Responsibility, Law & Governance
Responsible AI
The umbrella term for deployment with a backbone: assess risks per use case, consider those affected, plan failure paths, name accountability. Not as an ethics poster but as operating practice — who checks, who stops, who is liable. The test is simple: can you explain the deployment publicly without squirming?
An example
Two companies have an AI policy. At one, it says that AI is used “responsibly, transparently and fairly.” At the other, it states which systems are in use, who is responsible for each of them, which decisions may not be automated, how an employee reports an error and what happens then. Only the second document changes anything when an error occurs.
A closer look
In practice this means four things: named owners, documented purposes and limits, logs of decisions, and a way for affected people to object. Everything else follows from these.
Why it matters
The difference between a statement of intent and control lies in four points that can be answered with yes or no: Is there a register of the systems in use? Does each have a named owner? Is there a reporting channel for misbehavior that employees know about? And is there a body that is allowed to shut a system down? Whoever has these four hardly needs the principles. Whoever lacks them has nothing, principles or not.
Commonly misunderstood
Principles papers without accountability change nothing. The test is whether anyone is authorized to stop a running deployment.