Adding a memory (forgetting) mechanism slightly decreases simulated learning gains overall yet improves alignment with human results, apparently by letting agents discard incorrect rules¶

Claim · All claims
Evidence · 1 study · theoretical r3 · q2 · i2 medium
Currency · newest study 2023 · not yet checked against its sources · not reviewed by a person · recheck by 2026: tool and model accuracy changes quickly

Subclaims¶

q2 i2 Introducing the memory mechanism decreased overall learning gains (13.7% vs 22.2%) but did not negate the worked-example benefit for skills, and forgetting agents more effectively utilized correct production rules. → Rachatasumrit 2023

Evidence¶

Rachatasumrit 2023¶

Rachatasumrit, N., Koedinger, K. R., & Carvalho, P. F. (2023). Evidence and theory for why the best example-problem ratio to optimize learning gain depends on knowledge content. https://osf.io/29wpe doi:10.31234/osf.io/eyxfw

q2 · i2 · theoretical · r3

Simulation comparison of AL agents with versus without forgetting (d = 0.7). The authors' interpretation, from a closer examination, is that forgetting agents better discard incorrect production rules and more often select correct ones due to stronger memory activation of later-learned correct rules.

"the introduction of a memory mechanism in simulated learners with forgetting slightly decreases learning gains (22.2% for simulated learners without forgetting vs 13.7% for simulated learners with forgetting), t (948) = 9.409, p < 0.0001, d = 0.7, but does not negate the benefits of worked examples over retrieval practice for skills"

Discussion¶

Related Claims¶

• In simulated learners with forgetting, study-practice yields higher skill gains while practice-only yields higher fact gains, mirroring the human pattern — related
• Ablations show each CogEvolution module contributes: removing ICAP perception, structured retrieval, or evolutionary update degrades mistake precision, learning-curve fit, and alignment — related
• In a human experiment, verbatim facts are learned better from practice-only training while general skills are learned better from example-integrated training (cross-over content-treatment interaction) — related