You might think that a bee foraging in your garden and a browser window running ChatGPT have nothing in common. However, recent scientific research is seriously considering the possibility that one or both are conscious.
There are many ways to study consciousness. One of the most common is measuring how animals (or artificial intelligence (AI)) behave.
But two new papers on the possibility of consciousness in animals and AI suggest new theories on how to test this. The theory falls somewhere between sensationalism and inevitable skepticism about whether humans are the only conscious beings on Earth.
intense discussion
Questions about consciousness have long sparked intense debate.
Part of the reason is that conscious beings can be morally significant, unlike unconscious ones. Expanding the realm of consciousness means expanding our ethical horizons. Even if we are not sure whether something is conscious, we can err on the side of caution by assuming that it is. This is what philosopher Jonathan Birch calls the precautionary principle of sensation.
The recent trend is to expand.
For example, in April 2024, a group of 40 scientists at a conference in New York proposed the New York Declaration on Animal Consciousness. The declaration, which has since been signed by more than 500 scientists and philosophers, states that consciousness is realistically possible in all vertebrates (including reptiles, amphibians, and fish), as well as in many invertebrates, including cephalopods (octopuses and squids), crustaceans (crabs and lobsters), and insects.
In parallel, the incredible rise of large-scale language models such as ChatGPT has made the possibility of machine consciousness a serious possibility.
Five years ago, the ironclad test for whether something was conscious was to see if you could talk to it. Philosopher Susan Schneider has suggested that if there is an AI that thinks convincingly about the metaphysics of consciousness, then it might be conscious.
If we followed these criteria, today we would be surrounded by conscious machines. Many people go so far as to apply the precautionary principle here as well. The burgeoning field of AI welfare is dedicated to figuring out if and when we should be paying attention to our machines.
But all of these arguments rely on largely surface-level behavior. However, that action may be deceptive. What matters for consciousness is not what you do, but how you do it.
See how AI works
A new paper in Trends in Cognitive Sciences, co-authored by one of us (Colin Klein), builds on previous work to focus on machines rather than AI behavior.
We also draw on the tradition of cognitive science to identify a plausible list of indicators of consciousness based on the structure of information processing. This means that we don’t have to agree on which of the current cognitive theories of consciousness are correct, and that we can create a useful list of indicators of consciousness.
Some indicators (such as the need to resolve trade-offs between competing goals in a situationally appropriate manner) are shared by many theories. Most other indicators (such as the presence of informational feedback) are only required by one theory but implied by the other.
The point is that all useful metrics are structural. They all have to do with how the brain and computer process and combine information.
What’s the verdict? Existing AI systems (including ChatGPT) are unaware. The emergence of consciousness in large-scale language models is not achieved in a way similar enough to us to justify the ascription of conscious states.
But at the same time, there is no barrier to AI systems (perhaps systems with very different architectures than today’s systems) becoming conscious.
What about the lesson? AI can act As if without consciousness There is Conscious.
Measuring insect consciousness
Biologists are also looking at the mechanisms by which we perceive consciousness in non-human animals: the workings of the brain.
A new paper in Philosophical Transactions B proposes a neural model of minimal consciousness in insects. This is a model that abstracts away anatomical details and focuses on the core computations performed by a simple brain.
Our key insight is to identify the types of computations our brains perform to create experiences.
This calculation solves an ancient problem from our evolutionary history that arises from having a mobile, complex body with many senses and conflicting needs.
The important thing is that we have not determined the computation itself, and it is not yet scientifically understood. But we will show it. It was done Once we know that, we will have a level playing field for comparing humans, invertebrates, and computers.
same lesson
The problem of consciousness in animals and computers appears to be moving in different directions.
In animals, the question is often how to interpret whether ambiguous behaviors (like a crab dressing a wound) indicate consciousness.
In the case of computers, you have to decide whether an apparently well-defined behavior (a chatbot speculating with you about the purpose of its existence) is a true indicator of consciousness or just role-playing.
But as the fields of neuroscience and AI advance, both are converging on the same lessons. That is, when determining whether something is conscious, how it works turns out to be more informative than what it does.
