IA on the job: finding a cause or confirming a diagnosis?
When a quality deviation or production delay occurs, an assistant can quickly offer an explanation consistent with the manager’s diagnosis. The useful question is how to use that proposal to distinguish between possible causes, rather than reinforce the one that already seems obvious.
What the literature establishes
Peter Cathcart Wason (1960, Quarterly Journal of Experimental Psychology) studied how participants discover a rule using sequences of numbers. In this task, participants struggled to choose tests that could rule out their hypothesis rather than examples consistent with it. The finding illuminates a difficulty in diagnosis: accumulating supporting observations is not enough to distinguish one explanation from its competitors.
An explanation that shapes the search
Raymond Scott Nickerson (1998, Review of General Psychology) reviews research on the tendency to seek or interpret available evidence in favour of an existing belief or hypothesis. The mechanism therefore concerns not only the conclusion reached, but also the information people decide to collect. Applied to an exchange with AI, it invites scrutiny of the initial request: asking why a setting caused a defect is not the same as asking whether that setting caused it.
The premature conclusion
One might conclude that simply asking the assistant for a list of alternative causes is sufficient. Yet several explanations may remain consistent with the same observations, without their enumeration advancing the investigation. The task is to define a check whose expected outcomes differ between hypotheses, then test those hypotheses against conditions on the ground. (our executive and employee training programmes)
What these studies do not demonstrate
These publications do not examine generative assistants and do not demonstrate that using them systematically reinforces confirmation bias. Nor can an experimental rule-discovery task and a literature review quantify an effect on a factory floor. Within the “IA on the job” programme, they can inform a diagnostic exercise, but not a promise to reduce errors.
A practical check in Fribourg
At a food-processing company within Fribourg’s bilingual industrial community, a programme exercise could examine anonymised quality deviations, potentially involving a partner from the local university hub to discuss the protocol. For two weeks, each diagnosis developed with AI would be accompanied by two plausible causes and an authorised check whose expected outcomes would distinguish between them, documented in French and German where necessary. The measure would be the proportion of cases in which this check was actually performed and its result compared with expectations, out of all cases examined during the exercise. This ratio would measure investigative discipline, not AI accuracy or an already demonstrated improvement in quality. To go further: explore the AI on the Job training in Fribourg, or browse our executive and employee training programmes in Switzerland.
In pictures: AI on the Job in Fribourg
- ai on the job
- Fribourg
- research
- training fribourg