IA on the job: who checks when everyone can?
An assistant produces a draft procedure, and the team is asked to review it: the arrangement seems prudent. Yet the management question is precise: does asking everyone to check ensure that someone actually does?
What the literature establishes
Bibb Latané, Kipling Williams and Stephen Harkins (1979, Journal of Personality and Social Psychology) showed, in experimental sound-production tasks, that individual effort could decrease when participants believed they were contributing to a collective output. This phenomenon, known as social loafing, does not mean that people are generally less conscientious in teams. It establishes that the conditions under which a contribution is made and distinguished can affect the effort invested.
The mechanism: a contribution that genuinely matters
Steven Karau and Kipling Williams (1993, Journal of Personality and Social Psychology) synthesised research on social loafing and proposed a model linking effort to the expected usefulness of an individual's contribution. Their analysis highlights, among other factors, the possibility of evaluating that contribution and the importance attached to the task. For reviewing AI-generated material, this suggests a working hypothesis: a collective request may be less motivating than a specifically assigned check.
The premature conclusion
One might conclude that appointing a reviewer is enough to make any generated document reliable. These studies do not support that conclusion: making a contribution identifiable provides neither the expertise, nor the time, nor the reference materials needed for verification. The assignment must therefore concern a bounded, feasible check, rather than a general responsibility to 'guarantee the AI'. (our executive and employee training programmes)
What the evidence does not allow us to assume
These studies address neither generative assistants nor document validation in precision manufacturing. Professional review combines technical understanding, evidence gathering and production constraints, extending well beyond the original experimental tasks. Counting the people consulted, or collecting their agreement, therefore does not measure the checks actually performed.
A practical check in Neuchâtel
As part of SHR's 'IA on the job' programme, an exercise could compare two review arrangements within a Neuchâtel team working in microtechnology, watchmaking or precision microelectronics, using comparable fictional drafts with no production use: a collective request or checks assigned to named individuals. Known discrepancies against approved references would be introduced into the drafts, and participants would alternate between the arrangements. The measure would be the proportion of those discrepancies correctly identified, reported alongside the time spent checking; this would allow a local assessment of the value of explicit assignment, without confusing a signature with verification. To go further: explore the AI on the Job training in Neuchâtel, or browse our executive and employee training programmes in Switzerland.
In pictures: AI on the Job in Neuchâtel
- ai on the job
- Neuchâtel
- research
- training neuchâtel