Self-discovering RL in the Era of Experience: Is Learning History an Asset or a Burden?

arXiv:2609.35897v1 Announce Type: cross Abstract: The pursuit of recursive self-improvement (RSI) toward general intelligence is divided between macro-level language model scaling and the interaction-driven principles of "Era of Experience". Yet, any self-improving architecture ultimately rests upon its underlying optimization engine: if general intelligence requires learning from grounded interaction, the reinforcement learning (RL) update rule itself must be capable of cumulative adaptation. While algorithm self-discovery has produced Disco103 that surpassed PPO to achieve SOTA benchmark performance -- its internal update machinery remains an uninspected black box. We present the first causal mechanistic audit of a self-discovered RL rule, structured directly around the five pillars of the Era of Experience: extended horizon, grounded reward scales, continuing streams, within-lifetime change, and exploration depth. By surgically pinning, freezing, and transplanting recurrent states while holding meta-parameters fixed, we test when learning history acts as an asset or a burden. Three findings organize the audit: (1) Recurrent history actively expands usable reward scales, sustaining a six-decade window versus three under zero-pinning. (2) Decoupling historical content from its maintenance reveals that the penalty of mismatched history stems from perpetual clamping; allowing imported state to evolve naturally attenuates this burden. (3) Under environmental change, controlling replay retention reverses the apparent adaptation advantage over DQN, demonstrating that external data turnover can confound internal plasticity. Validated through capability thresholds and ported to a second rule (OPEN), this work grounds macro-RSI ambitions in micro-level learning dynamics, establishing a foundational audit standard for next-generation, self-evolving RL algorithms.

Title:Self-discovering RL in the Era of Experience: Is Learning History an Asset or a Burden?

Abstract:The pursuit of recursive self-improvement (RSI) toward general intelligence is divided between macro-level language model scaling and the interaction-driven principles of "Era of Experience". Yet, any self-improving architecture ultimately rests upon its underlying optimization engine: if general intelligence requires learning from grounded interaction, the reinforcement learning (RL) update rule itself must be capable of cumulative adaptation. While algorithm self-discovery has produced Disco103 that surpassed PPO to achieve SOTA benchmark performance -- its internal update machinery remains an uninspected black box. We present the first causal mechanistic audit of a self-discovered RL rule, structured directly around the five pillars of the Era of Experience: extended horizon, grounded reward scales, continuing streams, within-lifetime change, and exploration depth. By surgically pinning, freezing, and transplanting recurrent states while holding meta-parameters fixed, we test when learning history acts as an asset or a burden. Three findings organize the audit: (1) Recurrent history actively expands usable reward scales, sustaining a six-decade window versus three under zero-pinning. (2) Decoupling historical content from its maintenance reveals that the penalty of mismatched history stems from perpetual clamping; allowing imported state to evolve naturally attenuates this burden. (3) Under environmental change, controlling replay retention reverses the apparent adaptation advantage over DQN, demonstrating that external data turnover can confound internal plasticity. Validated through capability thresholds and ported to a second rule (OPEN), this work grounds macro-RSI ambitions in micro-level learning dynamics, establishing a foundational audit standard for next-generation, self-evolving RL algorithms.

Artificial Intelligence (cs.AI); Machine Learning (cs.LG)

arXiv:2609.35897 [cs.AI]

(or arXiv:2609.35897v1 [cs.AI] for this version)

https://doi.org/10.48550/arXiv.2609.35897 Focus to learn more arXiv-issued DOI via DataCite (pending registration)

analysis_scripts/build_data_figures.py

analysis_scripts/figure_style.py

analysis_scripts/openscience.mplstyle

analysis_scripts/plot_sources/behavior_panorama.py

analysis_scripts/plot_sources/legacy_plots.py

analysis_scripts/plot_sources/recovery_panorama.py

analysis_scripts/plot_sources/replay_detail.py

analysis_scripts/plot_sources/scale_panorama.py

analysis_scripts/plot_sources/scale_transplant_detail.py

evidence_manifests/behavior_panorama.json

evidence_manifests/legacy_plots.json

evidence_manifests/recovery_panorama.json

evidence_manifests/replay_detail.json

evidence_manifests/scale_panorama.json

evidence_manifests/scale_transplant_detail.json

figure_data/app03_interface.data.json

figure_data/app04_streams.data.json

figure_data/app06_exploration.data.json

figure_data/app07_baselines.data.json

figure_data/integrated_behavior.data.json

figure_data/integrated_scale_interface.data.json

figure_data/replay_main.data.json

figure_data/scale_main.data.json

figure_data/transplant_main.data.json

figure_data/unified_behavior_summary.data.json

figure_data/unified_mapping.data.json

figure_data/unified_recovery_catch.data.json

figure_data/unified_recovery_overview.data.json

figure_data/unified_scale_full.data.json

figure_data/unified_state_scale.data.json

(26 additional files not shown) You must enabled JavaScript to view entire file list.

arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.

Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.

Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.

analiza aiqbench

Self-discovering RL in the Era of Experience: Is Learning History an Asset or a Burden?

Subiecte conexe