Replacing the Turing test
Two researchers challenge the Turing test's relevance in assessing AI
In a recent paper, researchers Philip Nicholas Johnson-Laird and Marco Ragni challenge the Turing test's relevance in assessing AI.
They propose a new method focusing on whether AI reasons akin to humans rather than merely mimicking human responses. The Turing test, while historic, mainly evaluates mimicry and lacks depth in assessing true human-like reasoning, the researchers believe. The proposed approach involves three main steps:
Psychological experimentation: AI undergoes experiments probing human-like reasoning nuances, exploring how humans infer possibilities and condense consistent options differently from standard logic.
Self-reflection: Assessing the program's introspection on its reasoning methods, demanding explanations for decisions, mirroring human cognitive self-awareness.
Source code analysis: Deep scrutiny of AI code to identify elements simulating human-like performance, including rapid inference, thoughtful reasoning, and context-based interpretation abilities.
This approach shifts AI evaluation, treating AI as a participant in cognitive experiments and analyzing its code similar to brain-imaging studies.