Partially observable Szilárd engines

Leo Szilard pointed out that Maxwell's demon can be replaced by machinery, thereby laying the foundation for understanding physical nature of information. Szilard's information engine still serves as a canonical example after almost hundred years, despite recent significant growth area. The role plays reduced to mapping observable data meta-stable memory, which is utilized extract work. While showed map implemented mechanistically, it was chosen priori. choice how construct meaningful memory constitutes demon's intelligence. Recently, shown this automated well. To end, generalized, partially engines were introduced, providing basis processing. Partial observability ubiquitous in real world systems have limited sensor types and acquisition bandwidths. Generalized run work extraction at different temperature, T' > T, from forming process. This enables combined treatment heat engines. We study characteristics intelligent observers introducing model displays richness, its simplicity. A minor change - inserting divider an angle results family Their analysis shows intelligence automated. For each angle, value T'/T, optimal found, enabling with minimal dissipation. Those memories are probabilistic maps, computed algorithmically. discuss they simple system, characterize their performance, compare quality naive, deterministic quantizations observable.