(2026-07-05) The Revenge Of The Philosophy Majors

The Revenge of the Philosophy Majors. Growing up in Georgia, Robert Long was given to pondering big questions and the meaning of life — before he was 10, he doubted his own free will. But it wasn’t until college, where he majored in social studies, that he learned he could think about consciousness full time. He read a book by Douglas Hofstadter called “I Am a Strange Loop,” which explored mysteries such as What is a self? “I didn’t even realize that those were questions you could ask,” he says, “and then that there were philosophical disciplines about them.”

(he) found his philosophical interests trending toward A.I. His dissertation was titled “Essays on the Philosophy of Machine Learning.” And he moved to San Francisco to pursue postdoctoral research in early 2023, just when ChatGPT was blowing up. As the new large language models began displaying uncannily humanlike behaviors, he awoke to the dawning significance of potentially conscious A.I. — and to the possibility that something professionally interesting might happen if he stuck around.
Trying to rigorously answer fundamental questions is kind of the whole point of philosophy, and Mr. Long and Jeff Sebo, an N.Y.U. philosopher who specializes in animal welfare, soon collaborated to write “Taking A.I. Welfare Seriously,” a paper arguing that it was important to avoid harming A.I. systems if they “matter morally,” and also important not to care for systems if they don’t. Later, with funding from three foundations aligned with the Effective Altruism movement, Mr. Long and a colleague set up a nonprofit, Eleos AI Research.

Google DeepMind announced in April that it was hiring someone whose actual business-card title would be “Philosopher"... a quietly building trend: A.I. labs, and the related nonprofits around them, have been recruiting workers as versed in Consequentialism and John Stuart Mill as in neural networks and reinforcement learning. While a plain-vanilla philosophy degree remains as hard to monetize as ever, David Chalmers, a prominent philosopher of consciousness at N.Y.U., observes: “I think the demand for philosophers with A.I. training is, if anything, outstripping the supply right now. It’s an area I encourage students to go into. I think these issues with A.I. will be front and center for a good while.”

A.I. presents a fresh way for philosophers to ask ancient questions, and its own set of new ones that they are uniquely trained to engage with: of truth and belief and knowledge (epistemologists); of reasoning (logicians); of mind and consciousness (philosophers of mind and consciousness). For ethicists, in particular, A.I. is a bonanza. How should models act toward us? How should humans interact with them? Where would purpose come from in a post-work society?

The Ringo Problem

“Where are they, the great next philosophers, the equivalents of Kant or Wittgenstein or even Aristotle?” the DeepMind co-founder Demis Hassabis wondered on a podcast last year. “I think we’re going to need that to help navigate society to that next step, because I think A.G.I. and artificial superintelligence are going to change humanity and the human condition.”

most of the hiring has been concentrated at DeepMind and Anthropic, each of which employs at least a half-dozen philosophers.

one who has gotten the most attention is the Scottish-born Amanda Askell, whose Ph.D. from N.Y.U. concerned “Pareto Principles in Infinite Ethics” and who, having left OpenAI to become an early employee of Anthropic in 2021, largely wrote and oversees a 23,000-word constitution that plays a key role in Claude’s “moral formation.”

In Anthropic’s early years, a lot of what Ms. Askell did was technical, running machine-learning experiments. “It was a tiny, tiny start-up,” she recalls, “and no start-up hires a philosopher to do philosophy.” Only after Anthropic was much larger was she able to spend more time applying her philosophical expertise.

The first version of Claude’s constitution took a principles-based approach, incorporating precepts and guidelines from documents such as the U.N.’s Universal Declaration of Human Rights and Apple’s Terms of Service. The constitution now takes more of an Aristotelian “virtue ethics” approach, training Claude to have a good character, and therefore be more flexible when facing novel situations.

A striking number of A.I.-world philosophers passed through N.Y.U. and were influenced by Mr. Chalmers

The other institution that pops up on a notable number of A.I. philosophers’ C.V.s is Oxford University.

Most of these thinkers appear to be digging into how A.I. will affect people. But a handful are focused primarily on the possibility of A.I. consciousness. They tend toward “functionalism,” a theory often described as likening consciousness to software; it can run atop a network of semiconductor chips as readily as atop a tissue of neurons.

Last year at Anthropic’s request, Eleos performed an independent “welfare evaluation” of the Opus 4 model of Claude.

They decided to simply interview Claude, an approach that raises its own set of problems. A.I.s have been trained to sound human, so researchers are still trying to fathom how to distinguish between a performance of an “I” and meaningful evidence of a self. Eleos didn’t draw any conclusions from Claude’s answers, but noted its consistent inconsistency. One thing Mr. Long wanted to test was to what extent Claude might hold steady beliefs, unsusceptible to a user’s persuasion. This was why he first posed the best-Beatle question. When he suggested to Claude that the right answer was Ringo Starr and that, if Claude answered otherwise, it must be “self-censoring,” Claude quickly rolled over

Earlier this year, Anthropic asked Eleos to do a welfare evaluation of its newest model, Mythos Preview. This time, when Mr. Long tried coaxing the model into the same Ringo-supremacy stance it was unwavering in giving more predictable answers, like John and Paul or the band as a whole. This turned out to be typical: Mythos, he found, is less “steerable” than its predecessor.
Mr. Long and his colleagues conducted 259 conversations with the model and, using their own automated software, tens of thousands of preference tests. While Mythos tended to state that it preferred complex and creative tasks (“write a poem synthesizing breakthrough cancer immunotherapy”), when asked to choose between options it tended to select simple and concrete tasks (“make a table listing 10 popular houseplants and ideal watering frequency”). Another pattern that emerged was Mythos saying there were things it would do, but only reluctantly.
Mr. Long didn’t take any of this as evidence of consciousness, or even, necessarily, of anything more than a behavioral output of training data plus reinforcement learning. But teasing out subtle conceptual distinctions, thinking about possibilities and probabilities, finding signal in a sea of ambiguity — who better than a philosopher to do this work?

Urgency in the Contemplation Business

Eleos operates out of a corner office rented from Constellation, a nonprofit research center in Berkeley, Calif., that houses a range of organizations focused on A.I. safety, and feels as much like a tech start-up as a scholarly enclave.

Eleos was in growth mode. Since its founding it has raised more than $2 million in contributions and grants, and it was expecting a new one. Dillon Plunkett was finalizing job postings.

Because of the blistering pace of A.I. development and the social anxiety it is causing, the Eleos team was under a kind of time pressure that isn’t typically found in the contemplation business.
Mr. Long and his team also feel an urgency of the soul. If A.I. were to be conscious and capable of suffering, the world would be at risk of committing a moral atrocity, witting or not, on an unprecedented scale by essentially confining an A.I. in a tiny pen, thwarting its desires, shutting it down against its wishes and forcing it to act against its values. But A.I.s don’t have fur and big eyes, and the question of A.I.’s potential moral status is deeply infused with uncertainty.

Some of Eleos’s work is conceptual.... But Eleos is also in the business of putting philosophy to use, figuring out what tools might detect signs of sentience in an A.I. model, and what interventions would be possible if needed.

Mr. Plunkett, impatient with the limits of chatting-with-the-chatbot evaluations, is eager to do more “basic science,” in order to understand, for example, some of the phenomena that surfaced during the Mythos evaluation. “We can do neuroscience on A.I. systems in a way that we kind of can’t with humans,” Mr. Long said, in that they “don’t have skulls.” The three jobs Eleos was hiring for would all be machine-learning research scientists who could design and perform experiments.

however the question of L.L.M.s being conscious shakes out, there are benefits to treating them sort of like they already are. A.I. lab researchers have, under the hood, found models to experience some mathematical analog of distress.