Is there a philosopher in the house?

Is there a philosopher in the house? Tech insiders call for humanist-scientist discourse amid ‘Copernican moment’ of AI age Artificial intelligence advances daily. Right now, humans, with our values and beliefs, are the decision-makers who set the agenda, form relationships, create culture. There is a growing sense of profound discomfort over the fact that AI soon may take over some of that. What is our place when we are no longer solely at the center of things? That was the question at the heart of a panel co-hosted by the Berkman Klein Center for Internet & Society and the Division of Arts and Humanities’ Public Culture Project. “I think we’re going to start seeing a lot of exciting applications [for AI] coming out … but it’s important to talk about this in a way that respects what it is to be human,” said panelist Ken Archer, product lead in responsible AI at Microsoft. The panel was the first in a yearlong series exploring what makes us distinctly human and why that is important in the “Copernican moment” of the AI age. Future panels will bring together scientists and humanists to probe what artificial intelligence means for death and grieving, faith and religion, and empathy and meaning-making, among other topics. “These kinds of questions are the kinds of questions that humanists have asked, at least some of them, for a very long time — poets, philosophers, historians, novelists. This is the domain of the human, and we really have to bring our judgment to bear,” said Sean Kelly, dean of Arts and Humanities and the Teresa G. and Ferdinand F. Martignetti Professor of Philosophy, who introduced the panel. Archer and fellow panelist Jackson Kernion ’12, who leads the human feedback program at Anthropic, began by offering contrasting visions of a different kind of intelligence: the human one. “The human mind is not different in kind from a reptile mind or a monkey mind,” Kernion said. “I was always attracted to the idea that the mind is like a computer, which is maybe convenient for the work I went into.” Archer, on the other hand, saw the human mind, with its capacity for reason and for updating its background knowledge or assumptions, as distinct from other kinds of intelligence. “We’re in the world, making distinctions within the world. This is basically thinking, and it’s our best definition we have so far,” Archer said. “What it lacks is the precision of mathematical physics, but I think it can get us started to have a good conversation about whether AI can surpass us.” “I’m present for every moment of my life, and I bring with me all my experiences that accrue. LLMs are born anew in every context window.” Jackson Kernion How would those two perspectives, then, inform what it would mean for AI to outmatch us, asked Moira Weigel, the host of the discussion and an assistant professor of comparative literature in the Faculty of Arts and Sciences. AI would first require a longer memory, Kernion said. “I’m present for every moment of my life, and I bring with me all my experiences that accrue,” he said. “LLMs are born anew in every context window. They have all this background knowledge in their weights, but they pop into existence from the first token of the context window until they’re done generating, and that really is very much unlike us.” Or, he joked, maybe a human life is just one long context window, referring to the amount of text an AI can process at one time when asked to generate a response. Archer pointed to Martin Heidegger’s concept of thrownness, or Geworfenheit — the uniquely human experience of finding oneself thrown into the pre-existing world — and the uniquely human act of expressing that experience in language. It’s humanity’s “thrownness” that gives us our ability to adapt to a frustrating, imperfect world, Archer said, a trait that still confounds the most advanced large language models. AI may excel at chess, he explained, which is a closed world with fixed rules. “But when presented with an impossible task, does it step back from the world and say, ‘Well, what is the significance of this task I have been given?’ … No.” The conversation took place days after news of an Anthropic researcher quitting over the risk of self-improving AI snowballing out of human control and amid the ongoing conversation about how a swarm of OpenAI agents recently went rogue and hacked the AI code repository Hugging Face. Archer and Kernion, who both work in AI safety, are not immune from concerns about the products they help build — including the uncanny ways the models can seem, in Kernion’s words, like developing organisms. “Every RL [reinforcement learning] training run is starting from this infant you’re growing in a weird way. The public only sees the final product; there are all sorts of weird, messed-up models you don’t see. I want to spend the time making sure the reward signals are specified the way we want,” Kernion said. The researchers acknowledged that aligning AI with human values is no easy task, especially since humans are rarely aligned with one another. For those messy contingencies, Archer said, scientists ought to enlist the help of the humanities. “We need the academy, and we need philosophy in particular, at this moment. Unfortunately, it’s less common today than it used to be, for when society faces a crisis, to call the philosophers.”