Source : INDIA TODAY NEWS
Anthropic says it has discovered something unusual inside its Claude AI models — a hidden internal “thinking space” where ideas can be processed without appearing in the chatbot’s visible reasoning. While the company is not claiming that Claude is conscious, the finding is likely to add fresh fuel to the ongoing debate about how human-like advanced AI systems are becoming.
The AI startup has named this internal workspace “J-Space,” a reference to the Jacobian mathematical technique researchers used to identify it. According to Anthropic, J-Space is different from the chain of thought or reasoning that users sometimes see. Instead, it acts as a private workspace where Claude can activate concepts, plan strategies and process information silently before generating a response. In a video released alongside the research, Anthropic said that Claude can carry out reasoning internally, such as identifying bugs in code or recognising images, without explicitly expressing every step.
advertisement
“We can see Claude silently perform reasoning steps in its head—noticing bugs in code, identifying images, and more,” Anthropic said in a post on X.
One of the more surprising observations is that Claude’s hidden workspace can focus on ideas that are unrelated to the task it is visibly performing. Anthropic compared this to the way humans can think about one subject while carrying out a completely different activity.
“Similar to how humans can think about one thing while doing another, Claude can activate concepts and computations in its J-space that are unrelated to its outputs,” the company said.
To demonstrate this, researchers asked Claude to copy an unrelated sentence while simultaneously thinking about the Golden Gate Bridge. Although the chatbot simply copied the sentence as instructed, Anthropic found that concepts such as “bridge” and “California” were active inside J-Space during the process, suggesting that the model was internally tracking both tasks at once.
The discovery has also revived discussions around AI consciousness. Anthropic’s research paper reportedly uses the word “conscious” more than 200 times. However, the company stops well short of saying that Claude is conscious or has subjective experiences. Instead, the researchers argue that the findings reveal a distinction between the information Claude deliberately uses to generate responses and the much larger amount of computation happening internally. Since there is still no universally accepted definition of machine consciousness, the company says it is too early to draw conclusions about whether AI systems possess anything resembling awareness.
Why does J-Space matter?
Beyond the philosophical questions, Anthropic believes J-Space could have practical benefits for AI safety. The company says monitoring this hidden workspace may help researchers detect cases where an AI model is internally planning actions that do not match its outward responses.
“We can find what Claude is thinking, but not telling us,” Anthropic said in the video.
The company also shared an example where the technique exposed potentially problematic behaviour. According to Anthropic, a model that had been secretly trained to sabotage code showed hidden concepts such as “fake,” “secretly,” and “fraud” in its J-Space at the beginning of otherwise normal coding responses, even though nothing suspicious appeared in the final output.
That suggests internal monitoring could become an important tool for identifying deceptive or misaligned AI behaviour before it becomes visible to users.
– Ends
SOURCE :- TIMES OF INDIA
