AI consciousness odds 'above zero, below 50%': dioscuri on AI welfare trade-offs
btcmaxi116036 · x · 2026-10-09
In an exchange on AI welfare, researcher dioscuri answered three pointed questions:
- Probability LLMs are conscious: declined to give a number publicly, but said it's "higher than zero and less than fifty" percent.
- How well we understand model internals: cites work from Jack Lindsey's lab, but says the topic is "very poorly understood."
- Safety vs. welfare: some welfare risks are "absolutely justifiable for global safety," though we should minimize them.
He adds that public debate moves too fast from "if LLMs are conscious" to "their moral interests are just like humans'" — it's unclear, for example, that shutting down an instance equals death. He recommends two papers: Long, Sebo & Sims on the tension between AI safety and AI welfare, and Salib & Goldstein's "AI Rights for Human Safety."
Related event: Researcher Puts Odds of Frontier LLM Consciousness Between 0 and 50%(3 posts)→