AI adoption: should every use be traceable?
AI tools can leave detailed records of prompts, revisions and outputs. The management question is specific: which records should be reviewed to control risks without turning every experiment into a demonstration of compliance?
What the literature establishes
Ethan Bernstein (2012, Administrative Science Quarterly) examines, in an industrial setting, the paradox of transparency that can hinder the learning and performance it is intended to support. His field study shows that limited protection from observation can facilitate experimentation and improvements in work. This finding does not establish that AI should be exempt from oversight; it challenges the assumption that greater visibility means better control.
Accountability does not have a single effect
Jennifer Lerner and Philip Tetlock (1999, Psychological Bulletin) show in their literature review that the effects of accountability depend, among other things, on when it is announced and to whom an explanation must be given. Depending on the conditions, it can encourage more thorough consideration or adjustment to an expected position. Applied to AI, this mechanism suggests that asking employees to explain their validation criteria is not the same instruction as having their manager review every prompt.
The premature conclusion
One might conclude that all traceability should be reduced to encourage adoption. That would confuse a protected space for experimentation with a lack of responsibility for an output put into use. The useful distinction concerns the purpose of oversight: reconstructing a sensitive decision is not the same as continuously observing each employee’s working process. (our executive and employee training programmes)
What these findings cannot settle
These studies do not address the current adoption of generative AI in Lugano businesses. They establish neither a retention period for prompts nor a universal list of records to collect. Applicable obligations, data confidentiality and the risks associated with each use must therefore be examined separately; an activity dashboard does not answer these questions.
A practical check in Lugano
As part of SHR’s programme « Conduire l’adoption de l’IA dans son équipe », a Lugano team could review all managerial requests to access its AI usage records over two weeks, without changing mandatory logs. In a bank, fashion business, trading company or Italian-speaking family SME, each request would be recorded in Italian, stating its purpose, the decision it should inform and the data actually needed. The measure would be the proportion of requests leading to a documented corrective, validation or risk-control action, rather than a simple observation of activity. This ratio would not prove that oversight is effective, but it would provide a documented basis for discussing access to records whose purpose remains unclear. To go further: explore the Leading AI Adoption in Your Team training in Lugano in the canton of Ticino, or browse our executive and employee training programmes in Switzerland.
In pictures: Leading AI Adoption in Your Team in Lugano
- leading ai adoption in your team
- Lugano
- research
- training lugano