Can AI give truly empathetic responses without knowing someone's character?

When a friend talks to you, they decide whether to soothe or amplify your emotions based not just on the situation but on who they deem you to be. If they think you don't stand up for yourself, they amplify your anger to encourage action. If they think you lean toward arrogance, they de-escalate your pride — even when your pride is warranted.

This means genuinely empathetic response requires two things AI currently lacks:

1. Prior knowledge of the interlocutor's character — not just their current emotional state, but their patterns, tendencies, and growth edges
2. A normative framework for character — deciding what character traits are desirable and which the interlocutor should develop or moderate

The "Computer says No" paper points out that even getting the emotion right (a difficult task) is insufficient. Consider: your friend tells you their nephew did well in maths, but you know the nephew cheated. The appropriate empathetic response depends on character knowledge about the friend — their capacity to handle the truth, their relationship with the nephew, their values about honesty.

This connects to Does any single persuasion technique work for everyone? and Why can't chatbots detect when users are ambivalent about change? — both show that generic interventions fail because individual context determines appropriate response. But the character-knowledge requirement goes deeper: it's not just knowing traits, but making normative judgments about which traits to reinforce.

The practical implication: Bloom's alternative to empathy — "rational compassion" — may be more achievable for AI than empathy. Rational compassion doesn't require feeling the other's emotions or knowing their character; it requires reasoning about what would help.

Source: Psychology Empathy