Accountability

Trust requires that there is a specific person who can be held responsible if something goes wrong. Not accountable in the abstract sense — accountable in the concrete sense: if I trust you with X and X goes wrong, I can tell you that you failed me, I can take action based on that failure, and you have to face the consequences of your choices.

Accountability is uncomfortable. That is the point. Accountability is what makes it real.

When an organization uses an AI system, someone has to own what happens. If the system produces harmful output, or fails silently, or is deployed inappropriately, someone eventually faces consequences. The question is: who? Is it the person who deployed it? The person who set the rules? The company itself? The designer of the model? This is often murky. But the murkiness is not accidental — it is structural.

An AI system cannot be held accountable in this sense because there is no "it" to hold accountable. Code is not sentient. Code does not make choices. Code executes instructions. When an AI system causes harm, you can investigate what happened, but you cannot look the system in the eye and hold it responsible. You cannot demote it. You cannot ask it to explain its choices. You cannot fire it and prevent it from working elsewhere.

This is why every use of AI in any consequential domain requires a human who steps into the accountability gap. Someone has to be able to say, "I am responsible for this outcome, and I will face the consequences if I got it wrong." Without that, you do not have trust. You have liability. And you have risk.