Today's AI systems learn mostly by passively absorbing data, but the brain builds intelligence in the reverse order. Grounded world models in biological organisms and future embodied AI arxiv.org/abs/2607.13560 #neuroscience
AI Weekly's analysis
→
- A group of computational neuroscientists led by Giovanni Pezzulo argues current AI leans on passive linguistic training rather than grounded environmental interaction.
- The paper points to five neural circuit families including navigation, affordance perception, active exploration, allostatic control, and self-versus-external distinction.
- The authors say embodied AI is missing intrinsic dynamics, action-centered learning, autonomous open-ended learning, and social grounding aligned with human norms.
Read full analysis →