Pan-Embodimentism

From the AI Research Compass wiki

Pan-Embodimentism: Ring title for extreme World scores: there is no intelligence without a body, and everything that computes must be understood as embodied.

Beliefs

Pan-Embodimentism is a ring title earned when the Language-World score exceeds 0.85, the far right edge of the chart, whatever the vertical position. It is the mirror image of Logocentrism. Its holders believe that cognition is grounded in sensorimotor interaction, that meaning comes from what a body can do with a thing, and that a model which has only read text will never truly know what “heavy” or “slippery” means. The “pan” is the provocation: even a language model, on this view, is embodied, just in a very poor body made of a keyboard and a context window.

Since only the horizontal extreme matters, the title is shared by giant robot foundation-model builders and by tinkerers of passive walkers alike. Its ultra-revolutionary shade reflects the claim that the language-centred mainstream has the foundations backwards, a view rooted in embodied cognitive science and enactivism.

Arguments

  • Embodied cognitive science argues that perception, action and environment jointly constitute cognition, a view supported by work on sensorimotor contingencies and morphological computation.
  • Language models still make errors about object permanence, physical support and spatial relations that embodied learners acquire early.
  • The volume of continuous sensorimotor experience a child accumulates far exceeds the text any model has read, which suggests what is missing from text-only training.

Critiques

  • Language models have acquired surprisingly rich world knowledge from text alone, which weakens the strongest form of the “no body, no mind” claim.
  • “Everything is embodied” stretches the concept until it no longer distinguishes any approach from another.
  • Embodied data is slow and expensive to collect, so the position can delay progress waiting for robots to catch up.

Representative works

Relationships

How to get it

Earned in the full or pro test when: Representation >0.85.

A first draft. Corrections and better arguments are welcome through a pull request. Edit this entry on GitHub Edit its data