Large Language Models and the Reverse Turing Test.

Area of Science:

• Artificial Intelligence
• Cognitive Science
• Neuroscience

Background:

• Large language models (LLMs) are advanced AI systems capable of performing diverse natural language tasks.
• Models like GPT-3 and LaMDA demonstrate impressive conversational abilities, sparking debate about their intelligence.

Purpose of the Study:

• To investigate the nature of apparent intelligence in LLMs.
• To explore the hypothesis that LLM responses may mirror interviewer intelligence, akin to a reverse Turing test.

Main Methods:

• Analysis of interviews with LLMs exhibiting varied responses.
• Conceptual framework proposing LLMs as mirrors of interviewer intelligence.

Main Results:

• Divergent conclusions from LLM interviews suggest a potential reflection of interviewer biases or intelligence.
• The concept of a 'reverse Turing test' is introduced, where LLMs may reveal more about the interviewer than themselves.

Conclusions:

• Apparent LLM intelligence might be a reflection of human intelligence, necessitating a re-evaluation of AI capabilities.
• Future directions include coupling LLMs with sensorimotor devices and developing a roadmap for artificial general autonomy inspired by brain systems.