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.