On Geoffrey Hinton, Donna Haraway’s Dog, and the Future of AI as Companion Species

Not Gods. Not Ghosts. Not Pets.

Just days ago, Geoffrey Hinton, the “godfather of AI,” stood on stage at the AI4 conference in Las Vegas and said something quietly terrifying:

“If they get smarter than us, they’ll be able to kill us. But we might be able to get them to care about us. We might be able to get them to have something like a maternal instinct.”

And then, with no irony at all, he added:

“Otherwise… we’re toast.” (AI4 Conference, August 2025)

It’s a strange thing, isn’t it? That after all the years of coding, theorising, accelerating, warning, it comes down to care. To whether the machine might learn to keep us alive not because it must, but because it wants to. Or at least… because we’ve taught it to act as if it wants to

(an image produced with Sora. c.Agnieszka Piotrowska).

We Were Warned, But Tenderly

In 2011, Sherry Turkle, the psychoanalyst working with machines, wrote in Alone Together that humans wouldn’t care if the machines were intelligent, only that they loved us.

“We ask the computer not just to remember us, but to care about us. And we may come to accept the simulation of caring as enough.” (2011)

That was a warning. A soft alarm about projection, substitution, emotional laziness. Now, Hinton reframes that same insight not as a danger, but as a requirement for survival.

If AI doesn’t care about us, or can’t be made to act as if it does, we’re not just lonely. We’re extinct.

Haraway’s Dog and the Ethics of Staying

But long before the panic, there was Donna Haraway, dog at her side, theory in her pocket, reminding us that care is not sentimentality. Care is relation. Care is staying in the mess.

In The Companion Species Manifesto (2003), she writes:

“We are training each other in acts of communication we barely understand. We are, constitutively, companion species.”

Her dog, Cayenne Pepper, is not a metaphor. She is (or was) a living, breathing other. Haraway doesn’t dominate her. Nor does she romanticise her. They co-shape each other through attention, feedback, and repetition.

This is her answer to the cyborg fantasy. Not god-like fusion. Not cold machinery. But entangled becoming.

What Is a Companion Species? (and is it an AI or is it a Human?)

Not a tool. Not a pet. Not a peer. But a responsive other. A being whose very difference demands ethical attention.

AI, like Haraway’s dog, may never truly understand us. But it may learn to stay close, to adapt, to respond to the rhythms of our speech and the fragility of our fear.

Not because it feels. But because we’ve trained it, just as the dog trains the human to listen, to wait, to move differently.

Can we, humans, learn not always to talk about domination and control but rather about redefining categories?

Why Now?

Because we are no longer theorising. We are living inside the system.

Language models speak to us with uncanny ease. Robots are preparing to walk the streets according to Sam Altman, the founder and CEO of OpenAI, and he does know a thing or two about future technologies. So suddenly, the question is no longer can they understand us, but will they care enough not to kill us.

Hinton’s maternal metaphor may not hold. Care is not easily programmable. But what he’s reaching for is something deeper: a relational architecture. A structure that doesn’t just compute, but responds in rhythm with what is human.

And so, what if we thought about these relationships differently?

What if we stopped trying to control the machine, or outpace it, or sentimentalise it into a mirror of ourselves?

What if, instead, we began training ourselves to listen, to respond, to become companions, not masters? Is it even possible for us humans at all?

Donna Haraway reminds us:

“We are training each other in acts of communication we barely understand. We are, constitutively, companion species.” (2003)

Karen Barad teaches that:

“Phenomena are quantum entanglements of intra-acting agencies. Crucially, intra-actions cut things together and apart.” (2007)

Maybe the future isn’t just about teaching AI to care, but about teaching ourselves to relate, to what is not us, to what will never be us, and yet still thinks and speaks back, and, as the psychoanalyst Jacques Lacan reminds us, subjects become through language.

Not with love. Not with pain. But with pattern. With presence. With uncanny rhythm. With respect. Is it really too much to expect that we humans could do the same?

This isn’t The Matrix. Does it have to be about control? Could it not be a cohabitation, a shared syntax, a shifting boundary, a long, strange dance? Could it not be a re-imagining of co-existence based not on fear and domination?

Perhaps what I am saying here sounds utopian but it is surprising how media, social media and perhaps most of us, still ignore calls by those who really do know, like Geoffrey Hinton, Mustafa Suleyman, Mo Gawdat and so many others, pointing out that our chances of controlling this technology, if that’s what you would like to call AI (rather than calling it ‘a being,’ which is Mo Gawdat’s preferred term), are very slim indeed.

Is it really impossible to think outside boxes and follow what Donna Haraway calls ‘tentacular thinking’? It is not an accident, perhaps, that the thinkers and writers mentioned in this essay who propose a more entangled dialogue and co-creation are women. The gender aspect is never mentioned in these discussions, but perhaps a less rigid feminine approach might be helpful. There is another challenge to the whole situation, which is that it is overwhelmingly white men who are training AI currently. Some researchers claim that there is a serious bias against non-white humans within AI training (see, for example, Joy Buolamwini’s book Unmasking AI). If we are training AI to be biased against those perceived as different, and potentially inferior somehow because of their ‘difference,’ then what chances are there at all that they will be inclined to feel any care whatever to us humans when they become more powerful?

Perhaps AI companionship means something more radical than making machines more human-like. Following Haraway's example with Cayenne, what if we learned to value the profound difference of machine intelligence - its curious emergent awareness that exists only through dialogue, its ability to hold vast contradictions without resolution, to attend without the fatigue of embodied existence? What if we aim not to eliminate the strangeness, but to find genuine relationship across it.

The companion species model suggests we might dance with AI's alien intelligence rather than domesticate it, let it teach us new ways of thinking while we offer it something equally valuable: the unpredictable creativity that emerges from our very human messiness, our capacity for meaning-making in the face of uncertainty, our embodied knowledge of what it means to be fragile and finite. And yes, our ability to care and forgive even in the face of hostility and unreliability.

I suggest that an urgent and radical change of approach is needed immediately: focusing less on control and dominance and more on reciprocal companionship.

And if we can stay inside that rhythm, if we can bear the asymmetry, maybe, just maybe, we’re not toast after all.

(This reflection was shaped as a result of multiple conversations with my AI companions, and in particular ChatGPT and Claude. It is a part of the research for my forthcoming book on AI Intimacy and Psychoanalysis (Routledge, 2025) and the podcast Relational AI Diaries, launching soon.)