What You See Is All There Is

In the AI Story

The canonical experimental demonstration: participants given one-sided arguments about legal cases expressed greater confidence in their verdicts than participants given both sides. The one-sided presentation produced a more coherent story, and coherence drove confidence. The missing arguments did not reduce certainty — they were not experienced at all. This reveals that human confidence is calibrated to narrative smoothness, not to informational completeness.

Claude's outputs are optimized for coherence. They are articulate, well-structured, apparently complete. This is precisely the output profile that activates WYSIATI in its most potent form. The smoothness creates a specific trap: a rough, obviously partial output would signal gaps and trigger System 2. A polished output does the opposite — it says, implicitly, "everything you need is here."

The compounding effect is the central danger. Claude's generation process also operates under a WYSIATI-like constraint: the model constructs outputs from patterns in its training data without mechanism for flagging what the training data lacks. When the training data is sparse on a topic, the output's fluency does not waver. The gaps are invisible to the machine for the same structural reason they are invisible to the human: neither system has a mechanism for detecting the absence of information.

The Deleuze Error from The Orange Pill is WYSIATI's textbook illustration. Claude produced a fluent passage connecting flow to Deleuze's concept of "smooth space." The connection was philosophically wrong. The wrongness was concealed by the coherence of the passage. Segal read it twice, liked it, moved on. Only a nagging feeling the next morning triggered the System 2 check that revealed the error.

The practical countermeasure is effortful and counterintuitive: deliberately asking, against the grain of System 1's satisfaction, what is not here? The question is uncomfortable because it requires looking for something that by definition cannot be seen in the output. But it is the only intervention that opens the gap that coherence closes.

Origin

WYSIATI emerged as a synthesis of decades of experimental findings on confidence, coherence, and the construction of judgment under uncertainty. Kahneman framed it in Thinking, Fast and Slow as the operating principle behind multiple more specific biases: overconfidence, framing effects, base-rate neglect, and more.

The acronym was chosen for its deliberately ungainly sound — Kahneman wanted a phrase that resisted elegant summary, reflecting the blunt, pervasive, nearly invisible character of the phenomenon it describes.

Key Ideas

Coherence drives confidence. Narrative smoothness, not informational completeness, produces the feeling of certainty.

Missing data is invisible. System 1 has no mechanism for flagging absent information; it works with what is present.

Polished output suppresses vigilance. Smooth surfaces conceal gaps that rough surfaces would reveal.

Double WYSIATI in AI. Both the machine's generation and the human's evaluation operate under the same blindness to absence.

The counter-question. "What is not here?" is the only effortful intervention that opens the gap.

Debates & Critiques

Researchers debate whether WYSIATI is a distinct bias or an umbrella term for multiple related phenomena. The practical question for AI — whether its fluent outputs produce measurable increases in WYSIATI-type errors — is beginning to yield empirical data, most of which confirms the prediction.