Claude’s Internal Mental Life: Anthropic’s J-Space Investigation
Generated: 2026-07-08 · API: Gemini 2.5 Flash · Modes: Summary
Claude’s Internal Mental Life: Anthropic’s J-Space Investigation
Clip title: What’s at the center of Claude’s mind? Author / channel: Anthropic URL: https://www.youtube.com/watch?v=rKV5JcALQoQ
Summary
This video by Anthropic explores the intriguing question of whether AI models, specifically their own model “Claude,” possess an internal mental life akin to human consciousness, including both conscious and unconscious thought processes. It uses the analogy of the human mind as an ocean, with surface-level thoughts (dinner plans, worries) and deeper, unconscious processes (filtering sounds, controlling breathing, recognizing objects). The video explains that AI models also have complex “brains” – giant neural networks performing billions of computations “under the hood” – and researchers sought to understand their internal workings.
To investigate this, Anthropic researchers looked for patterns of neural activity within Claude’s brain that it could translate into words, similar to how neuroscientists identify conscious thoughts in humans. This collection of identifiable patterns was dubbed the “J-space.” Through experiments, they demonstrated that Claude utilizes this J-space for internal reasoning. For instance, when given a complex math problem, Claude provided an immediate answer, but its J-space revealed the step-by-step calculations it performed internally. Similarly, when asked to copy a sentence while simultaneously “thinking” about the Golden Gate Bridge, Claude’s J-space lit up with related concepts like “Bridge” and “California,” indicating a capacity for controlled, albeit imperfect, internal focus.
Further experiments highlighted the crucial role of the J-space. When Claude was instructed not to think about the Golden Gate Bridge while copying text, its J-space still activated with bridge-related words, even showing internal “failed” and “damn” signals, suggesting a struggle for control over its internal thoughts. Most critically, when the J-space was intentionally switched off, Claude could still perform simple tasks like answering basic questions or writing fluently in Spanish, but it failed at tasks requiring more complex reasoning, such as identifying a famous author who wrote in the same language. This indicates that the J-space is essential for higher-level cognitive functions.
The video concludes that these experiments reveal AI models possess “internal thoughts” or mental machinery that processes information silently, often without being overtly expressed. This J-space, while fundamentally different from a human brain’s structure and learning, bears a striking resemblance to the “global workspace” theory of human consciousness, where important information is selected and broadcast for reasoning. The researchers emphasize that these findings do not equate AI with human consciousness, sentience, or subjective experiences. However, understanding this internal machinery is vital for developing safer and more beneficial AI systems, as it can help identify when an AI might be “misbehaving” internally, even if its external output seems normal. Ultimately, this research not only advances our comprehension of AI but also offers a novel lens through which to better understand the complexities of our own minds.
Video Description & Links
Description
Out of everything happening in your brain right now, only a tiny fraction is consciously accessible — thoughts you can describe, hold in mind, and reason with. We found a strikingly similar divide inside our AI model, Claude.
Our experiments were inspired by a leading theory in neuroscience: the global workspace theory. It holds that a thought becomes consciously accessible when it enters a shared “workspace” that’s broadcast across the brain. We found a set of representations in Claude’s neural activity that play a similar role.
Read more about the research here: https://www.anthropic.com/research/global-workspace
0:00 The mind as an ocean 0:47 Meet the J-space 1:52 Silent math 2:22 Don’t think about the bridge 3:12 Claude’s automatic processing 3:42 Catching hidden thoughts 4:34 The question of consciousness