Human likeness is a misplaced benchmark for artificial intelligence

Recent claims that large language models pass the Turing test span different procedures, but share a common focus on human outputs, rather than the processes that lead to those outputs. Here we challenge the inference from successful imitation to intelligence. Similar outputs can arise from different processes and conceal differences in reasoning, generalization, and adaptation. More fundamentally, we question whether human likeness should guide AI development at all. Prioritizing imitation can reproduce human biases, facilitate manipulation, and favor replacing people over augmenting their capabilities. AI systems should expand human welfare, not merely impersonate us. We recommend three coordinated changes: designing systems that strengthen human capabilities and independent judgment; requiring funded projects to assess benefits, costs, and their distribution; and guiding institutional procurement and renewal by evidence of AI’s effects on human welfare.