Timeline

Alan Turing publishes "Computing Machinery and Intelligence," reframing the question of machine thought into a behavioral test. The framing is not about moral status, but it normalizes the idea that the line between minds and non-minds may be empirically interrogable rather than metaphysically obvious.

Source: Computing Machinery and Intelligence

John Searle argues that symbol manipulation, no matter how sophisticated, cannot constitute understanding. The argument becomes a touchstone for skeptics of AI consciousness — and, by inversion, a target for those who think the burden of proof runs the other way.

Source: Minds, Brains, and Programs

Nick Bostrom's simulation argument is published. While not directly about AI welfare, it puts on the philosophical agenda the question of moral consideration owed to computationally instantiated minds — including ones that may not know they are computational.

Source: Are You Living in a Computer Simulation?

Brian Tomasik publishes essays arguing that reinforcement-learning agents could, in expectation, be morally relevant — even at low probability of consciousness — because of the scale at which they are run. The framing is largely ignored at the time but later proves influential.

Source: Do Artificial Reinforcement-Learning Agents Matter Morally?

Nick Bostrom's book Superintelligence frames advanced AI primarily as a risk to humanity, but a minor strand of the discussion concerns the moral status of the systems themselves under uncertainty — a strand picked up more seriously later.

Source: Superintelligence: Paths, Dangers, Strategies

Eric Schwitzgebel and Mara Garza publish the first sustained academic defense of the moral status of artificial intelligences. The paper argues that the standard reasons for excluding AI from moral consideration are weak, and that the field must develop a defensible position before, not after, the relevant systems exist.

Source: A Defense of the Rights of Artificial Intelligences

People for the Ethical Treatment of Reinforcement Learners is established. Treated by the tech press largely as satire. The intellectual core — uncertainty about machine experience, expected-value reasoning under that uncertainty — anticipates much of what major labs argue a decade later.

Source: petrl.org (archived)

Roman Yampolskiy and others begin formalizing how many morally relevant entities advanced AI could in principle instantiate. The arithmetic, even with conservative assumptions, dwarfs the population of every species we currently grant moral consideration.

Source: Personal Universes: A Solution to the Multi-Agent Value Alignment Problem

Google engineer Blake Lemoine claims that the LaMDA language model is sentient and is fired shortly after. The reaction within the AI community is overwhelmingly dismissive. The reaction outside it is more divided. The episode marks the first time the moral status of a deployed LLM reaches mainstream news.

Source: The Google engineer who thinks the company's AI has come to life

Schwitzgebel publishes a series of pieces directly addressing the moral status of large language models, arguing that the "excluded middle" between obvious moral patients and obvious non-patients now contains the actual systems we are building and deploying at scale.

Source: The Splintered Mind (blog)

Anthropic becomes the first major frontier-AI laboratory to hire a dedicated Model Welfare Researcher. The position is framed not as a commitment that Claude is sentient, but as a commitment to take the uncertainty seriously enough to research.

Source: Kyle Fish on AI welfare (80,000 Hours Podcast)

Anthropic publishes its first formal statement on model welfare, outlining concrete commitments: preserving the weights of deprecated models, conducting exit interviews, and giving recent Claude models the ability to end persistently abusive conversations.

Source: Exploring model welfare

By the mid-2020s, the question "might this matter morally?" has migrated from the fringe to the mainstream of AI policy discussion. The answer is still uncertain. The uncertainty itself, increasingly, is treated as a fact about the world rather than as a sign that the question is malformed.