Definition
Konfirmační bias (confirmation bias) is the systematic human tendency to seek, interpret, weight, and remember information in ways that confirm already-held beliefs, and to symmetrically discount, ignore, or misremember information that disconfirms them. The bias operates across all three stages of belief processing — acquisition (which sources are consulted), interpretation (how ambiguous evidence is read), and retention (which findings are remembered). All three stages reinforce each other: yesterday's biased acquisition becomes today's biased prior.
Within the Kompendium epistemické a kognitivní anihilace, confirmation bias is the most heavily exploited single vulnerability in the entire taxonomy. It is the structural prerequisite for the EA-1.1 family (information-environment manipulation):
- EA-1.1.1 (Selective exposure) — directly exploits acquisition-stage confirmation bias: the target self-administers a curated information diet because disconfirming sources feel cognitively aversive.
- EA-1.1.2 (Saturation tactics) — exploits availability + confirmation: vivid confirming exemplars are easily retrieved, disconfirming statistics are not.
- EA-1.1.3 (Echo chamber engineering) — engineers the social environment so that even casual cross-checking returns confirming results, removing the natural correction that diverse social networks normally provide.
- EA-1.3.1 (Immunization against falsification) — installs cognitive moves ("they would say that, wouldn't they") that route disconfirming evidence as further confirmation.
The kompendium's diagnostic principle: confirmation bias is universal and incurable at the individual level. Defense requires structural rather than psychological measures — pre-committed exposure to disconfirmation, calibration scoring, and external review.
Context
Within the Prismatic Platform, confirmation bias is treated as a first-class structural problem, not a personal weakness:
- The Decision Engine explicitly tracks source diversity in its information graph: a decision based on a low-diversity source set receives a confidence penalty regardless of internal coherence. High coherence + low diversity is a confirmation-bias signature, not a strength.
- The platform's calibration loop compares predicted vs. realized outcomes specifically to detect the systematic over-confidence that confirmation bias produces. A subject whose confirming predictions outperform their disconfirming predictions is showing signature bias even without content-level analysis.
- NABLA Infinity implements mandatory disconfirmation queries: any high-confidence belief must be paired with an explicit list of sources that would disconfirm it; absence of such sources is itself an alarm.
- The Color Teams Red Team builds adversarial information environments to validate that platform-level diversity checks survive deliberate gaming. Czech-language treatment of confirmation bias in EA context: /kompendium/04-mechanismy/biasy/konfirmacni/.
The principled defense, drawn from falsifikace (Popper) and Lakatosian research programmes, is to prefer falsification attempts over confirmation searches. A finding that a belief survives a serious attempt to refute it is epistemically heavier than ten confirming examples. The platform UX exposes this asymmetry: disconfirmation entries weigh more in the Decision Engine's scoring than confirmation entries of equal source quality, mirroring the underlying probability theory (P(E|H) is weak; P(¬E|¬H) is strong when E was a genuine prediction).
See Also
- Kognitivní bias — Parent concept
- Falsifikace — Principled corrective from Popper
- Lakatosův výzkumný program — Distinguishes legitimate vs. degenerative protection of beliefs
- Epistemická anihilace — Threat model exploiting this bias
- Taxonomie EA — Specifically the EA-1.1.x family
- Metakognice — Reflective layer that catches active bias
- Kognitivní anti-rekurzivita — Hardened endpoint
- Kompendium — Full Czech-language taxonomy
Connect & Contribute
Created by Tomáš Korcak (korczis) | Open Source under GHL