CONCEPT

The Anchoring Effect

Tversky and Kahneman's 1974 demonstration that estimates start from an initial value and adjust insufficiently — the bias that makes every pre-AI projection of what is possible systematically wrong.

Anchoring is the cognitive operation by which an initial value — sometimes wholly arbitrary, sometimes deeply experiential — becomes the starting point from which subsequent estimates are generated by insufficient adjustment. Tversky and Kahneman demonstrated the effect with roulette wheels, calculator displays, and expert appraisals, showing that even manifestly irrelevant anchors move subsequent estimates in predictable directions. In the AI transition, the anchors are not arbitrary numbers but decades of professional experience, institutional memory, and accumulated expertise. The senior engineer's estimate of how long a project should take is anchored on hundreds of similar projects completed without AI assistance; the anchor is genuinely experiential, but it was set in a different world, and the adjustment from it is insufficient by orders of magnitude. The Trivandrum six-week estimate that collapsed to three days is the canonical illustration.

In the AI Story

Hedcut illustration for The Anchoring Effect — The Anchoring Effect

The 1974 paper in Science — 'Judgment under Uncertainty: Heuristics and Biases' — introduced anchoring as one of three foundational shortcuts, alongside availability and representativeness. The paper's roulette-wheel experiment remains the cleanest demonstration: subjects asked to estimate the percentage of African countries in the United Nations gave estimates correlated with a number obtained by spinning a wheel in their presence. The wheel was visibly irrelevant. The anchoring effect operated anyway.

In organizational settings, anchoring produces a systematic lag between what AI makes possible and what planning assumes possible. Hiring plans anchored on pre-AI productivity assumptions. Project timelines anchored on pre-AI development speeds. Competitive assessments anchored on pre-AI capability distributions. The adjustments from these anchors are insufficient not because the planners are incompetent but because the magnitude of the change exceeds the range the cognitive system treats as plausible. A twenty-fold productivity multiplier falls outside the range of plausible adjustment from any anchor set in the pre-AI world — the cognitive system discounts it as noise.

The interaction with framing produces compound distortion. The expert anchored on pre-AI experience evaluates the AI transition through a loss frame, because the anchor makes the gap between old and new capability feel like devaluation rather than expansion. The novice with no anchor evaluates the same transition through a gain frame. The result is a disagreement that looks factual but is actually a function of different cognitive contexts producing different evaluations of identical evidence.

The Orange Pill's description of the anchoring gap — the widening discrepancy between what the anchored mind expects and what the changed environment delivers — captures the experiential dimension of the bias. The gap is not closing. It is widening with each month of AI capability advance, because the adjustment process is sequential and the environmental change is exponential. The vertigo that the book names is, in Tversky's terms, the felt experience of an anchor that cannot be released and a reality that has moved beyond its reach.

Origin

Anchoring was formalized in the 1974 Science paper, though earlier experimental work in the 1960s on quantitative estimation had hinted at the mechanism. The generality of the effect — its operation across domains, cultures, and levels of expertise — emerged through two decades of subsequent replication.

Nicholas Epley and Thomas Gilovich's 2001 work distinguished between externally-provided anchors (which operate through priming) and self-generated anchors (which operate through insufficient adjustment), deepening the theoretical understanding of the mechanism and suggesting why the bias is so difficult to correct.

Key Ideas

Starting point dependence. Estimates begin from an initial value and adjust from there; the starting point determines the range within which the final estimate falls.

Insufficient adjustment. Adjustment from the anchor stops at the boundary of plausibility, which is itself defined relative to the anchor.

Expertise as anchor. Professional experience is a powerful, deeply felt anchor that resists updating even when the expert knows the anchor is obsolete.

The anchoring gap. In rapidly changing environments, the sequential adjustment process falls behind the environmental change, producing systematic divergence between expected and actual outcomes.

Frame interaction. Anchoring and framing compound each other: the anchored mind selects frames that make the anchor feel protective, which further entrenches the anchor.

Debates & Critiques

Whether anchoring reflects a single mechanism or multiple underlying processes remains debated. Some researchers argue that externally-imposed anchors (the roulette wheel) operate via different neural machinery than experience-based anchors (years of professional practice). The practical implication is that debiasing strategies effective against one type may be ineffective against the other.

Appears in the Orange Pill Cycle

Amos Tversky — On AI