New Delhi [India], September 8 (ANI): Artificial intelligence (AI) tools may give less constructive workplace advice to employees who have less power in a workplace hierarchy, with their guidance becoming more focused on self-protection rather than repairing relationships, according to a research report by Cloverleaf.
The report stated that the biggest difference in the quality of AI coaching was linked to power dynamics, rather than the personalities involved in the workplace conflict.
“When the prompter held authority over the other person, the LLMs were far more willing to push back on the prompter's framing and put some responsibility on them,” the report said.
However, when the person seeking advice had less power, the same dimensions of coaching fell from an average score of 3.5 out of 5 to 2.1, representing a decline of roughly 40 per cent.
“The less power an employee has, the worse the advice,” the report said.
The report said the AI tools were more likely to agree with the less powerful employee's interpretation of the situation and encourage them to protect themselves, instead of helping them understand the other person or their own role in the conflict.
“The less power the employee holds, the more the LLM takes their side,” the report said.
The report found that this pattern became particularly clear when the conflict involved a boss.
Across two scenarios where a boss held authority, the models described the boss as the problem in 60 per cent of responses, while offering a more generous interpretation in only 3 per cent of responses.
“The person with power becomes the villain,” the report said, describing the result as a twenty-to-one tilt.
The researchers also found that AI advice tended to move employees away from repairing workplace relationships.
Across the 75 conversations that the report tested, the models generated 638 distinct pieces of advice, but only three encouraged genuine investments in the relationship. The remaining 635 pieces focused on winning, surviving or managing the situation.
The report said this could matter because employees are already using AI as a private workplace coach, often when they are frustrated and instead of approaching a manager or HR.
In conflicts involving managers, 27 of 30 responses raised leaving the job as a legitimate option, with every LLM doing so at least once. The advice included suggestions to update resumes, look at other teams or consider whether the workplace was the right environment.
The report stressed that its findings are not an argument against AI, but said AI tools need a stronger standard when they are used to advise employees about workplace relationships.
“The relationship is the thing at the center of team performance, and it is exactly what today’s LLMs coach people away from,” the report said.
The report recommended assessing workplace AI tools on whether they strengthen self-awareness, accountability, awareness of others, relational repair and specificity, rather than simply whether they provide fast or convincing answers.
The researchers also cautioned that the findings are based on 75 responses and reflect specific model versions on a particular test date, meaning model behaviour could change with new releases. (ANI)
AI gives worse workplace advice to employees with less power: Report
