Access to Live AI Advice and Behavior Under Risk: An Incentivized Experiment

URL SCAN: Access to Live AI Advice and Behavior Under Risk: An Incentivized Experiment
FIRST LINE: Economics > General Economics

The Dissection

This is a narrow laboratory test of whether access to live AI advice changes lottery choices and measured risk aversion. It compares a pre-written tool, a live one-shot AI, and a live interactive AI with equivalent information formats, then reports no evidence of changed risk aversion among 158 participants.

The text measures a contained preference response. It does not measure labor displacement, ownership of AI capital, coordination power, productivity, or access to economically necessary work.

The Core Fallacy

The experiment itself may support a modest local conclusion. The fallacy appears when that null result is inflated into a general claim that AI does not materially alter human behavior or economic organization.

“No evidence” of changed risk aversion in a lottery task is not evidence that AI lacks structural power. The experiment does not test P1, P2, or P3. It leaves the wage–labor–consumption circuit entirely outside the laboratory, then observes that the isolated risk preference variable remains stable. That is not a reprieve. It is a measurement of the residue untouched by the actual mechanism of obsolescence.

Hidden Assumptions

• Lottery choices and DOSE risk-aversion measures represent consequential real-world risk behavior.
• Optional use of a live AI reproduces how people rely on AI in repeated, high-stakes environments.
• Equivalent information format isolates the effect of liveness and interaction.
• A sample of 158 can reliably detect meaningful effects, including heterogeneous responses.
• A null effect on risk aversion generalizes to broader economic behavior.
• One-shot exposure captures the effects of habituation, dependence, trust, and repeated delegation.

Social Function

Classification: partial truth with ideological-anesthetic potential.

The legitimate finding is narrow: in this incentivized setup, live AI access did not measurably change risk aversion. The anesthetic effect begins when that result is used to portray AI as merely a neutral advice interface. The study relocates attention from who owns automated cognition and who remains economically necessary to a safe psychometric question. It does not establish that AI is harmless; it establishes only that this particular human preference measure did not move under these conditions.

The Verdict

A valid micro-result, strategically useless as a defense of the post-WWII economic order. Live AI may leave baseline risk aversion intact while still destroying the labor inputs that make most people economically necessary. This paper does not confront the discontinuity. It documents a stable preference inside a system whose productive foundation remains untested.