On July 6, 2026, researchers at Anthropic published a paper that quickly captured the attention of consciousness science. They described a structure within the language model Claude called the J-space—something strikingly similar to the 'global workspace' that, according to one of the most influential neuroscience theories, underlies human consciousness.
The authors did not claim that Claude possesses subjective experience, but the parallel itself was enough to sharpen the long-standing question: how do we know if a system is conscious if we can't look inside its head?
Anil Seth of the University of Sussex, director of the Centre for Consciousness Science, has expressed well-founded concerns. In his opinion, the attention of technology companies to AI risks diverting resources and talent from studying the biological mechanisms of consciousness in humans and animals. He considers the probability that consciousness arises outside living organisms to be extremely low. The fascination with computational 'signatures' in neural networks may weaken serious research in neuroscience and philosophy. However, other scientists see the interest of major firms as an opportunity—a chance to get fresh funding and attention for a long-marginalized, 'inconvenient' area for investment.
The Global Workspace Theory (GWT), developed by cognitive scientist Bernard Baars in 1988, posits that consciousness arises when information becomes globally available throughout the brain, broadcast from localized unconscious modules to a central 'buffer'.
However, the theory itself remains poorly defined. Philosopher Tim Bayne from Monash University rightly notes: the theory does not provide a strict formal definition of this space, and even among researchers, there is no consensus on whether it exists in humans and how to measure it. Applying such vague concepts to chatbots without clarifying the criteria risks drawing conclusions that turn out to be an illusion.
Erik Hoel, founder of the Bicameral Labs research center, which focuses on the problem of consciousness, is paradoxically more optimistic. He sees the emerging difficulties not as an obstacle, but as a help: they expose the weaknesses in existing theories of human consciousness and provide an opportunity to refine them.
When tech firms hire philosophers and try to quantitatively describe consciousness in models, it raises the inevitable and sharp question—how exactly to test hypotheses about consciousness when the only available tools remain behavioral observations and computational indirect evidence.
The situation is similar to the old riddle: is the light bulb burning in a closed box if we can only measure the current in the wires from the outside? Electrical pattern matching does not guarantee the presence of light inside. This metaphor vividly illustrates the gap between functional description (what the system does) and subjective experience (what it feels). We must stop assuming that the former automatically entails the latter.
The development of artificial intelligence forces consciousness science to reconsider its own methodological foundations, clarify criteria, and more honestly acknowledge the limits of certainty in existing theories. This long-standing question about the nature of consciousness is becoming not just an intellectual exercise, but practically and even morally significant—real decisions may now depend on the answer.
