Asking AI to confirm a decision already made?
Asking AI to analyse a project can make a managerial preference look like an independent conclusion. The useful question is specific: how can we distinguish a consultation that tests a decision from one that supplies arguments in its favour?
What the literature establishes
Charles G. Lord, Lee Ross and Mark R. Lepper (1979, Journal of Personality and Social Psychology) studied how supporters and opponents of capital punishment assessed conflicting research findings about its deterrent effect. Participants tended to assess evidence consistent with their initial position more favourably, and exposure to opposing findings did not produce the expected convergence. Having arguments on both sides therefore does not guarantee an even-handed assessment.
Reasoning can serve a desired conclusion
Ziva Kunda (1990, Psychological Bulletin) reviews motivated reasoning: the desire to reach a particular conclusion can shape how people search for, construct and assess justifications. This tendency remains constrained by the ability to produce a seemingly reasonable justification; it does not require conscious manipulation. With AI, the managerial risk would be to request an analysis while selecting questions and answers that make an existing preference defensible.
The premature conclusion
One might conclude that consulting AI necessarily reinforces a manager’s convictions. Yet these studies do not examine AI and cannot establish that effect. Instead, they invite scrutiny of the consultation process: asking why a project deserves to proceed is not equivalent to asking what findings would justify abandoning it. (our executive and employee training programmes)
What current practice cannot verify
A well-argued final memo rarely reveals discarded prompts, rejected objections or criteria changed along the way. Counting arguments for and against is no substitute: their evidential weight may differ considerably. The experiment by Lord and his colleagues and Kunda’s review illuminate a reasoning risk without measuring its prevalence in AI-assisted professional decisions.
A practical check in Zurich
In a Zurich team working in financial services, insurance, tech or the European headquarters of an international group, test this protocol on upcoming AI-assisted decisions: before consulting the tool, record the preferred option and an observable finding that could challenge it. Afterwards, ask a colleague to check whether that finding was investigated using a verifiable source and addressed in the decision memo, then calculate the proportion of files meeting both conditions. This rate measures the traceability of the test, not the absence of bias or the ultimate quality of decisions. Within SHR — Swiss Human Resources’ programme « Manager à l'ère de l'intelligence artificielle », this protocol can serve as an exercise in distinguishing justification from investigation. To go further: explore the Managing in the Age of Artificial Intelligence training in Zurich in the canton of Zurich, or browse our executive and employee training programmes in Switzerland.
In pictures: Managing in the Age of Artificial Intelligence in Zurich
- managing in the age of artificial intelligence
- Zurich
- research
- training zurich