Cognitive offloading and the speedup illusion in human-AI interaction

URL SCAN: arXiv cs.CY — "Cognitive offloading and the speedup illusion in human-AI interaction" (Submitted 22 May 2026)

FIRST LINE: "Large language models (LLMs) have the potential to boost human productivity by speeding up task completion -- provided users know when to offload cognitive work to them."

THE DISSECTION

This is a behavioral economics study documenting a specific cognitive miscalibration: humans dramatically overestimate the time savings AI provides while accurately forecasting their own independent performance. N=1,237 participants predicted AI-assisted work would be significantly faster than independent work — but completion times were equivalent. Participants also reported lower subjective effort despite equivalent elapsed time.

The paper frames this as a calibration problem: users don't know when to offload efficiently. The implied solution: better mental models of AI capability so humans can deploy it optimally.

The paper is studying a dead-end optimization.

THE CORE FALLACY

The entire research frame assumes the relevant variable is human task completion speed — that the bottleneck is human decision-making about when to use AI. This misidentifies the location of obsolescence entirely.

Under the Discontinuity Thesis, the question is not whether humans accurately forecast AI time savings. The question is whether the tasks themselves survive at scale. The paper treats cognitive work as an ongoing enterprise that humans will continue performing — possibly faster, possibly with better AI calibration — and asks how to make that transition smoother.

It does not ask: what happens when the work stops existing?

The behavioral artifact the paper documents — the "speedup illusion" — is not a solvable calibration problem. It is a symptom of a deeper structural misalignment: human cognitive architecture evolved to model other humans, not probabilistic systems with non-human performance characteristics. This misalignment is not a bug. It is the mechanism by which displacement proceeds without triggering defensive response.

HIDDEN ASSUMPTIONS

1. Productivity as the operative variable. The paper assumes faster task completion retains economic meaning. Under P3, productivity gains for individual cognitive workers become irrelevant when the labor category itself is being automated.

2. Worthwhile work continues existing. Every sentence assumes the tasks participants are forecasting about remain necessary. The entire experimental design presupposes the continuity of cognitive labor markets.

3. Human judgment remains the frame. The study treats miscalibrated human expectations as the problem to be solved — rather than as the natural cognitive state of a population undergoing structural displacement.

4. Individual optimization is the relevant level. The paper addresses how one person decides when to use AI. It does not address the coordination failure that occurs when millions of people make this decision simultaneously in a labor market context.

SOCIAL FUNCTION

This paper performs transition management — specifically, it normalizes the idea that human-AI collaboration is a solvable engineering problem for individuals. It locates the problem in human cognition (miscalibration) rather than in structural economic displacement (AI capability trajectories). This framing keeps the research fundable, the recommendations actionable, and the implied future comfortable.

It is also an example of prestige signaling within institutional lag — a 2026 paper studying "frontier" questions about human-AI calibration while the structural question of whether that calibration matters is already being settled by capability curves outside the study's frame.

THE VERDICT

The speedup illusion is real. The miscalibration is genuine. The behavioral finding is clean.

But the paper is studying the wrong variable with the wrong urgency. It documents a cognitive artifact at the exact moment that the tasks this artifact concerns are being automated out of relevance.

The Discontinuity Thesis reads this study as an inadvertent census of the psychologically displaced: a population already experiencing the subjective phenomenology of cognitive obsolescence — lower effort reports, distorted time perception, systematic miscalibration about their own productive value — before the structural displacement has fully arrived. The cognitive offloading isn't a strategy. It's a rehearsal for permanent irrelevance.

The paper's subjects aren't learning to use AI better. They're learning to measure their own redundancy more accurately.