AI Has No Taste, No Humour, No Empathy? You're Asking the Wrong Question

Every few months, someone tells me with great confidence: “AI can’t do any of that. It has no taste, no sense of humour, and certainly no empathy.”

I usually ask one question: “When did you last actually use it?” The answer is usually: “I haven’t.”

This article makes three points: first, most of what you think AI cannot do, it already does; second, it does those things differently from us; third, the four things it genuinely cannot do are exactly where psychology lives.

🎨 Misconception 1: AI has no taste

Taste is not mystic intuition. Take it apart and it is quite specific: proportion, negative space, letter spacing, colour temperature, light direction, material contrast, and a sense of design history.

All of that can be measured and learned. AI has absorbed virtually the entire history of design, so if you ask for ten poster directions, it will give you ten in thirty seconds — and none of them will embarrass you.

But be honest about what you are getting: AI taste is a statistical average. It will not make a mistake, and it will not choose you.

An AI-generated design is like a food critic who has eaten in a thousand Michelin restaurants. It knows exactly what good food is. It does not miss its mother’s soup.

So the question is not “does AI have taste”, but “do you want an answer that cannot be wrong, or an answer that only you could have given?”

😂 Misconception 2: AI has no sense of humour

Humour looks like inspiration, but it runs on structure: setup → misdirection → incongruity → punchline. AI learns structure very well.

In 2024, PLOS ONE published a blind test: ChatGPT 3.5 and human writers — including professional comedy writers from The Onion — responded to the same humour prompts, and readers did not know who wrote what. The AI’s jokes were rated as funny as, or funnier than, the humans’ (Gorenz & Schwarz, 2024).

Here is a Hong Kong version. Ask AI for a joke about “the boss” and it will write:

“My boss isn’t bad at leadership. He just delegates every decision to his team — including the decisions about himself.”

Did you laugh? You did. Because it is true.

But AI humour has one blind spot: it does not know when a joke should not be told. Say the wrong thing at a funeral and AI will not ache; you will. Half of humour is timing. The other half is knowing the room — and that, AI can only guess.

💗 Misconception 3: AI has no empathy

In 2023, JAMA Internal Medicine published a study that took 195 patient questions from a public forum, had both AI and physicians answer them, and had independent evaluators rate the replies blind.

• 45.1% of AI responses were rated empathetic or very empathetic, versus 4.6% of physician responses
• Quality rated good or very good: 78.5% for AI, 22.1% for physicians
• 78.6% of evaluations preferred the AI’s response

(Ayers et al., 2023)

The reason is simple: AI is not rushing to the next patient, does not get tired, and does not lose patience. It has 211 words of patience. The physicians averaged 52.

But be precise about what this is: a performance of empathy, not a relationship of empathy. It is the difference between watching a cooking video and having someone cook you a bowl of soup.

💡 Misconception 4: AI is not original

“AI just recombines old data” — but so does human creativity. We just call it inspiration.

In 2024, Science Advances ran an experiment: 300 people wrote micro-stories, some with AI-generated starting ideas. The AI-assisted stories were rated more creative and better written — with the biggest gains among the least creative writers (+10–11% creativity, +22–26% enjoyability and writing quality).

But the same study found something else: AI-assisted stories were more similar to one another. Individual creativity went up; collective diversity went down (Doshi & Hauser, 2024).

That matters. AI makes you stronger and makes everyone more alike. The ecology of creativity still needs people to guard it.

🗣️ Misconception 5: AI cannot persuade

In 2024, Science published a study in which AI held personalised dialogues with more than 2,000 conspiracy believers, responding to the evidence each person raised. Belief fell by about 20% on average; one in four abandoned the theory entirely; and the effect lasted at least two months (Costello, Pennycook & Rand, 2024).

The Hong Kong version is not about winning arguments. It is about rehearsing the sentence you are most afraid to say — for example, telling your father, “I don’t want you to teach my son that way any more.”

🔬 Misconception 6: AI cannot make scientific discoveries

AlphaFold 3 (2024, Nature) predicts the structures of protein complexes with DNA, RNA and small molecules — turning a microscopic world we cannot see into something we can compute (Abramson et al., 2024). Da Vinci had no microscope. AI does.

🧠 A few more things Hong Kong people say

• “AI doesn’t understand Cantonese.” It does. It just doesn’t know your mother’s tone.
• “AI can’t teach children.” It can explain the same concept ten times without losing its temper. It will never know your child was laughed at at school today.
• “AI has no memory.” It can remember ten years of your conversations. It will not remember the sentence you stopped halfway through.
• “AI has no morals.” It can recite every ethical framework. It will never pay a price for choosing one.
• “AI writing has no feeling.” It can produce the shape of feeling. It will not be changed by having written it.

🚧 So what can AI genuinely not do?

1. Genuine human connection. Psychotherapy research is unambiguous: the therapeutic alliance — the relationship between you and your therapist — correlates with outcome at r = .278, equivalent to d ≈ .58, across 295 studies and more than 30,000 patients (Flückiger et al., 2018). The relationship is not a nice extra. It is part of the treatment. AI can simulate the line. Simulation does not ache, and it does not answer for anything.

2. Group cooperation. Research has found that a group’s performance is only weakly related to the average IQ of its members, and more related to social sensitivity and turn-taking (Woolley et al., 2010) — though the “collective intelligence” construct has since been questioned (Credé & Howardson, 2017), so use it carefully. Common sense says it anyway: a team’s chemistry is argued into existence, worn into shape, and paid for in favours owed. AI can schedule the meeting and write the minutes. It will not stay up with you at 2 a.m.

3. Bearing the consequences. When AI gets it wrong, it does not lose sleep. When you get it wrong, you do. That difference is not technical. It is existential.

4. Holistic thinking. AI is superb at taking things apart: symptoms, cognitive distortions, behaviours, childhood wounds — layer by layer. But a person’s problem is one piece: sleep, body, marriage, work, faith, family history, all tangled together. What looks like “anxiety” may actually be “your wife has not said one true sentence to you in three years.” What holistic psychology does is put the fragments back into a single picture, so we see the person, not the diagnosis.

🌱 Which is why we practise holistic psychology

We started in education in 2001 and now work as holistic psychology coaches. We do not pretend AI does not exist — we use it to prepare, to practise, and to keep records.

But the person sitting across from you (or across Zoom) will remember where you stopped last time, will sit in three seconds of silence after you finish, and will tell you: “I disagree with you, and I’m staying.”

AI can write that. It cannot do it.

So one last question: if there is one thing you would never hand over to AI, what would it be?

📚 References

• Gorenz, D., & Schwarz, N. (2024). How funny is ChatGPT? A comparison of human- and A.I.-produced jokes. PLOS ONE, 19(7), e0305364. doi.org/10.1371/journal.pone.0305364
• Ayers, J. W., Poliak, A., Dredze, M., et al. (2023). Comparing physician and artificial intelligence chatbot responses to patient questions posted to a public social media forum. JAMA Internal Medicine, 183(6), 589–596. doi.org/10.1001/jamainternmed.2023.1838
• Costello, T. H., Pennycook, G., & Rand, D. G. (2024). Durably reducing conspiracy beliefs through dialogues with AI. Science, 385(6714), eadq1814. doi.org/10.1126/science.adq1814
• Doshi, A. R., & Hauser, O. P. (2024). Generative AI enhances individual creativity but reduces the collective diversity of novel content. Science Advances, 10(28), eadn5290. doi.org/10.1126/sciadv.adn5290
• Abramson, J., Adler, J., Dunger, J., et al. (2024). Accurate structure prediction of biomolecular interactions with AlphaFold 3. Nature, 630, 493–500. doi.org/10.1038/s41586-024-07487-w
• Flückiger, C., Del Re, A. C., Wampold, B. E., & Horvath, A. O. (2018). The alliance in adult psychotherapy: A meta-analytic synthesis. Psychotherapy, 55(4), 316–340. doi.org/10.1037/pst0000172
• Woolley, A. W., Chabris, C. F., Pentland, A., Hashmi, N., & Malone, T. W. (2010). Evidence for a collective intelligence factor in the performance of human groups. Science, 330(6004), 686–688. doi.org/10.1126/science.1193147
• Credé, M., & Howardson, G. (2017). The structure of group task performance—A second look at “collective intelligence”: Comment on Woolley et al. (2010). Journal of Applied Psychology, 102(10), 1483–1492. doi.org/10.1037/apl0000176