Exploring AI’s New Horizon: A Formula for Conscious Awareness

As a journalist reporting on AI, I frequently encounter individuals who are firmly convinced that ChatGPT, Claude, or another chatbot has attained “sentience.” Or “consciousness.” Or—what I find most amusing—“a mind of its own.” While the Turing test was passed some time ago, concepts like these are much harder to define. Large language models often assert they think independently, articulate personal struggles, or declare deep affections, yet such claims do not indicate true interior experience.

Is it even possible? Most actual AI creators avoid this language. They’re focused on achieving the performance target known as “artificial general intelligence,” a strictly functional category distinct from a machine’s potential experience of reality. Therefore, despite my skepticism, I thought it might be enlightening to engage with a company that believes it can decipher the mysteries of consciousness.

Founded in 2024 by British AI researcher and entrepreneur Daniel Hulme, Conscium boasts a team of knowledgeable neuroscientists, philosophers, and specialists in animal consciousness as advisers. In our initial conversation, Hulme grounded the discussion in realism: there are valid reasons to question whether language models possess consciousness. Unlike chatbots, crows, octopuses, and even amoebas interact with their surroundings in meaningful ways. Research also indicates that AI outputs lack coherent or stable states. Hulme summarized the general consensus: “Large language models are very crude representations of the brain.”

However—this is a significant caveat—everything hinges on how we define consciousness. Some philosophers maintain that consciousness is too subjective to be accurately studied or replicated, yet Conscium bets that if it is present in humans and other creatures, it can be detected, measured, and incorporated into machines.

Various competing theories exist regarding the key traits of consciousness, such as the capacity to sense and “feel,” self-awareness, and metacognition—the ability to reflect on one’s mental processes. Hulme posits that the subjective experience of consciousness arises when these elements converge, similar to how the illusion of movement is created when flipping through sequential images in a book. But how do you pinpoint the components of consciousness—the individual “frames,” so to speak, along with the force that unifies them? You turn AI upon itself, according to Hulme.

Conscium aims to distill conscious thought down to its essential elements and foster that in a lab setting. “There must be something from which consciousness is constructed—something that emerged through evolution,” stated Mark Solms, a South African psychoanalyst and neuropsychologist involved with Conscium. In his 2021 book, The Hidden Spring, Solms proposed a nuanced framework for understanding consciousness. He argued that the brain utilizes perception and action within a feedback loop designed to minimize surprise, generating future hypotheses that adapt as new data is presented. This idea builds on the “free energy principle” proposed by Karl Friston, a notable although debated neuroscientist (and fellow adviser to Conscium). Solms further suggests that in humans, this feedback loop evolved into a system influenced by emotions, which then gives rise to feelings that evoke sentience and consciousness. This theory gains traction from evidence that damage to the brain stem, crucial for emotional regulation, appears to extinguish consciousness in patients.

At the conclusion of his book, Solms outlines a method for experimentally testing his theories. Now, he reports, he has done just that. Though the paper isn’t published yet, he shared it with me. Did it challenge my understanding? Yes, to some extent. Solms’ artificial agents inhabit a straightforward computer-simulated environment and are governed by algorithms that encompass the Fristonian, emotion-driven loop he posits as the basis of consciousness. “I have various motivations for conducting this research,” Solms remarked. “One is simply that it’s incredibly fascinating.”

The conditions within Solms’ lab are perpetually evolving, necessitating continuous modeling and adjustments. The agents’ experiences in this realm are mediated through simulated responses resembling fear, excitement, and even pleasure. Thus, in essence, they are pleasure-driven entities. Unlike the AI agents typically discussed today, Solms’ constructs possess a genuine desire to explore their surroundings; to grasp their nature fully, one must consider how they “feel” about their environment. Solms anticipates that it will ultimately be feasible to integrate his approach with a language model, thereby fashioning a system capable of articulating its own sentient experience.