Bad AI Consciousness Arguments, Turned into Bingo

Summary

In an opinion essay, Henry Shevlin turns 24 recurring objections to AI consciousness into a bingo card and argues that many of them end the discussion by assertion rather than by argument. His moderate position is that AI consciousness is possible, current models are probably not conscious, and people should remain cautious because the relevant theories and evidence are unsettled. The objections he addresses include claims that language models are merely programmed, matrix operations, next-word predictors, copies of human behavior, or role-players. He also examines claims that consciousness requires life, a body, feelings, free will, continuous activity, or biological implementation, arguing that each requirement needs support from a substantive theory rather than a slogan. Other sections criticize weak analogies with phones or radios, capability-based dismissals such as a model’s difficulty counting letters, and evidential standards that would also make it difficult to establish that animals or other people are conscious. Shevlin says indirect evidence can still matter, notes that a 2023 indicators report found no current system satisfying its indicators while identifying no obvious technical barrier to future systems, and distinguishes model training from a complete explanation of model cognition. He also rejects dismissing the debate as anthropomorphism, hype, distraction, or “AI psychosis,” because those claims address motives or social behavior rather than consciousness itself. The essay acknowledges that it focuses on anti-consciousness arguments and that pro-consciousness arguments can also be poor, including inferring consciousness directly from intelligence. It closes by treating the Chinese Room as a genuine but frequently misused argument, not a debate-ending slogan.