Whenever you interact with a big language model (LLM)—certainly one of the systems behind chatbots akin to ChatGPT and Claude—it could possibly feel as if you’re in touch with one other conscious mind. But are you, really?
Some outstanding scientists, akin to Geoff Hinton and Richard Dawkins, claim you’re. But most experts remain skeptical, arguing that the impressive cognitive capacities of LLMs occur within the absence of consciousness.
Recently, researchers at Anthropic, the corporate behind Claude, waded into this debate with an interesting finding. They claim Claude has a normally invisible set of representations of knowledge that guide its internal reasoning and its verbal output.
That is where it gets interesting. The researchers argue this finding could be understood by way of an influential theory of consciousness called the worldwide workspace theory.
What Is the Global Workspace Theory?
First proposed by the psychologist Bernard Baars in 1998 and further developed by the neuroscientist Stanislas Dehaene and his collaborators, this theory holds that consciousness involves the activity of a “global workspace.” It is a type of processing hub within the mind or brain that integrates and broadcasts information, allowing it for use for reasoning, behavior control, and speech.
In a glossy video explaining the work, Anthropic depicts the contents of Claude’s “global workspace” as sailing ships afloat on an enormous sea of unconscious mental activity.
How should we react to those developments? Do they supply evidence for artificial consciousness? If that’s the case, how strong is that evidence?
What Is a Global Workspace?
We are able to start by asking whether Claude does indeed have a “global workspace.” This will not be straightforward, for the idea gives no formal definition of a world workspace.
The notion is characterised only informally. The (typically implicit) assumption is that any computational workspace “similar enough” to a human’s will qualify as a “global workspace.” But how similar is analogous enough?
Claude’s workspace may indeed have much in common with ours, but there do look like differences.
For instance, the brain’s workspace is sustained by recurrent loops—signals cycling back through the identical circuits over time. In contrast, Claude’s workspace evolves over a single go through the network.
A related difference concerns how representations enter a workspace. Advocates of worldwide workspace theory have long argued that in humans, a process called “ignition” occurs by which a non-linear process amplifies and sustains neural representations, allowing them to enter the workspace. So far as we all know, nothing comparable occurs in Claude’s case.
Do these differences matter? The reply will not be clear. Global workspace theory is predicated on data drawn from adult humans. There are questions on how far the notion could be—or ought to be—prolonged.
Does a Global Workspace Imply Consciousness?
But let’s suppose Claude does have a world workspace. To work out whether that may be evidence for Claude being conscious we’d like to think about the status of the worldwide workspace theory of consciousness.
There isn’t a doubt it’s certainly one of the most influential theories of consciousness, however it’s hardly uncontroversial amongst experts. (In a reasonably extreme understatement, Anthropic’s paper remarks that “the worldwide workspace model will not be universally accepted.”)
Many consciousness experts argue that computational properties alone are enough for consciousness. Even amongst those that think that consciousness is inherently computational, global workspace theory is barely certainly one of many options.
‘Conscious Access’ and Subjective Experience
What’s more, there are questions on whether global workspace theory is basically a theory of consciousness within the relevant sense in any respect.
In an influential paper on artificial consciousness, the neuroscientist Dehaene and his collaborators advance the idea as an account of what they call “conscious access”—the provision of knowledge for recall, the voluntary control of behavior, and verbal report. Crucially, they leave open the query of whether global workspace theory ought to be understood as an account of the subjective or experiential components of consciousness.
But when global workspace theory is only a theory of “conscious access,” then its implications for the factitious consciousness debate lose much of their significance. After we ask whether Claude is conscious we don’t need to know whether it has “conscious access”—as an alternative, we wish to know whether there’s anything, subjectively speaking, that it’s prefer to be Claude. Global workspace theory doesn’t speak to that query if we treat it as nothing greater than an account of “conscious access.”
So Has Artificial Consciousness Arrived?
Even taking these complications into consideration, there isn’t any doubt that Anthropic’s findings are noteworthy. Global workspace theory could be understood as a theory of subjective experience, and Claude may indeed prove to have something akin to a “global workspace.”
None of that is evidence that artificial consciousness has arrived. But it surely’s not unreasonable to think these findings do move the dial—if only ever so barely—in the factitious consciousness debate.
But when that’s right, then it’s puzzling why Anthropic is kind of so upbeat about these developments. As Anthropic recognizes, the creation of artificial consciousness could be a momentous event with wide-ranging social, ethical, political, and legal ramifications.
If chatbots are conscious then we would wish to take their interests seriously. It could now not be permissible to treat them as mere machines; as an alternative, we would wish to consider their welfare.
Should Anyone Even Be Attempting to Do This?
Anthropic remarks that “it’s time to start out fascinated with whether we ought to be constructing conscious machines.”
I agree we’d like to have that discussion, but we must always also pause work on constructing machines that may potentially be conscious. If Anthropic were serious, it will surely down tools reasonably than plough ahead with its try and develop conscious AI.
A moratorium on AI research that is likely to be thought to steer to conscious AI would, in fact, be removed from straightforward. There are questions on the range of research it will affect and who might implement it. But when we don’t close the stable door now we’d find that the horse has already bolted.