Dartmouth, 70 years on: AI now answers quickly, and human judgment to assess it becomes the scarce resource

By Dartmouth · November 4, 2026.

From October 29 to 31, Dartmouth hosted the conference "AI at 70: What Must Remain Human?", seventy years after the summer workshop that gave artificial intelligence its name and launched it as a field. According to the university itself, more than 1,000 people registered and 24,000 followed the live streams.

The central idea was put forward by President Sian Leah Beilock: anyone can get a convincing answer from an AI in seconds, so the hard skill now is judgment. That means knowing whether the answer is correct and complete, understanding why, and being able to defend it. Beilock called for AI literacy to go beyond using tools efficiently and to include knowing when to use them and when to set them aside.

Industry speakers offered a qualified optimism. Peter DeSantis, senior vice president at Amazon, claimed that five engineers with AI do what used to take 50, and ten times faster; that figure is his own, with no supporting data in the article. Michael Breen of Anthropic argued that AI can extend the capacity to act autonomously to far more people, but acknowledged three risks: dependence that weakens human abilities, erosion of professional expertise, and concentration of power among those who control the systems. Mira Murati of Thinking Machines Lab advised students against spending their degrees trying to guess which jobs AI will take.

There were also critical voices. Columnist Ezra Klein described himself as "pretty skeptical," for now, about integrating AI heavily into education, and bet that value will return to the attention span needed to read a book. Molly Crockett, a Princeton professor, criticized the "hype" driven by industry and the US government, which she said diverts investment from academic science and rests on a mistaken view of how science works. Student Emma Lefferts noted that AI relies on real human labor and natural resources, and that there are present-day risks for specific communities.

The parallel hackathon, organized by Dartmouth's AI club with support from Anthropic, brought together 187 students in 40 teams. The winner was Wheelio, an app that finds accessible routes on campus for wheelchair users and uses AI to detect barriers. Dartmouth is also now surveying employers to measure whether entry-level jobs are changing because of AI; the next student survey will take place in August 2027.

Our take. The underlying message is right: when producing text, code or analysis gets cheaper, what becomes scarce is verifying, deciding and taking responsibility. That is consistent with what we see in the job market: routine work gets automated, while judgment, relationships and accountability gain value. Among junior profiles, however, there is a paradox the conference barely resolves: judgment is built by doing the basic work that AI now absorbs. If that rung is removed, how does anyone learn to judge? That is why the emphasis on reading, writing and reasoning without assistance makes pedagogical sense, not just nostalgic sense.

It should also be read with caution. This is a university reporting on its own event, with an AI lab backing the hackathon and speakers with clear interests. The productivity figures are claims, not measurements, and the data point that really matters, what happens to entry-level jobs, is still being collected.

In the short term, the transition will be uncomfortable: some students will outsource their thinking and some employers will cut the base of the pyramid. In the long term, if education manages to produce people who use AI without depending on it, the promise still holds: more capability within everyone's reach, along with the health, science and free time that makes possible. That Dartmouth is asking what must remain human is a good sign; the test will be whether classrooms turn it into practice and not just rhetoric.