Abductive Doubles
An abductive double is an output that looks like the product of genuine discovery but fails one of the three logical requirements of abduction. The Peirce volume identifies three varieties: the unmotivated hypothesis (a clever connection that responds to the prompt rather than to a genuine anomaly), the overdetermined hypothesis (a suggestion so well-supported by training data that it carries no genuine explanatory risk), and the simulated surprise (meta-level astonishment at the machine's capability mistaken for object-level encounter with anomaly). Each variety presents a distinct risk, and each is diagnostically identifiable through careful attention to whether the surprise is genuine, whether the hypothesis is responsive, and whether the inference carries real risk.
In the AI Story
The unmotivated hypothesis is the most common. The AI generates a connection — a structural analogy, an unexpected example, a reframing — that is clever and well-articulated. But it does not respond to a genuine anomaly in the inquiry. It responds to the prompt. The human asked for a connection; the machine produced one. The connection may resolve the difficulty as described in the prompt while missing the actual difficulty entirely. Segal's Deleuze episode in The Orange Pill is the paradigm case: a beautifully constructed connection to a philosophical concept the machine had deployed incorrectly.
The overdetermined hypothesis is harder to detect because it is typically accurate. It is correct, well-supported, relevant. It is also inert. It does not open new lines of inquiry. Genuine abduction involves venturing beyond the evidence — the hypothesis is a guess, and the knowledge that it is a guess motivates testing. The overdetermined hypothesis involves no such venture. The comfort it produces is the diagnostic sign. Genuine abduction is exhilarating, unsettling, and uncertain; overdetermined hypotheses are merely comfortable.
The simulated surprise is the most insidious. The human experiences a frisson of recognition at the machine's output — I hadn't thought of that. The frisson has the phenomenological texture of the surprising fact that initiates abduction. But the surprise may be about the machine's capabilities rather than about the subject matter. The distinction matters because only object-level surprise initiates genuine inquiry into the subject; meta-level surprise initiates only admiration of the tool.
The Peirce volume proposes a three-step diagnostic protocol: Is there a genuine surprising fact? Does the hypothesis respond to the specific anomaly? Does the hypothesis carry genuine explanatory risk? The protocol cannot be automated — it requires the inquirer's judgment at every step, exercising precisely the evaluative capacities that distinguish genuine inquiry from its simulation.
Origin
The concept is a direct extension of Peirce's distinction between genuine abductive inference and its formal imitations. Peirce himself warned that not every proposed hypothesis constitutes real abduction — the hypothesis must respond to a specific anomaly and carry genuine plausibility to initiate inquiry.
The Peirce volume sharpens the classical distinction into a diagnostic taxonomy for AI-assisted work, naming the three varieties and specifying their detection.
Key Ideas
Unmotivated. Responds to the prompt, not to a felt difficulty in the inquiry.
Overdetermined. Statistically safe, correct, and inert — no explanatory risk, no productive discomfort.
Simulated surprise. Meta-level astonishment at the machine mistaken for object-level encounter with anomaly.
Diagnostic protocol. Three questions about surprise, responsiveness, and risk that no automation can perform.
Appears in the Orange Pill Cycle
Further reading
- Erik Larson, The Myth of Artificial Intelligence (Harvard, 2021)
- Harry Frankfurt, On Bullshit (Princeton, 2005)
- Bent Flyvbjerg, "The Big Dig Test" (2025) — on confident wrongness in AI output
- Catherine Legg, "Peirce on Signs, Sentiments, and the Limits of AI" (forthcoming)