Who leads the trade? Responsibility, algorithmic influence, and regret in financial human-algorithm collaboration
Year: 2026
Type: article
Abstract: Financial decision-support is moving toward co-pilot designs, in which an algorithm advises and the person decides. Where authority is shared, so is authorship of the outcome. This study separates what the person decided from how the outcome turned out by recording the decision first: in a trading simulator, 88 participants committed to a direction and a stake before each recommendation appeared, then rated the responsibility they attributed to themselves, the influence they attributed to the algorithm, and their regret after the result (1232 trials). The advisor’s recommendations were fixed in advance and correct on 42.9% of rounds, isolating perceived agency from competence. Advice moved the committed plan on 50.4% of conflicting trials against a 3.2% within-subject baseline, and cost accuracy (45.0% to 41.9%, p = . 012 ). Losses were associated with lower self-attributed responsibility ( b = − 0.49 ) and higher attributed algorithmic influence ( b = + 0.58 ). In exploratory analyses of conflict trials, what the participant did outweighed what happened to them by roughly nine to one: abandoning the plan shifted responsibility by b = − 4.12 against −0.46 for losing, and the two did not interact. Affect did not follow attribution: abandoning the plan raised rather than lowered regret ( OR = 1.98 ). Direct experience with the advisor lowered stated willingness to use it for real money ( d z = − 1.04 ), while stated attitudes did not reliably predict behaviour. A person can assign most of an outcome to the algorithm and still feel the regret themselves.
Author: Anna Kovalchuk
Language: English
Subfield: Human Factors and Ergonomics
Field: Social Sciences
Domain: Social Sciences
Open Access status: gold
Estimated APC paid: $1,540
OpenAlex ID: W7224272689