Picture an investor deciding where to put their money, only knowing whether each company earned an A, a B, or a C. If that’s all the information you had, would you feel comfortable investing?
This is James Genone’s argument for wanting to get rid of grades. On the latest episode of Educast 3000, Genone, who has served as Senior Vice Chancellor for Learning Strategy at Northeastern University, says that’s what a university hands an employer every time it sends a transcript: a single letter that flattens everything worth knowing about what a graduate can do.
But the conversation is about something bigger than grading. Genone is a philosopher and cognitive scientist who's spent his career on how people actually learn. His read is that higher education's tools for both measuring and delivering learning are cruder than anyone likes to admit. AI, he argues, is the first technology with a real chance to sharpen both, if institutions design it to.
That "if" is the whole game. The same kind of tool can deepen learning or gut it, depending on how it's built. Designed one way, AI walks a student through a problem and leaves them better able to solve the next one on their own. Designed another, it hands back a finished answer when a student pastes in the assignment, and they learn nothing. The difference is in the design.
A philosopher who quit lecturing
The conviction behind all of this traces back to one semester. In 2013, teaching philosophy and cognitive science at Franklin & Marshall College, Genone stopped lecturing entirely, flipping the classroom in favor of what he called “heavy duty student engagement.” He kept the same assignments, reused some of the same tests from earlier versions of the course, and handed the class time over to his students. The results weren’t just higher scores, but higher engagement in learning. "Not so much just better performance, but actually better growth," he said.
The research backs the instinct. A 2023 meta-analysis in the journal Higher Education pooled 104 studies covering nearly 16,000 college students in the humanities and social sciences, and found active instruction raised assessment scores by about half a standard deviation over traditional lecturing, roughly half a letter grade. Giving students an active role does real cognitive work. (Unsurprisingly, Genome is not our first podcast guest to talk about how AI can be designed to encourage effort that increases learning!)
That semester set the direction. Genone went on to help build Minerva University as a founding faculty member, led product at the AI startup Atypical AI, and lead AI strategy for teaching and learning at Northeastern. The throughline through all of his work is the question “what actually makes a student learn, and how do you give more students access to it?”
The tutor test: AI as coach, not answer key
Ask Genone where the line sits between AI that helps and AI that hollows out learning, and he reaches for one of the oldest models of teaching and learning that we have. "If your tutor is sort of writing your students' essays or solving their homework for them," he said, "they're obviously not playing the role correctly. That's not what we hire tutors for." A good tutor finds where a student is stuck, learns what challenges and motivates them, and offers the right problem at the right moment with steady feedback. An AI assistant needs to be able to do the same.
His favorite example comes from his own house. His 15-year-old daughter, an early adopter, started taking pictures of her quizzes that came back with wrong answers, handed them to a tool like Gemini, and asked it to explain where she went wrong and then generate new problems just like them so she can practice until the idea sticks. This kind of investigate, identify, and iterate cycle is exactly what a good tutor would do during a session.
What Genone's daughter found by instinct, a pair of Harvard researchers built on purpose, for the same reason. When Gregory Kestin and Kelly Miller designed the AI tutor for their physics students, they fed the model the same feedback notes Kestin planned to deliver in class, told it to keep answers brief, and instructed it to give away only one step of a solution at a time. Those constraints forced the same coaching loop: the diagnosis followed by the deliberate practice, rather than letting the tool short-circuit the work. Across 194 students who learned one unit each way, the AI tutor produced more than twice the learning, in less time.
The tutoring model matters so much because we have known its power for 40 years. In a 1984 paper, the educational psychologist Benjamin Bloom reported that students tutored one to one, with feedback along the way, outperformed 98% of students in a conventional classroom. The catch, which Bloom called out, was cost. Genone's argument is that careful AI design, with human judgement, is the first real chance to give that kind of attention to every student, not just the few who can afford a tutor.
Where AI in higher education goes wrong
Genone sees schools failing at AI adoption from opposite directions. Some offer no real guidance and leave each faculty member to invent their own rules; this “wild west” leaves students with, well, wildly different experiences from one class to the next. Others swing to mandates, by either banning AI outright or requiring it everywhere, including at least one university he knows of that forbade faculty from stopping students who wanted to use it. Neither extreme is really serving students.
"You need a thoughtful, deliberate, scaffolded, structured approach," he said, one that pulls faculty in to the process. The work he is proudest of from Northeastern looked like this. He convened groups of faculty, often as communities of practice, and let early adopters teach their peers, whose voices their colleagues already trusted. He set up working groups to build shared frameworks for questions like AI readiness and assessment. Department and college leaders then carried those frameworks into courses, so students got a consistent experience across programs. That doesn't mean AI belongs in every course, and he is clear about that; but it does mean every student should be able to engage with it in a way that will build the AI literacy their field will expect.
What comes next
Looking a few years out, Genone expects the pace of AI in higher education to come from the labor market more than from regulators. Government bodies have too broad a remit to go deep on this, he argues, and institutions differ too much for one rule to fit them all. Employers will set the tempo instead. Graduates will be expected to arrive AI-literate, and universities that don't prepare them will feel it in their students' job outcomes, which then feeds back as pressure on the institutions themselves.
His bet for closing that gap is experiential learning: the practicums, projects, and work simulations that let students learn and do at the same time, and he wants far more of it carrying real academic credit. "Experiential learning is that bridge between the world of employment and the world of learning in higher education," he said.
One idea connects everything Genone argues for: students learn when they do real work and get feedback on it. That's what a good tutor provides and what a skills-based record captures that a transcript can't. Experiential learning is real work plus feedback at full scale — a student doing the actual work of a field before they've left school for it. Giving every student that kind of hands-on feedback was too expensive to offer widely, which is why it's lived at the margins, alongside 1-1 tutoring. Genone's wager is that AI, designed with care and kept under human judgment, is what finally changes the economics, making both the personal attention of a tutor and the real-world practice of a career more affordable to offer every student.
Hear the full conversation with James Genone on Educast 3000.
