How Will We Know When AI Becomes Conscious?

0:00Today on the AI Breakdown, we look at a new paper attempting to create a systematic way to determine when AI becomes conscious conscious. The AI Breakdown is a daily podcast and video about the most important news and discussions in AI. Go to Breakdown.network for more information about our YouTube, our Discord, and our newsletter. One of the big questions in artificial intelligence is, of course, about consciousness. And at what point machines or systems like AI actually become sentient, become conscious? Today on the AI Breakdown, we are exploring a new paper which attempts to create some type of systematic framework for answering that question.

0:40The massive paper was called Consciousness and Artificial Intelligence, Insights from the Science of Consciousness, and had 19 co-authors spanning from artificial intelligence experts to philosophers with everything in between. Notable contributors include Robert Long from the Center for AI Safety, Turing Award winner Yoshua Bengio from the University of Montreal, and other contributors from New York University, the University College London, the University of California, Irvine, and beyond. They begin the paper, The question of whether AI systems could be conscious is increasingly pressing. Progress in AI has been startlingly rapid, and leading researchers are taking inspiration from functions associated with consciousness in human brains in efforts to further enhance AI capabilities.

1:21Meanwhile, the rise of AI systems that can convincingly imitate human conversation will likely cause many people to believe that the systems they interact with are conscious. The attempt they're making then is to move from the realm of feeling and sentiment and sensibility into the realm of science for determining consciousness. And to do so, they're examining a number of different neuroscientific theories of consciousness. Now, even before we got to ChatGPT, there were already some pretty serious conversations around exactly this issue. You might remember in 2021 when Google engineer Blake Lemoine made headlines by claiming that Lambda, which was a chatbot that he had been testing, was sentient.

1:57That ended up getting Blake fired and igniting a firestorm around exactly this type of conversation. One of the authors, Robert Long, said, When Blake Lemoine was fired from Google after being convinced by Lambda that marked a change, if AIs can give the impression of consciousness, that makes it an urgent priority for scientists and philosophers to weigh in. Science magazine helped explain their methodology a little bit further. They write, How does one go about probing the phenomenal consciousness of an algorithm? Unlike a human brain, it offers no signals of its inner workings to detectable with an electroencephalogram or MRI.

2:28Instead, the researchers took a theory-heavy approach. They would first mine current theories of human consciousness for the core descriptors of a conscious state, then look for these in an AI's underlying architecture. To be included, a theory had to be based on neuroscience and supported by empirical evidence, such as data from brain scans during tests that manipulate consciousness using perceptual tricks. It also had to allow for the possibility that consciousness can arise regardless of whether computations are performed by biological neurons or silicon chips. So which theories did they end up thinking had something to offer this conversation around AI consciousness?

3:00The first was something called recurrent processing theory. They write, RPT are sometimes referred to as local as opposed to global theories of consciousness because they claim the activity of the right form in relatively circumscribed brain regions is sufficient for consciousness. RPT, they say, is primarily a theory of visual consciousness. It seeks to explain what distinguishes states in which stimuli are consciously seen from those in which they are merely unconsciously represented by visual system activity. Continuing, they write, The theory claims that unconscious versus conscious states correspond to distinct stages in visual processing.

3:32An initial feed-forward sweep of activity through the hierarchy of visual areas is sufficient for some visual operations, like extracting features from the scene, but not sufficient for conscious experience. When the stimulus is sufficiently strong or salient, however, recurrent processing occurs. in which signals are sent back from higher areas in the visual hierarchy to lower ones. This recurrent processing generates a conscious representation of an organized scene, which is influenced by perceptual inference, processing in which some features of the scene or percept are inferred from other features.

4:01Next up, they look at global workspace theory. They write, The Global Workspace Theory of Consciousness, GWT, is founded on the idea that humans and other animals use many specialized systems, often called modules, to perform cognitive tasks of particular kinds. These specialized systems can perform tasks efficiently, independently, and in parallel. However, they are also integrated to form a single system by features of the mind which allow them to share information. This integration makes it possible for modules to operate together in coordinated and flexible ways, enhancing the capabilities of the system as a whole.

4:32GWT claims that one way in which modules are integrated is by their common access to a global workspace, a further space in the system where information can be represented. Information represented in the global workspace can influence activity in any of the modules. The workspace has a limited capacity, so an ongoing process of competition and selection is needed to determine what is represented there. GWT claims that what it is for a state to be conscious is for it to be a representation in the global workspace. Another way to express this claim is that states are conscious when they are globally broadcast to many modules through the workspace.

5:03Now beyond that, they also categorize a group called higher order theories. They write,

5:16This is accounted for by an appeal to higher-order representation, a concept with a very specific meaning. Higher-order representations are one that represents something about other representations, whereas first-order representations are ones that represent something about the non-representational world. This distinction can be applied to mental states. For example, a visual representation of a red apple is a first-order mental state, and a belief that one has a representation of a red apple is a higher-order mental state. The other theories that they talk about include attention schema theory.

5:46Attention schema theory is another example of a higher-order theory of consciousness, because, as they say, it claims that consciousness depends on higher-order representations of a particular kind, in this case, representation of our attention. They write, The attention schema theory of consciousness, AST, claims that the human brain constructs a model of attention which represents and may misrepresent facts about the current objects of attention. This model helps the brain to control attention in a similar way to how the body schema helps with control of bodily movements. Another theory they explore is called predictive processing.

6:16They write, Predictive processing claims that the essence of human and animal cognition is minimization of errors made by a hierarchical generative model in predicting sensory stimulation. In Perception, this model is continually generating predictions at multiple levels, each influenced by predictions at neighboring levels and in the immediate past, and by prediction error signals which ultimately arise from sensory stimulation itself. They write, Although PP is not a theory of consciousness, its popularity means that many researchers regard predictive processing as a plausible necessary condition of consciousness.

6:45Mid-brain theory, they say, rests on a different part of the brain, writing, While the neuroscientific theories of consciousness we have discussed so far focus primarily on cortical processes, Merkur 2007 argues that the cortex is not necessary for consciousness. Another influential theory, they say, in animal consciousness literature is unlimited associative learning. They write, The proposal here is that the capacity for unlimited associative learning, UAL, is an evolutionary transition marker for consciousness, A single feature that indicates that an evolutionary transition to consciousness has taken place in a given lineage.

7:15The hallmarks of consciousness, according to UAL, include

7:31Now, this is a massive paper that goes deep into understanding and explaining all of these, as well as into examining in what ways current AI systems do or don't match elements of these theories of consciousness. At the conclusion, they suggest that none of the current AI architectures is likely to be conscious, at least at this time. However, their goal is to create a framework that can be used to apply as we get closer and closer to that huge milestone. The implications, of course, of this are huge and much more than just whether one Google engineer gets fired. AI consciousness, for example, would bring up serious questions of rights, enfranchisement, things that have so far been firmly in the realm of sci-fi, but which might not be for long.

8:12Anyway, I will of course leave a link to this extremely dense but really interesting paper in the show notes, and hopefully this gave you some food for thought for your weekend. Thanks as always for listening or watching, and until next time, peace!

8:37Thank you.