MIT recently released the report from their ad-hoc committee on AI and education, which is a great read and worthy of the “bold ideas” MIT is known for. My colleague Matthew Rascoff, who just ran a fascinating panel session at the INSPIRE 2026 convening, interviewed the committee’s co-chairs, Sam Madden (head of CS there, and Berkeley mafia :-) and Eric Klopfer (director of the Scheller Teacher Education Program).

Matthew was kind enough to give me a pre-release listen of the interview, which will be released as an episode in the Higher Ed miniseries he’s producing for the EdTech Insiders podcast. It goes live November 3, so I can’t reproduce it here, but I took away some interesting points from the interview, some of which are less obvious in the report than others.

The importance of humanism

The democratization of frighteningly powerful technology that can change the world carries great responsibility. Accepting that responsibility requires understanding of history, culture, context. Our students must be humanists. Reading philosophy may not help you directly “on the job,” but it prepares you to read critically and follow logical arguments. Understanding history may not help you “on the job,” but it will give you context for the potential effects of the things you are about to build.

This point also came up in the panel Matthew moderated, and our own Narges Norouzi, who works at the intersection of AI and education and is serving as Associate Dean in Berkeley’s College of Computing, Data Science, and Society, even went so far as to propose that a “4 year degree” be subdivided into (say) a 2-year degree in computing and two 1-year degrees in the humanities or social sciences. Berkeley’s Data 8 and connector courses, and the associated data science modules that Eric Van Dusen and his team have been creating, are a step in this direction. And at the FISEE 2026 I recently attended (and co-organized), this was one of the points that had broad agreement even among those who disagreed vigorously about the discipline-specific approaches for teaching software engineering in the age of GenAI.

For a long time, a typical response to “students should be better steeped in the humanities” was “it’s so hard to learn the technical stuff that there isn’t enough time for that in a 4-year degree.” We no longer have that excuse.

Grades and transcripts as “legacy code”

Everyone seems to hate grades, and to agree that they create perverse incentives and contribute to students’ transactional view of education (“I’m here to get a job, for which I need a degree and good grades from a good college”). So why do we still have them? Matthew compared them to “legacy code”: there’s a whole ecosystem of higher education and career paths that is built around the concept of grades and transcripts, and we need to interoperate with it. Sam Madden mentioned how at Brandeis students now get a second “shadow transcript” in which faculty or instructional staff attest to extracurricular or co-curricular skills that don’t emerge from the transcript alone, such as internship skills, hackathons, etc. Unlike a LinkedIn profile, which a student could self-update, adding things to the shadow transcript requires faculty sign-off. Sam also told the story of a particular Harvard political science professor who would give students two grades at the end of his course: one that goes on the transcript and ensures they’ll have a good shot at law school, and a second unofficial one that represents the professor’s unvarnished and annotated opinion of the student’s work in the course. (At least this instructor is baldly admitting to grade inflation up front. His honesty should be more refreshing than it is.)

But again, this overlaps an INSPIRE topic–transcripts are lossy, not useful as indicators of specific achievement, and create perverse incentives, so we need to start putting mechanisms in place that will lead to their long-term de-emphasis.

Why “Socratic” AI tutors partially miss the point.

MIT, and its peer institutions (mine included), are working on all manner of “Socratic” AI-based tutors that help students struggle with material rather than enabling cognitive offloading. But these only work if students are actually motivated to learn rather than motivated to get a particular grade in the easiest way possible. My colleague Dan Garcia and I were discussing on the way home from the workshop that the social contract with students has broken down: the character of so many current students makes it acceptable to commit acts of academic misconduct that would have been unthinkable when we were undergraduates, and they have no remorse about it when confronted—their main reaction is to avoid getting in trouble. It’s gotten so bad that campus counsel has advised certain faculty not to proceed with cases in which there was obvious and blatant cheating, because forgetting to dot any I’s or cross any T’s in the student misconduct procedure could land the university in an expensive courtroom fight that it was unlikely to win anyway. It is a sad state of affairs when student cheating becomes the same kind of moral dilemma as fighting organized crime.

Perhaps learning more about history and philosophy would improve this aspect of students’ character—but I’m not sure.

AI tools aren’t collaborative

Especially with AI able to do detail work like pounding code, employers care more than ever about so-called “soft skills,” including collaboration skills, participating in or managing a team, and so on. Yet AI tools as currently formulated are designed for individual use. As Matthew pointed out, a lot of people work on collaborative software projects where AI-assisted coding is happening; but the actual collaboration is via existing mechanisms such as GitHub—not facilitated by the AI tools themselves.

What could we do to specifically involve AI in the teaching and practicing of collaboration skills?

I don’t have anywhere to go with this but I wanted to capture my thoughts while still top of mind, and this is too long (and contains too many links) to be a post on Computing Educators, where you should almost certainly go if you want to discuss!

Updated: