OREL-group/Developmental-AI · GitHub

Place to host theoretical synthesis for Developmental AI

Continual Embodied Learning

Developmental Braitenberg Vehicles

Training on Viewpoint Networks, in Observer-dependent Collective Behavior

Artificial Neural Networks vs. Biological Neural Networks

Dynammical Heterochrony

Allostasis Machines

Meta-brains Models

Gibsonian Information

• Path Dependence vs. Innateness (mixed features).

• canalization function.

• epigenetic landscapes vs. gradient descent (quantitative trajectories, energy minimization).

• population-based approach (agents with diversity rather than pure minimization).

• are developmental trajectories always maximally efficient? Phenotypic buffering, pleiotropy, minimally viable phenotypes (not developmentally lethal).

• theory development (@ansonzlim), fleshing out and defining different components and their relations.

• computational verification / principle "testing" (inspiration from free energy principle)

• real-world impacts: Gary Marcus "Robust AI" ; Lisa Feldmen Barret, clinical approaches

Next steps?

1. Differentiate Developmental AI from "Regular AI", and what we mean by "development" (sequential acquisition vs. emergent morphogenesis)?

• what is DevAI (guiding document)?

• examination of the various conceptual components and start fitting them together - a Wiki?