Actually, if you can program it to take inputs of anonymized employee satisfaction surveys, and objective employee satisfaction data (attrition, absenteeism, etc), it could work.
Especially if the AI’s target goals are public information. Nobody would work for a company that set the “employee happiness” and “corporate ethics” dials to 0 and the “improve net profit” dial to 100.
Not only that, it has been proven again and again that treating well workers actually yields positive results, considering the IA would have the best for the Company as a goal instead of the pure greed of current CEO/Stakeholders, there are big chances that IA CEO would treat workers way better than current status.
The problem is if the IA goal is not the best for the Company, but the best for the Stakeholders short term, then we would be fucked 😅