IA on the job: does revising a proposal make us less critical?
An AI-generated memo rarely becomes a deliverable without revision, and the time spent revising it seems to justify confidence in the result. The managerial question is more precise: how can we check whether we are defending the document’s quality rather than the work we have invested in it?
What the literature establishes
Michael I. Norton, Daniel Mochon and Dan Ariely (2012, Journal of Consumer Psychology) show experimentally that taking part in making an object can increase the value people assign to it. In their experiments, this effect depends in particular on successful completion: expending effort is not sufficient under every condition. These findings concern assembled objects, not professional documents revised with AI.
Ownership can also change evaluation
Daniel Kahneman, Jack L. Knetsch and Richard H. Thaler (1990, Journal of Political Economy) show in trading experiments that possessing a good can change its valuation. This mechanism is distinct from the effort of making something: it cautions against treating assigned value as a straightforward reading of an object’s properties. Applied to a revised memo, it suggests a hypothesis to test rather than an established finding: becoming the author of its final version could make impartial evaluation harder.
The premature conclusion
One might conclude that employees should be prevented from reworking AI proposals to preserve their critical distance. That would confuse a risk of attachment with an absence of added value: revision can correct errors and contribute essential knowledge. The issue is to distinguish the contribution to production from the final assessment, not to eliminate that contribution. (our executive and employee training programmes)
What the evidence cannot establish
These studies establish neither the existence nor the magnitude of an ownership effect when revising AI-generated texts at work. Professional documents involve compliance constraints, domain-specific criteria and collective consequences that cannot be reduced to the valuation of an object. A preference for one’s own version may therefore reflect a genuine improvement, a particular attachment, or both.
A practical check in Zurich
As part of “IA on the job”, a Zurich team in financial services, insurance, tech or a European headquarters could test this issue using non-sensitive internal memos produced with an approved tool. For each memo, the reviser and a colleague not involved in the work would separately assess the initial and revised versions against a predefined rubric covering accuracy, coverage of requirements and clarity; the colleague would receive the versions in random order, with their provenance concealed. The measure would be the gap between the score improvement attributed to the revisions by the reviser and that attributed by the colleague, alongside the specific defects identified. A recurring gap would not prove ownership bias, but would provide a verifiable reason to introduce independent review before circulation. To go further: explore the AI on the Job training in Zurich, or browse our executive and employee training programmes in Switzerland.
In pictures: AI on the Job in Zurich
- ai on the job
- Zurich
- research
- training zurich