AI systems make decisions that affect lives. Who gets to decide whether an AI system gets deployed is itself a consequential decision — and it's often made by the people least affected by it.
A city council votes to deploy facial recognition at transit hubs. The police wanted it. The vendor sells it. The transit authority operates it. But the people most affected — commuters in low-income neighborhoods where police deployment is already heavy — weren't asked. When a wrongful arrest follows, the question of who held authority to make that decision suddenly becomes urgent.
This module teaches you to map power in AI deployments — and to argue, rigorously, about who should have had decision-making authority when they didn't.
The city council of Midvale, population 340,000, called a vote on Tuesday, March 15, 2024, at 7 PM. The resolution was to authorize facial recognition surveillance cameras at six major transit hubs across the city. The vote passed 5-4.
The decision had been initiated by the police department, which had requested the system a year earlier. Their stated goal: identify wanted suspects, catch human trafficking victims, and locate missing people. The transit authority was tasked with operating it. The vendor was a private surveillance company. The cost was $2.8 million in grant funding from a federal smart-cities program.
What the council didn't do: they didn't hold community meetings in the neighborhoods where cameras would be concentrated (predominantly low-income, predominantly Black and Latino). They didn't commission an independent audit of the technology's accuracy. They didn't establish an oversight board with residents or civil liberties representation. They didn't clearly define what "identify suspects" meant, or what the grounds were for flagging someone as a person of interest. They didn't answer the question: if facial recognition identifies someone incorrectly, what protections exist before an arrest is made?
Three months later, a 28-year-old man was arrested at a transit hub. He matched a facial recognition alert. The charges were dropped after two days — he had been misidentified. He sued the city. The case raised a question the council had never formally debated: who actually made the decision to deploy facial recognition? The council voted yes. But did that vote mean the council understood what they were voting on? Did the affected communities have a right to be consulted? And who should have authority over a system that could lead to wrongful arrest?
Every AI deployment involves four layers of decision-making authority. Understanding who holds power at each layer is the foundation of thinking clearly about governance.
The developer decides what the technology can do. A vendor who sells facial recognition that is 99% accurate for white men but 70% accurate for Black women has already made an ethical choice — before any city council voted. The developer cannot escape responsibility by claiming "the client decides how to use it." The developer chose what capabilities to build and what accuracy tradeoffs to accept.
The deployer (in Midvale's case, the city council) decides whether this specific tool gets used in this specific context. But do they decide with full information? Have they consulted the people most affected? Have they established oversight? A council vote with no community input and no clear accountability structure is a decision made in an information vacuum.
Regulators (government agencies, courts, oversight boards) set the legal boundaries within which AI systems can operate. In the US, there is almost no federal regulation of police use of facial recognition. That absence is itself a power choice — it cedes authority to local police chiefs and vendors to decide how surveillance systems operate.
The people most affected by an AI system — the ones who can be arrested because of its decision, who lose employment because of its judgment, who are denied services because of its output — have the least formal authority. But they have the most at stake. Fair governance requires bringing their voice into the decision before deployment, not just in lawsuits after harm.
The question is never "who gets to make the final call?" It's "who was missing from the room when the decision was made, and should they have been there?"
When analyzing who had authority (or should have had authority) in an AI deployment, three questions cut through the noise.
In Midvale, the city council could say yes. But who could say no? The affected communities had no formal power to block deployment. The civil liberties groups had no seat at the table. The police could push for implementation without community consent. Unequal authority is the core problem.
The commuters who would be surveilled weren't consulted. The people in neighborhoods where police deployment is heaviest weren't asked if they wanted facial recognition at their transit hubs. Fair process requires that affected people have a voice in the decision, not just legal recourse after harm.
Once facial recognition is live, who monitors it? Who has authority to shut it down if it's causing harm? Who investigates misidentifications? If the answer is "nobody outside the police department," then accountability is a fiction.
These questions aren't about perfect democracy — they're about ensuring that power is exercised with the input of the people whose lives are on the line.