Davinci the Dualist: The Mind-Body Divide in Large Language Models and in Human Learners

Area of Science:

• Cognitive Science
• Artificial Intelligence
• Philosophy of Mind

Background:

• Humans intuitively perceive a mind-body dualism, viewing the mind as separate from the physical body.
• While learning contributes to dualism, innate core knowledge systems may also play a role in its emergence.
• The debate persists on whether dualism acquisition relies on innate cognition or is solely learnable through experience.

Purpose of the Study:

• To investigate the origins of mind-body dualism by examining a large language model (LLM) devoid of innate core knowledge.
• To determine if dualism can emerge from experience alone, independent of inherent cognitive systems.
• To assess the dualist tendencies in LLMs and their relationship with the model's learning potential.

Main Methods:

• Probed a large language model (Davinci, part of GPT-3) for evidence of a mind-body divide.
• Compared dualist tendencies between GPT-3 and its successor, GPT-3.5 (text-davinci-003).
• Assessed the models' selective consideration of mental states (epistemic states) as disembodied.

Main Results:

• Davinci, an LLM without core knowledge, exhibited dualist tendencies, indicating a predisposition towards a mind-body divide.
• Dualist bias increased with the learner's inductive potential, with GPT-3.5 showing a stronger bias than GPT-3.
• LLMs selectively treated thoughts as disembodied and rejected the persistence of the psyche after death, yet showed dualist tendencies regarding life.

Conclusions:

• The mind-body divide is partly learnable from experience, independent of innate core knowledge.
• LLM findings challenge nativist theories that attribute dualism solely to innate cognitive systems.
• While not definitive for human acquisition, LLM results necessitate a higher burden of proof for innate theories of dualism.