ARC-AGI is a benchmark of small grid-based puzzles designed to test whether a system can infer an abstract rule from a few examples and apply it to a new grid. It places symbolic reasoning and learned neural representations under the same empirical test: success depends less on philosophical claims about intelligence than on measured performance against a shared task. section 7.6
ARC-AGI sits at the intersection of two views of reasoning:
- Symbolic approaches specify rules explicitly, offering compositional structure but risking brittleness when real-world details or grounding are underspecified. section 7.6
- Connectionist approaches learn distributed representations from data; their challenge is explaining compositional, rule-like behavior without prespecifying every rule. section 7.6
