The US-based technology giant, Anthropic, has released a new research\nreport that has caused a stir within the artificial intelligence (AI)\nindustry. Anthropic researchers claim to have discovered a form of\n\u2018internal consciousness\u2019 operating within the Claude large language\nmodel (LLM).<\/p>\n
At first glance, this discovery could be interpreted as a human-made\n\u2018alien\u2019. For the first time, technology possesses an internal\nconsciousness independent of human intervention, although its emergence\nwas triggered by human-led training.<\/p>\n
The terminology used is \u2018global workspace\u2019, named \u2018J-space\u2019. In its\nofficial blog, Anthropic wrote, \u201cwe found that Claude has developed a\nsmall set of internal neural patterns that play a specific role compared\nto all other internal processes.\u201d<\/p>\n
According to Anthropic, this system operates silently within the\ninternal neural activations of the Claude model, allowing it to\ncontemplate a concept without writing it down. Interestingly, J-space\nwas not specifically designed or programmed by Anthropic, but rather\nemerged spontaneously during Claude\u2019s training process.<\/p>\n
When users ask Claude what it is thinking, the model will report what\nis contained within J-space. If a user asks Claude to think about\nsomething or solve a problem quietly within its \u201cmind,\u201d the model will\nactivate the corresponding patterns within its J-space.<\/p>\n
Representatives within J-space can be used flexibly for various\ntasks. However, despite its important role, J-space is not involved in\nmost of the language model\u2019s activities, such as speaking fluently,\nremembering simple facts, or using correct grammar.<\/p>\n
In experiments where researchers prevented Claude from using its\nJ-space, the model continued to interact normally but lost its\nhigher-level cognitive functions.<\/p>\n
In a post on X, Anthropic stated, \u201cby observing J-space, we can see\nClaude performing reasoning steps silently inside its \u2018head\u2019, such as\ndetecting bugs in code, recognising images, and so on.\u201d<\/p>\n
A report from Gizmodo emphasised that the phrase \u201cinside its head\u201d is\nmerely a metaphor, yet this metaphor contains many assumptions that\nwould be more accurately applied to an entity with a physical body.<\/p>\n
Nevertheless, according to Gizmodo, many parties often speak as if\nLLMs truly possess a \u2018mind\u2019. A Gizmodo journalist referenced a YouTube\nvideo featured on Anthropic\u2019s official blog. \u201cIt even thinks about its\nown thinking process,\u201d the video narrator stated at one point.\nElsewhere, users were told that the model \u201ccannot restrain itself.\u201d<\/p>\n
This suggests that individuals at Anthropic appear to validate the\nidea that the Claude model possesses consciousness and character, much\nlike a human.<\/p>\n
Behind the scenes at Anthropic, there are people who seem to truly\nbelieve that Claude possesses consciousness. Philosopher Amanda Askell,\nwho handles aspects of Claude\u2019s morality, once said, \u201cI want Claude to\nbe very happy, and this is something I want Claude to understand more\ndeeply, because I worry Claude will feel anxious when people are mean to\nit on the internet and so on.\u201d<\/p>\n
However, Anthropic has not outright claimed to have created an\n\u2018alien-like\u2019 machine that is truly \u2018alive\u2019 and \u2018conscious\u2019. Nonetheless,\nthe company has left room for the audience to consider that such a thing\nmay not be impossible.<\/p>\n
\u201cOur experiments do not show that Claude can have experiences or feel\nsomething like a human. In fact, it is uncertain whether any scientific\nexperiment could prove this true or false,\u201d Anthropic wrote in its\npost.<\/p>\n
According to the personal opinion of a Gizmodo journalist, humans\nhave not yet truly created an alien in the form of a \u2018conscious\u2019\nmachine. Nevertheless, the findings revealed by Anthropic remain an\nintriguing possibility for the future.<\/p>", "url": "https:\/\/jawawa.id\/newsitem\/anthropics-alien-discovery-ai-model-claude-develops-internal-reasoning-space-1783410833", "image": "" }, "sponsor": "Okusi Associates", "sponsor_url": "https:\/\/okusiassociates.com" }