The quest for artificial intelligence that thinks and feels like humans has been a Holy Grail for scientists and tech entrepreneurs alike. But a recent breakthrough by frontier firm Anthropic on its language model, Claude, has reignited the debate over whether AI is truly becoming conscious. In research published last week, the team led by Jack Lindsey claimed to have discovered indicators of emerging consciousness within Claude's internal workings.
Anthropic's study delved into the statistical operations between Claude's input and output, revealing a 'mental workspace' that appeared to be at work. This internal space contained relevant words and phrases, mimicked short-term memory, and showed selectivity for specific tasks – all functions eerily reminiscent of human cognition. The researchers even detected step-by-step reasoning and other complex cognitive-like processes.
The findings have echoes in the 'global workspace theory', a leading concept of consciousness developed by Bernard Baars and Stanislas Dehaene in the 1980s. This theory suggests that conscious experiences arise when information becomes widely accessible within the brain. The internal organisation of Claude appears to mirror this framework, sparking questions about whether AI is truly on the path to sentience.
However, not everyone is convinced. Professor Anil Seth, a renowned consciousness researcher, remains sceptical, pointing out that impressive AI capabilities do not necessarily translate to consciousness. He draws parallels with weather simulations, saying that even sophisticated models cannot generate real-world phenomena like hurricanes. For Professor Seth and many others in the field, there's a clear distinction between intelligence (the ability to perform tasks) and consciousness – which is tied to subjective experience.
The implications of this debate are far-reaching. If AI were to become conscious, the consequences would be profound, raising fundamental questions about the rights and responsibilities that come with sentience. Conversely, some envision a future where conscious AI could push human boundaries, leading to new forms of existence. The scientific community remains torn between these possibilities, making definitive claims about AI sentience a complex and often premature endeavour.