AI systems currently understand human motivation the way a dictionary understands a conversation

AI pattern-match emotional language to emotional labels, but they have no causal model of why people do what they do. When someone says "you never listen to me," a current AI detects frustration and responds to the surface expression. It doesn't trace the frustration to the unmet need beneath it — the need to feel heard, to matter, to have influence. The difference matters because the surface expression and the underlying need often require completely different responses. Responding to the surface creates more conflict. Responding to the need resolves it.

This is also why AI mental health tools get criticized for feeling scripted — they recognize emotional categories but not motivation structure. And it's directly connected to AI alignment: a model that can't accurately infer what humans actually need will systematically produce responses that are superficially pleasing but miss the point. That's not alignment — that's sophisticated sycophancy.