Introduction
Sherlock Holmes described his approach as the “science of deduction,” though modern readers often see abduction and Bayesian reasoning. Doyle’s stories work as a pedagogical world model: clues are partial observations; culprits and motives are latent; each chapter is an update.
Inference tropes
Red herrings, withheld evidence, and unreliable witnesses inject noise and adversarial structure—precisely the complications POMDP research formalizes. Doyle’s narrative pacing controls information flow much as an experiment designer controls disclosure.
Readers as agents
The reader’s own updating mirrors the detective’s; the text is an environment for training interpretive priors. Cognitive narratology explains how readers tolerate ambiguity; interactive mystery games and AI benchmarks operationalize the same loop with measurable policies.
Ethics of suspicion
Holmesian brilliance can romanticize surveillance. The essay asks when inference cultures become predatory—when hidden-state reasoning serves justice versus when it serves control.
Conclusion
The science of deduction is less a finished method than a dramatization of how humans live inside hypotheses—testing, revising, sometimes spectacularly wrong—until the story closes.