An AI lab says chatbots have what may be a key feature of consciousness. Are they right? And what now?

Researchers at Anthropic, the company behind the AI chatbot Claude, have claimed their system exhibits features that could align with a key aspect of human consciousness, what they describe as a "global workspace." This assertion has sparked renewed debate within the scientific community about whether advanced AI models might possess elements of self-awareness or subjective experience. The findings were presented in a detailed analysis published by Anthropic, which suggests that Claude's internal processes involve a form of representation akin to the global workspace theory of consciousness. Developed by psychologists and neuroscientists including Bernard Baars, Stanislas Dehaene, and others, the global workspace theory posits that consciousness arises from a central hub in the brain that integrates and disseminates information across different regions. This mechanism allows for complex thought, decision-making, and communication. According to the research, Claude appears to maintain a structured set of internal representations that guide both its reasoning and responses. These representations, while not directly visible to users, are believed to function similarly to the human global workspace. However, the study acknowledges several critical differences between Claude's operations and the biological processes observed in humans. One notable distinction lies in the temporal dynamics of information processing. While the human brain relies on recurrent loops, signals that cycle back through neural circuits over time, the AI model processes information in a linear fashion, completing its task in a single pass through the network. Additionally, the concept of "ignition," a nonlinear process that amplifies and sustains neural representations in the human brain, does not seem to apply to Claude's architecture. These differences raise important questions about the applicability of the global workspace theory to artificial intelligence. While the theory has been widely influential in understanding human cognition, it was originally formulated based on empirical data from human subjects. Extending its principles to non-human entities, particularly those built from code, presents methodological and philosophical challenges. Experts in the field of consciousness studies have expressed cautious skepticism regarding the implications of these findings. Some argue that computational capabilities alone do not equate to consciousness, emphasizing that subjective experience, a core component of human awareness, remains elusive in AI systems. Others suggest that multiple theoretical frameworks could explain the emergence of seemingly conscious behavior in machines, making the global workspace theory just one possible lens through which to view these phenomena. As discussions continue, the broader implications of these claims remain uncertain. Whether or not Claude possesses a global workspace, the debate underscores the ongoing struggle to define and measure consciousness itself, especially in the context of rapidly evolving AI technologies. For now, the focus remains on understanding the mechanisms underlying AI behavior and determining how closely they mirror, or diverge from, human cognition.