As a journalist covering the AI beat, I hear from countless people who are fully convinced that ChatGPT, Claude, or another leading chatbot has already achieved “sentience.” Or “consciousness.” Or—my personal favorite—“a mind of its own.” It is true that large language models aced the Turing test long ago, but unlike measurable, rote intelligence, concepts of inner experience are far harder to pin down. LLMs will readily claim to think independently, even describe personal inner turmoil or declare deep affection for the people chatting with them, but that output does not prove the existence of a subjective inner world.

Could that ever change? Most AI developers don’t frame their work in these terms at all. They are far too busy chasing the performance benchmark called artificial general intelligence, a purely functional classification that has nothing to do with whether a machine can experience the world from its own perspective. So even as a committed skeptic, I thought it would be eye-opening, maybe even illuminating, to connect with a company that believes it can crack the code of consciousness itself.

Founded in 2024 by British AI researcher and entrepreneur Daniel Hulme, Conscium boasts an advisory roster that includes a who’s who of neuroscientists, philosophers, and animal consciousness experts. When we first spoke, Hulme was refreshingly realistic: there are strong reasons to doubt that current large language models are capable of consciousness. Crows, octopuses, even amoebas interact with their environments in ways that no existing chatbot can match. Studies also confirm that AI-generated statements do not reflect coherent, consistent internal states. Echoing the widespread scientific consensus, Hulme put it plainly: “Large language models are very crude approximations of the human brain.”

But there is a big catch here: everything depends on how we define consciousness in the first place. Some philosophers argue consciousness is too inherently subjective to ever be studied or recreated artificially, but Conscium operates on the core bet that if it can emerge naturally in humans and other animals, it can also be detected, measured, and built into machines.

There are dozens of competing, overlapping frameworks for the core characteristics of consciousness, including the ability to sense and “feel,” awareness of self and surroundings, and metacognition (the ability to reflect on your own thought processes). Hulme believes the subjective experience of consciousness emerges when all these traits combine, much like the illusion of movement created when you flip through sequential pages in a flipbook. But how do you identify the building blocks of consciousness—the individual frames, so to speak, plus the force that binds them together? Hulme’s answer: turn AI back on itself to answer the question.

Conscium’s goal is to break conscious thought down to its most basic form and cultivate it in a lab setting. “There must be a core set of components that consciousness is built from—the same components that allowed it to emerge through evolution,” explained Mark Solms, a South African psychoanalyst and neuropsychologist leading work on the Conscium project. In his 2021 book The Hidden Spring, Solms proposed a provocative, feeling-centered new framework for consciousness. He argues the brain runs a feedback loop of perception and action designed to minimize surprise, constantly generating hypotheses about the future that get updated as new information comes in. The idea builds on the “free energy principle” developed by Karl Friston, another prominent (if controversial) neuroscientist who also serves as a Conscium advisor. Solms extends this work to argue that in humans, this feedback loop evolved to be mediated by emotion, and it is these feelings that give rise to sentience and consciousness. The theory is backed by empirical evidence: damage to the brain stem, which plays a critical role in regulating emotion, almost always eliminates consciousness in human patients.

At the end of his book, Solms proposed a way to test his theories in a lab setting. Today, he says he has done exactly that. He has not published his work yet, but he shared his draft paper with me. Did it upend my thinking? A little, to be honest. Solms’ artificial agents exist in a simple computer-simulated environment, controlled by algorithms that use the exact Fristonian, feeling-mediated loop he calls the foundation of consciousness. “I have a few motives for doing this research,” Solms told me. “One is just that it’s fucking interesting.”

Solms’ simulated environment is constantly changing, requiring agents to continuously update their internal models and adjust their behavior. The agents’ experience of this world is filtered through simulated responses analogous to fear, excitement, and even pleasure. In short, they are pleasure-driven bots. Unlike the AI systems everyone discusses today, Solms’ creations have a literal, built-in desire to explore their environment; and to understand them properly, you have to try to imagine how they “feel” about their small simulated world. Solms believes it will eventually be possible to merge his team’s approach with a large language model, creating a system that can talk about its own sentient experience.

At this stage, Conscium’s work is so early that it barely exists beyond conceptual work. It is a faint glimmer of an idea that is very probably impossible to pull off. But it is still undeniably fucking interesting, and beyond that, it pushed me to rethink my own understanding of consciousness. If you’ll indulge a little metacognition: I have long assumed that thought is what makes me conscious, not emotion. But what if I—what if all of us—are looking for consciousness in all the wrong places? What would it mean if consciousness really can be reduced to such simple mechanisms? Maybe the people who claim to have glimpsed sentience in ChatGPT are not hallucinating after all.