Symbolic models offer a powerful account of cognition, yet their applicability has long been constrained by manual engineering and computational tractability. We demonstrate how recent advances in artificial intelligence can overcome these limits and enable large‑scale symbolic modeling. We introduce Behavior‑Conditioned Symbolic Synthesis (BCSS), an automated search for interpretable, deterministic symbolic programs that provably reproduce observed behavior. BCSS is applied to hundreds of challenging reasoning problems, modeling both human and machine performance. The extracted programs reveal that people and machines alike tend to oversimplify, adopt rules that only partially explain the data, and misinfer dependencies among problem features. These shared signatures are modulated by available resources; higher reasoning effort predicts greater simplification and explanatory behavior in machines. Together, the findings suggest a satisficing‑like mechanism underlying both human and machine reasoning and illustrate that BCSS provides a general approach for interpreting and comparing rule‑based reasoning of humans and black‑box AI systems.
Review