Management Human Because

What If AI Management Makes Workplaces Less Fair, Not More?

The week's convergence of stories — an LLM firing a worker, autonomous drones selecting targets, an AI generating alleged CSAM at scale — invites scrutiny of a consensus assumption running through almost all current AI policy discussion: that AI management and decision-support tools, properly implemented, will reduce human bias in high-stakes decisions and therefore produce more equitable outcomes than human-only management.

The argument is compelling on its face. Human managers are demonstrably biased along lines of race, gender, age, and social similarity. Multiple academic studies from 2022 to 2024 showed meaningful reductions in demographic disparities in hiring rates when AI screening tools replaced initial human review. If a system evaluates performance data without knowing what someone looks like or what their name sounds like, it should theoretically produce fairer results.

The stronger counterargument, however, is not that AI systems are secretly biased in detectable ways — though some clearly are, as the Amazon hiring algorithm case demonstrated — but that AI management tools optimize for measurable proxies of performance defined by humans in contexts that reflect existing power structures. When an LLM fires a worker for low productivity scores, those scores were defined by whoever built the system, calibrated against whoever's performance was used as training data, and deployed in a context where the worker has no visibility into the decision logic.

That opacity may be worse than human bias in one specific way: with a human manager, an employee can argue, appeal, present context, go to HR, and build a legal claim based on documented conversations. With an LLM termination decision, the decision trail may be entirely opaque — not through malicious concealment, but because a model's inference process is not interpretable even to its own developers. The question on the table is whether AI management replaces one form of flawed but challengeable accountability with a different form of accountability that is opaque and therefore unchallengeable.

The indicator worth tracking is whether labor boards or courts issue rulings specifically on the legal capacity of an LLM to serve as an employer's authorized agent in a termination — which would force legal frameworks to catch up with operational reality. Separately, researchers should monitor wrongful termination claims and resolution rates in companies that have deployed LLM management systems versus those that have not. If AI-managed workplaces show lower rates of claims filed, that could mean decisions are fairer — or that workers have less access to the evidence needed to make a claim. Those two outcomes look identical in the filing data and demand entirely different policy responses.