Intrinsic Goals for Autonomous Agents: Hierarchical Exploration Captures Future Volitional Action in Drosophila

DIAYN, a hierarchical competency algorithm, achieves the strongest joint behavioral and neural prospective coding performance with Drosophila melanogaster, matching future volitional action during exploration.

• Why it matters: The mechanisms enabling intrinsic, autonomous behavior without external cues remain unclear, and linking them to neural representations of future decisions is a key gap.
• What they did: Using whole-brain recordings identifying spontaneous exploratory decisions more than 10 seconds in advance, the authors train intrinsically motivated agents with no external reward using 20 intrinsic motivation algorithms and evaluate frozen policies in a virtual tethered-ball assay.
• The result: DIAYN learned policies closely capture individual variability in fly exploration strategies and its latent activity closely captures the sum-difference rule in interhemispheric encoding of future kinematics, enabling reverse-engineering of animal autonomy.