Klyubin, A., Polani, D., Nehaniv, C. (2005). Empowerment: A Universal Agent-Centric Measure of Control. 2005 IEEE Congress on Evolutionary Computation, 1, 128-135. doi:10.1109/CEC.2005.1554676.
Read in full on 2026-11-03 (NOTE-1511; THEORY-1344).
Key takeaways
- Empowerment = max_{p(aⁿ)} I(Aⁿ_t; S_{t+n}) (eq. 3), open-loop, in bits.
- Control the agent cannot perceive does not count (the neurosurgeon example, p. 129).
- Evolved sensors change representation with the niche (Figure 3).
Standing in the record
The first statement of empowerment. Identified in the curation entry of 2026-10-18 as essential for CLAIM-354 and left unregistered then.