There’s a difference. When you program a machine it follows rigid logic. It’s predictable.
When you train a machine it does not. It can make its own inferences and operate outside of strict parameters. It can also make bad inferences, what we call AI hallucinations.
I don’t know that what you’re saying is wrong about avoiding responsibility, but programmed is not the right word for what’s basically a genie in a bottle. And we still hold accountable the member of the tribe that lets the genie out, or should anyway.