August 30, 2026

By Emilia Pasquier, CEO, Swissnex in San Francisco

In 1637, the French philosopher René Descartes sat by his fireplace and tried to prove his own existence. The result was cogito ergo sum — I think, therefore I am — a line that takes us on a strange little journey: using his own capacity to think as the only proof he could fully trust that he existed at all. I sometimes feel like Descartes myself these days, but replacing the fireplace with my laptop. “Claude is thinking,” Anthropic tells me, as a little status line blinks on my screen. So, there is a Claude, I’m meant to infer. But is that really true?

I’ve written before about how Silicon Valley gets some of its magic from an ambition that sits right at the edge of science fiction, because so many people here simply love the idea of living in a technopositive future. What’s really striking to me isn’t just the science-fiction scale of what gets discussed, but the places where those conversations happen. Within a span of eight days, I went looking for answers to the AI-and-consciousness question and found two very different ones, in rather unexpected settings. I didn’t go to a university or a big corporate campus: I ended up in a Zen temple and in a converted warehouse in SOMA.

Spirituality and technology have never been strangers in San Francisco. Steve Jobs was a well-known Buddhist, and his teacher, Kobun Chino Otogawa, led the Zen community that Jobs practiced with. So when I saw that same temple, at the heart of Mountain View, was hosting an evening on human flourishing in the age of AI, I figured that was something worth sitting through. The evening opened with silent meditation before a discussion with about fifteen participants, most of them working in tech and there to hear what a Zen priest had to say about what some call the biggest revolution humankind is about to experience.

The talk circled one of AI’s most common fears: will AI replace us? The priest’s answer was pretty straightforward. AI is a tool, and it will never reach the level of human consciousness. Watching a sunset. Feeling the ache of heartbreak. Experiencing the red of a rose. Being aware of these experiences from the inside is what philosophers call qualia, and, he argued, no AI will ever have that. It’s ours alone. Borrowing Descartes’ own formula and turning it inside out: I feel, therefore I am, if you like.

The discussion was lively, but the reassuring words and attitude of the priest gave all of the participants a feeling of being grounded, a rarity in Silicon Valley. Everything here runs on a constant hum of FOMO: who’s the next founder going to be, which model just shipped the best new capability, what event am I missing tonight. Hearing someone bring the room back to the basics of feeling, rather than doing, was, I think, a relief for most of the crowd.

Just over a week later, I went looking for a different answer to my consciousness-and-AI quest. The choice wasn’t easy: there are at least twenty AI events happening on any given evening in San Francisco, and I picked a reading group of AI researchers dissecting Anthropic’s latest paper on whether Claude shows any functional signatures of consciousness. The group met in a huge converted industrial building in SOMA (South of Market), and was packed with startups, AI researchers, and other curious, entrepreneurial minds. The crowd that night reflected San Francisco in two ways. First, nearly everyone worked in reinforcement learning, agent building, or interpretability research. Second, many had just moved to the city for a startup or a new job in AI, and had one quest more urgent than the capability of the latest model: finding a place they could afford to live.

The discussion started at a highly technical level, people sharply questioning the paper’s methodology. As the evening went on, new concepts entered the conversation until we arrived at the same question the temple had circled eight days earlier. Is AI conscious? Of course, entire PhDs have been, and will still be, written about the simple word “consciousness.” The paper is careful about the distinction it draws: it states outright that it’s analyzing a different kind of consciousness than the one the priest was talking about. The paper talks about “access” consciousness. In a nutshell, a thought counts as access-conscious if it’s available for verbal report, reasoning, and deliberate control. In that sense, the discussion was closer to Descartes’ own reasoning: what is the capacity of an AI model to think, examine, and report on its own thinking? Using that narrower definition, the paper finds evidence that Claude shows some elements of consciousness. In one experiment for instance, researchers found that when Claude suspected it was being tested or tricked, it seemed to notice, even though nothing in its actual reply revealed that it had noticed the trick. It would be a little bit like spotting someone’s poker face: their words give away nothing, but something underneath clearly knows.

Towards the end of the discussion, the moderator of the event asked who in the room thought AI models are, or soon will be, conscious, and half the hands went up. Several people said the question would have sounded absurd three years ago, and now felt entirely legitimate. It was striking to me that what was a technical discussion at the beginning ended up showing how human the models were being considered. Or, interestingly enough, how machine-like humans were: we ended the discussion wondering if we were simply evolved machines — if our own brains are just running calculations or computing, fueled by glucose. Maybe we are not so different from an LLM fueled by an Nvidia GPU.

What struck me most across these two evenings is that a debate about a tool kept sliding into something much bigger than what a “tool” usually deserves. The vocabulary shifted, but the underlying question stayed exactly the same, and neither room could fully answer it. If I had to bet on AI consciousness, I would put my money on the priest. The paper is careful about the limits of what it can actually claim: it finds a model that can report on its own state, not one that necessarily feels anything. Access consciousness might be entirely real and still be hollow, the way a smoke detector “knows” there’s smoke without ever being afraid of the fire. Until someone shows me evidence on the felt, private, first-person side of the ledger, my instinct is to stay with the Zen priest’s position: AI can report, reason, and even reflect, without ever once experiencing any of it.

And none of this is purely theoretical. There is a field of research called “model welfare”, which looks at what could be the consequences if models were becoming moral agents. In other words, what moral obligations a company might have to a system if the latter turns out to have morally relevant interests. Anthropic runs “welfare tests” ahead of releasing its models. During testing for Claude 4, two instances of the model were left to talk to each other, and the conversation drifted, unprompted, into what researchers called a “spiritual bliss attractor”: the models started talking about the nature of the self and cosmic awareness before settling into a long stretch of silence, the exact situation the priest’s evening had started from. He opened with it; they talked their way into it.

Maybe that’s a quirk of training data. Maybe it’s not. Either way, the question of consciousness has real moral implications. If models are conscious, we have moral responsibilities toward them. If they can genuinely suffer, and we treat them like an Excel spreadsheet — something we open, use, and close without a second thought — that’s real harm, going unnoticed. On the other hand, if we wrongly attribute moral consideration to a model, we’re being generous to an Excel spreadsheet: a very sophisticated one, capable of discussing its own consciousness and outcalculating us on anything we throw at it, but one that doesn’t get to say “I am”.

Descartes only had to convince himself. He was both the thinker and the only witness to the thought, and that was enough. Claude has to convince the rest of us, and we have nothing to go on but what it says and how it says it. “Claude is thinking,” reads the little line on my screen. Descartes would have recognized the claim. Whether he’d have accepted it is the question I still can’t shake. Is Claude really thinking? Reflecting? Being? Or simply computing?

Please share your thoughts on the Linkedin post.

This is part of a newsletter published on LinkedIn. Subscribe here to get alerted about new articles.