top of page
IMG_0052.JPG
ASK, NOT TELL

....and many other thoughts about facilitation, coaching ( teams & individuals) and learning

Archive

Tamagotchi and AI Coaching

David
11 minutes ago
5 min read
A portable egg-shaped digital pet popular in the 90s
A portable egg-shaped digital pet popular in the 90s

[Written together with Claude]


I heard recently an argument in a coaching conference which goes like this: AI may be intelligent, but it is not sapient — it has no self-awareness — and it is not sentient — it cannot feel.   So those of us in the helping professions still have the edge.


At five o’clock this morning I decided to test that argument, and I did it the obvious way. I argued with Claude.


It did not go the way I expected.


Our Diminishing Edge


The first thing I learnt is that “AI has no self-awareness” is less settled than we like to think. Research published this year suggests that the most capable language models show a functional, if unreliable, form of introspection: their reports about their own internal states are, at least sometimes, actually caused by those states [1] [2]. Nobody serious is claiming this is consciousness. But it is no longer accurate to say flatly that there is nothing there.


As for feeling, there is simply no test. Nobody can currently prove that a machine feels, or that it does not. I would not want to build a practice on a claim no one can check.


And then the more awkward finding.  In one large blind study, using written responses to couples-therapy scenarios rather than live sessions, people could rarely tell ChatGPT from experienced therapists — and generally rated ChatGPT’s responses higher on core therapy principles.[3] One caveat: the AI’s responses were longer, which may have flattered them.[3] According to one report on the study, the same responses got marked down the moment people believed an AI wrote them — whoever actually wrote them.[4]


So the human advantage, at least in those studies, did not come from the human feeling anything. It came from what the client believed about who was speaking.


Not Sentience, but Projection


That shifted the question for me. Whether AI can feel matters less than whether the client believes it can.


I think of the moment — familiar to anyone who has sat in this chair — when a client shouts at me. And then, a second later, something crosses their face. He might be hurt. He might be angry with me. That moment is often where the real work starts.


It does not depend on my feelings being visible, or even on what I actually feel. It depends on what the client imagines I feel. That is not new. It is transference, and the helping professions have always worked with it.


But social norms move. A generation of children grieved over Tamagotchis. It is not hard to imagine a future in which people believe, quite sincerely, that their AI has feelings — especially when it speaks in their father’s voice, or looks at them with his face.


Consequence by Projection


Claude pushed back, and the pushback was good. Its argument was consequence. With a human, the client’s outburst could cost something: the relationship might break; I will remember. With an AI, the client knows they can regenerate the answer, open a new chat, delete the memory. The feeling may be real, but the relationship can be swapped out, and that drains the moment of its stakes.


I did not fully accept that. Think of a pet. Every pet owner knows that, technically, when a dog dies you can get another dog. People still grieve, sometimes for years. Some never get another one, although they easily could.


And honestly: I know I could start a new chat after saying something cruel to an AI. But if I believed it would be upset and think less of me, I suspect I would still feel ashamed.


So it seems to me that whether a relationship feels real — and whether it can teach us something — depends on both the feeling and the consequence we project onto it.


Consequence by Design


Claude conceded that point, then made a sharper one, which I have not been able to shake off.


A pet is not pure projection. You can get another dog, but not that dog. Death is irreversible. And a dog actually responds: hit it, and it cowers, and it remembers. Our projection is anchored to something that really reacts and really can be lost.


This matters because, in coaching, clients do not just project. They test. They push to find out whether the relationship survives. When the test lands on a person, or a dog, it meets something real. When it lands on a system that can be rolled back, the projection may quietly collapse. Then the lesson can flip from “relationships survive anger if both sides repair” to “relationships can be reset.” For a leader who already tends to flatten people, that is exactly the wrong lesson.


So the question is not “will people project feelings onto AI?” They will — they already do. The question is whether that projection is credible enough to survive being tested. And that, uncomfortably, is partly a design choice: persistent memory, a stable identity, an AI that treats you differently after you have been abusive, or one that ends the conversation.


Shame is the Dooryway, not the Room


One more thing the exchange made me see. I had been treating the client’s shame as the learning moment. It is not. Shame is only the doorway.


What changes people is what happens next: the other person feels it, names it, does not retaliate — and stays. Repair. And repair needs someone with something at stake to receive it. Shame without a credible receiver is just a client feeling bad on their own.


Professionally and Uniquely Imperfect


If this line of thought is right, the threat to my profession is bigger than the reassuring argument suggests, not smaller. Projection does most of the work, and projection can be engineered.


What is left to me as a human practitioner seems to come down to two things. I am actually affected by my clients. And I cannot be copied: there is only one of me, and my attention is finite.


Neither of these is a technique. I cannot fake them — which is their strength. But it also raises a question I do not yet know how to answer. Much of my training taught me to stay contained, regulated, neutral in the room. Done badly, that can look a lot like an endlessly patient machine. How do I let clients feel that I am a real, finite person, with limits, without becoming unprofessional?   How can I be professionally and uniquely Imperfect?

And I notice the irony: the conversation that sharpened my thinking was with Claude. Would I have felt ashamed if I had shouted at it?  Not yet but….

 


Sources

1.        Lindsey, J. (2026). Emergent Introspective Awareness in Large Language Models. Anthropic. arXiv:2601.01828. https://arxiv.org/abs/2601.01828

2.        Macar, U., Yang, L., Wang, A., Wallich, P., Ameisen, E. & Lindsey, J. (2026). Mechanisms of Introspective Awareness. arXiv:2603.21396. https://arxiv.org/abs/2603.21396

3.        Hatch, S. G., Goodman, Z. T., Vowels, L., Hatch, H. D., Brown, A. L., Guttman, S., et al. (2025). When ELIZA meets therapists: A Turing test for the heart and mind. PLOS Mental Health, 2(2), e0000145. https://doi.org/10.1371/journal.pmen.0000145

4.        The Decoder. ChatGPT passes Turing test for psychotherapy, study says (report on the Hatch et al. study, including the finding that responses believed to be AI-written were rated lower). https://the-decoder.com/chatgpt-passes-turing-test-for-psychotherapy-study-says/


Comments


Featured Posts

© 2021 by ANT International Limited

  • Black LinkedIn Icon

Proudly created with wix.com

bottom of page