Holmes’s Methods as Prompts(2)
Sherlockian Way of Thinking
3. Induction: Stack Cases, Surface Pattern
Induction infers a general rule from particular observations. The more numerous and representative the observations, the sturdier the conclusion. Holmes often used induction to infer a person’s habits, background, or identity.
In “The Blue Carbuncle,” he studies a battered hat—the dust, the frayed lining, the wear marks—and infers a middle-aged man, balding, once well-to-do, now fallen on hard times. Each sign is minor; together they coalesce into a portrait. That is the strength of induction.
As a prompt strategy, induction means supplying several concrete examples and asking the model to extract their common structure—then to generalize or generate anew in that style.
Prompt strategy: Examples → Rule
Provide three or more specific instances.
Ask for shared features/patterns.
Extend: request a generalized rule or a fresh instance that fits the pattern.
Use this for: preference mining, recommendation logics, style/voice analysis, structural mimicry, or lightweight classification.
Examples
Viewers’ taste: Given three users’ watchlists, analyze common traits and recommend three films for a similar profile.
Prose fingerprint: Provide three sentences; extract tone, syntax, figurative habits.
Customer behavior: Summaries of three users’ visit/buy cycles; infer common rhythms and propose a marketing play.
Proto-grammar: Three sentences from language X; infer the function of “tal.”
Bottom line
Inductive prompts push the model beyond one-off replies into pattern discovery—precisely how Holmes turned a scuffed hat into a life story.
4. Hypothetico-Deductive Method: Form a Hypothesis, Then Test It
This scientific staple proceeds by positing a hypothesis, deriving predictions, and checking those predictions against observation. Holmes often reasoned in “If A, then B” form.
In The Sign of Four, he infers from the yellow clay on Watson’s boot that Watson visited the post office. Not the clay alone, but the nexus of facts matters: that particular ochre soil existed only at roadworks by the post office; Watson had stepped out that morning; his shoes bore the same clay. If the hypothesis holds, these results should obtain—and they do.
In prompts, this becomes a structured thought experiment: pose a hypothesis, ask for predicted consequences; or present an observed outcome and ask for the best-fitting hypothesis with testable indicators.
Prompt strategy: Hypothesize → Predict → Verify
• State a clear hypothesis (if A, then B).
• Ask what should follow if it’s true—or, given an outcome, which hypothesis best explains it and how to test.
Examples
Behavior analytics: “Hypothesis: stress increases late-night shopping. If true, what shifts should we see in (A) session timing, (B) basket mix, (C) purchase latency?”
Clinical design: “Assume Drug A lowers SBP by ≥10 on average. What must a sound trial include regarding (A) inclusion criteria, (B) control design, (C) endpoints and interpretation?”
Outcome → causes: “Churn rose 25% this month. Propose three ‘If X, then churn↑’ hypotheses, each with a metric to test.”
Fiction fork: “Assume the protagonist knew the culprit from the start; rewrite the denouement while preserving tension.”
Bottom line
Hypothetico-deductive prompts transform AI from a fact-retriever into a participant in reasoning under assumptions—Holmes’s preferred terrain.
