Measuring human likeness of artificial intelligence to natural intelligence

Natural intelligence (NI) is considered the biological foundation and ultimate aim of artificial intelligence (AI). Leveraging the steep increase in computational power and high-performance algorithms, many AI-empowered applications have claimed to attain human-level natural intelligence and behave like humans in certain tasks. However, these claims are primarily based on the classical Turing test and heavily rely on the binary self-reported ratings. Here, we introduced comprehensive behavioural and neural measurements to quantify the continuous distinctions between AI and NI. An adaptive-learning nonverbal Turing test was used to imitate the certainty and variability of human behaviours simultaneously, and typical LLM-based implementations were conducted as validation. Our results clarify the human behavioural and neural sensitivity of evaluating AI human likeness, extending Turing's testing criteria, and illustrate an empirical basis for pointing out directions of various human-like AI applications to approach NI.