AI assistants: is treating every request the same way fair?
A no-code assistant can route training requests, assign priorities or draft HR responses. The management question is specific: how can we check that this initial routing does not disadvantage some employees for reasons unrelated to their request?
What the literature establishes
Jason A. Colquitt (2001, Journal of Applied Psychology) provides validation evidence for distinguishing four dimensions of organisational justice: distributive, procedural, interpersonal and informational. A favourable outcome, a procedure considered fair, respectful treatment and an adequate explanation are therefore not interchangeable. For an AI assistant, this distinction provides an assessment framework, not proof of fairness.
Explanations matter, but do not replace outcomes
Jerald Greenberg (1990, Journal of Applied Psychology) studied responses to temporary pay cuts in an industrial setting. His field study links the quality of the explanation accompanying these cuts to perceptions of inequity and observed theft. It suggests that the way an unfavourable decision is presented matters, without establishing that an explanation is sufficient to make that decision fair.
The premature conclusion
One might conclude that programming an identical rule and a polite explanation for every request is enough. That would confuse consistency of treatment with the relevance of the criteria: a uniform rule may overlook a legitimate job constraint or indirectly favour people who are familiar with administrative terminology. The task is to define which differences should count, and which should not change the routing. (our executive and employee training programmes)
What the evidence does not establish
These studies concern neither generative models nor no-code agent-building platforms. They therefore do not demonstrate that an assistant improves or undermines fairness, and a measure of perceived justice is not a discrimination audit. In practice, checking only the courtesy of responses leaves unanswered the question of how favourable and unfavourable routing decisions are distributed.
A practical check in Fribourg
In SHR's programme « Créer ses assistants et agents IA sans coder », an exercise set in Fribourg could use a fictional training-request triage process spanning the agri-food sector, bilingual industry and the university hub. Create pairs of synthetic requests with identical needs, eligibility and job constraints, changing only a factor declared irrelevant, such as semantically equivalent wording in French or German. Across a predetermined number of repetitions, measure the proportion of pairs routed differently, retaining the inputs, outputs and configuration version. Each discrepancy should be reviewed before deployment: this test checks a bounded property of the process, not the overall fairness of the system. To go further: explore the Building AI Assistants and Agents Without Coding training in Fribourg in the canton of Fribourg, or browse our executive and employee training programmes in Switzerland.
In pictures: Building AI Assistants and Agents Without Coding in Fribourg
- building ai assistants and agents without coding
- Fribourg
- research
- training fribourg