, among others, were asked by Dylan Hadfield-Menell, associate professor of EECS, during a panel he moderated that brought together a diverse 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 morals, but still interpret the rules. A reasonable person, although not the best person who ever lived. When it comes to AI, it should not be modeled as perfect. The AI should be doing what we tell it to do, while using its own morals to interpret according to our moral standards.”
Bailey Flanigan, an assistant professor of political science with a shared appointment with the MIT Schwarzman College of Computing at EECS, took this step further. To him, the most important issue in AI alignment is “resolving fundamental questions of who has the authority to manage different types of AI systems in the first place.”
Joining Flanigan on the panel was Bernado Zacka, associate professor of political science. Given the momentum of AI and the complex designs of institutions, Zacka expressed, “one of the most urgent problems is to understand the intelligence contained in the systems we’re replacing, and why they work the way they do.”
As the pressure to deploy increases, it may sound as if humans are building the aircraft as they fly them, although the panelists overall seemed optimistic about the trajectory of AI alignment, stressing how important the human element is in shaping these systems.
Loading versus lifting
As students at all levels of education begin to use AI, questions arise as to whether there is a way to ethically integrate AI tools while maintaining academic accuracy and rigor. In a panel on AI and education, MIT faculty and Marta McAlister, director of Gemini for Education, explored how AI is already being used in their classrooms and discussed ways it can support learning while remaining consistent with instructional and curricular goals.
Professors Eric Klopfer and Samuel Madden, co-chairs of MIT’s Ad Hoc Committee on AI Use in Teaching, Learning, and Research Training, ran into the main problem that AI is being used to free up work, rather than being used to help integrate concepts being taught.
Madden, head of the computer science division at EECS and MIT College of Computing Distinguished Professor, described the process of intellectual struggle, where learning is done through a series of trials and failures. He said, “students now, when they hit that wall, their first instinct is to question AI. They don’t see this as the best in the process, and they haven’t found the skill that you’re evaluating.” The question then becomes how educators maintain the process of cognitive struggle so that it provides enough challenge to combat the urge to use AI.
Klopfer, who serves as the director of the Scheller Teacher Education Program and the Education Arcade at MIT, echoed similar sentiments, saying that critical thinking is not an important step in career development. As for where to start in keeping the materials challenging enough, Klopfer suggested examining the curriculum as a whole. “Some important content has to go. We continue to add, instead of dividing or pruning,” he said.
President Justin Reich, director of the Teaching Programs Lab and associate professor in the Comparative Media Studies Program/Writing, noted that even though young people know AI is bad, they are not necessarily stopping their use of AI. However, by inviting them into a conversation about how AI is used and incorporating virtual exchanges with instructors, students can be more empowered to choose how they use these tools and why.
Regardless, AI tools and their implementation should not be considered a one-size-fits-all policy. Pat Pataranutaporn, Asahi Broadcasting Corporation Career Development professor of Media Arts and Sciences and head of the Cyborg Psychology research group at the MIT Media Lab, said, “AI is not just one thing. It can and should be designed differently to encourage things like creativity and critical thinking. What we measure, and how, shouldn’t be about getting the answer right. We have to think about it so the student can really learn these days.”
Is imitating human thinking as good as the real thing?
With a slide deck that included chess grandmasters and movie references, Kleinberg’s keynote, titled “AI’s World Models, and Ours,” explored situations where AI systems have unintentionally failed us due to mismatches between the world’s system model and ours.
To demonstrate this point, Kleinberg used chess, where modern chess engines can compete at superhuman levels, but when paired with human partners, their strategies cannot be understood or compared to their human counterparts. This evacuation would have led to confusion. Kleinberg used the example of “The Fellowship of the Ring,” where Gandalf, a powerful wizard, assigns a very dangerous and important quest to a ragtag group of hunters. For those familiar with the story, the group is unexpectedly left without guidance by Gandalf, sending them into a temporary state of extreme chaos.
When the chess engine passes the turn to its human counterpart, the human struggles to follow the predictable pattern of moves that the engine has been following until now. “The danger with human-algorithm teams is that when a human takes over, the algorithm knows what it wants to do next, but the human doesn’t,” explains Kleinberg.
These simulations show the difference in how AI understands the world – through predictive simulations, pattern recognition, and limitations – to mimic human thinking versus the internal, integrated knowledge that comes with human knowledge, and whether these systems really understand the world in which they operate. But the question remains that if the game still leads to checkmate, does it matter?
