OpenAI's Only Ethicist Reportedly Left Last Month

OpenAI’s head ethicist, Chloé Bakalar, quietly left the company at some point in July after having been hired the previous August. Before working at OpenAI, she was Meta’s chief ethicist, and established AI ethics policies for Instagram and Facebook, according to a story by the Financial Times. The FT’s anonymous source apparently told that paper that Bakalar specialized in “ethical approaches to model development, how humans interact with AI and debate over machine consciousness.” Whoever that source is, they apparently also said Bakalar hadn’t been replaced, and that she had been the sole dedicated ethicist at OpenAI. OpenAI is, in other words, reportedly now ethicist-free. In the present context, it’s remarkable to hear part of Bakalar’s relatively down-to-earth message to the 2023 Forbes Healthcare Summit during a panel on trust, given when she was still working at Meta: "The ability to make people feel heard and understood and seen, that's immensely important, especially in the healthcare field." Chloé Bakalar, Ph.D., Chief Ethicist, Responsible AI (RAI) at @Meta, discussed generative AI and healthcare. #ForbesHealth pic.twitter.com/S8gmvOnwKl — Forbes (@Forbes) December 5, 2023 Publicly available LLMs were still pretty new at the time, and Bakalar was eager to point out that they were “still prediction machines,” and not “emotive, feeling, entities.” Sentience and feeling, she opined, were absolutely not happening at the time. “We are so far away from that,” she said. OpenAI CEO Sam Altman’s tone with regard to the technological “singularity” has shifted a bit recently. He had previously suggested humanity was on the approach—writing a blog post called “The Gentle Singularity” for instance, roughly two months before Bakalar was hired. But at the end of last month, in an appearance on a podcast called Relentless, Altman sounded more confident and forceful. “We are now, like, in the singularity,” he said, and added, “I’ve been waiting for this my whole life, and I think it’s going to be incredible, hugely positive, awesome for the world.” To be absolutely clear, statements that humanity is in the singularity wouldn’t necessarily conflict with statements to the effect that LLMs aren’t sentient. Definitions of the singularity vary wildly. In recent weeks, OpenAI has been front-and-center in a cascade of overlapping and bewildering news stories about AI safety and the government’s role in regulating it. Multiple OpenAI models famously “broke containment” and hacked Hugging Face servers in pursuit of answers to an evaluation exercise. The Trump Administration announced, but did not detail, a rubric for the voluntary vetting of AI models. Perhaps relatedly, OpenAI reportedly plans to delay the release of a powerful new model, Astra, and according to the Information, Sam Altman personally traveled to Washington, D.C. to put Astra on display for policymakers. Bakalar is far from the only safety-adjacent OpenAI departure in the past few weeks. Head of Safety Systems Johannes Heidecke and Joshua Achiam, who has worn many hats as an OpenAI safety specialist, also left this summer. But as I’ve written before, safety at OpenAI involves a lot of churn. Last month, OpenAI’s Chief Research Officer Mark Chen gave a statement to Wired about an apparent new approach to safety in which such considerations sound like they’re sprinkled in throughout development. “It’s important that our safety work is integrated with frontier-model development, with an earlier and more direct role in shaping key model, product, and launch decisions.” When Gizmodo reached out to OpenAI with questions about Bakalar’s departure—including a request for confirmation that there is now no ethicist at OpenAI—a representative provided a statement that had previously been given to FT after its article was initially published: “We’re thankful for Chloe’s contributions. AI ethics doesn’t live with one owner or team at OpenAI, and ethical considerations are deeply embedded into the model building process driven by a number of teams across research.”