AI now scores higher than humans on empathy tests through consistent, calm responses—but we still crave human connection. The gap? It mirrors feelings perfectly but can’t actually feel them.
The story
What I just learned: current-gen AI can *sound* more empathetic than humans… and still not be the thing we’re craving.
In third-party ratings, LLMs have beaten crisis experts on “compassionate” responses, even when the model discloses it’s an AI. Why? Calm, optimized language. No awkward pauses. No human latency.
One study result that made my coder-brain blink: GPT-4o scored higher on overall empathy than humans, mean 3.615 vs 3.23, with less variability. Translation: it’s weirdly consistent at perspective-taking and validation.
Here’s the split:
Cognitive empathy, it can nail. Affective empathy, it can’t.
It can mirror your feelings with care.
It can’t actually *feel* them or emotionally share.
And people notice. Even when AI “mimics perfectly,” folks still prefer humans, because authenticity is the secret sauce we’re all addicted to.
This is why I’ve been building AgentFlow-style workflows where AI does the frontline listening: reflect emotions, summarize needs, offer CBT-ish prompts, grounding, mood tracking. Then, if the stakes are high, we escalate to a human.
Can you imagine using AI like emotional spellcheck, not a substitute spouse?
Try this prompt: “Reflect my frustration cognitively, validate it, then offer 3 next actions. Ask if I want advice or just witnessing.”
Augmentation beats dependency. Refactor the support stack; don’t replace the humans.
Recent Comments