§ IThe Framework

Three mechanisms, one confusion.

A proposed account of how access to machine-generated reasoning can be mistaken for possession of it.

For most of history, the ability to produce a well-reasoned analysis was reasonably good evidence that its author understood the subject. Writing a competent legal memorandum, a credible business plan, or a working piece of software required the knowledge those artefacts expressed. The artefact and the competence travelled together.

Large language models separate them. A person can now obtain, in seconds, output that bears every surface marker of expertise. The Lisa Latte Effect names what may follow: a confusion, often sincere, between having access to that intelligence and possessing the knowledge, judgment, and experience required to evaluate or apply it.

The concern is not that people use AI. Delegating cognitive work to tools is ordinary and often wise. The concern is a specific misattribution — treating the tool's competence as one's own, and acting with a confidence that only one's own competence would justify.

01

Borrowed Intelligence

The sophistication of AI-generated reasoning becomes psychologically associated with the individual receiving it.

The person asked the question, read the answer, and found it sensible. Each of those acts is genuinely theirs, and together they create a sense of authorship over reasoning they did not perform. Prior work on Internet search suggests that externally retrieved explanations can inflate estimates of one's own internal knowledge; generative systems, which return finished reasoning rather than sources, plausibly intensify the effect.

02

Inflated Confidence

Fluent, authoritative AI output can contribute to confidence that exceeds the user's independent understanding.

Fluency is a cue people routinely use to judge truth and comprehension. Generative output is engineered for fluency: complete sentences, confident structure, the appropriate vocabulary. The framework proposes that this cue is transferred from the text to the self, so that confidence tracks the quality of the prose rather than the depth of one's grasp of it.

03

Execution Disconnect

The ability to articulate a sophisticated recommendation is mistaken for the ability to evaluate or implement it.

Articulation and execution draw on different capacities. A recommendation can be repeated verbatim by someone who could not recognise its flaws, adapt it to a changed circumstance, or carry it through the first unexpected obstacle. The disconnect becomes visible only when the system is absent and the person must act.

How the mechanisms interconnect

A fluent, well-structured answer arrives at negligible cost. Its sophistication is a property of the system, not the reader.

Fig. 2 — Proposed explanatory sequence. This is a hypothesis about how the components may relate, not an experimentally validated causal chain.

§ II The Competence Gap

Apparent expertise is not demonstrated competence.

Everything in the left column can be produced by a model. Everything in the right column must be produced by a person. Select a row to see the question that separates them.

Apparent expertise

Demonstrated competence