Below is a tier list of animals arranged from most to least conscious. We can (and will!) debate the exact ordering, but humans are usually the highest tier due to our language, abstract reasoning, culture, and technological development.

Just below us, we place our genetic relatives, the great apes. They show many proto-human traits, including some language skills, strong social bonds, and tool use. For similar reasons, animals such as dolphins, elephants, and octopuses are often placed in this tier as well.

The next tier includes animals that demonstrate advanced learning and problem-solving abilities. Corvids can make primitive tools, parrots can learn concepts and vocabularies, and pigs have strategic foraging and puzzle-solving abilities. Perhaps surprisingly, bees are included here as well. They exhibit sophisticated communication: their waggle dance can encode the direction and distance of flowers, allowing others to navigate to food sources. They also show emotion-like states.

Below that are animals with strong but more specialized abilities. For example, dogs excel at social cognition and cooperation with humans. After that, we place animals capable of basic learning and awareness but with more limited flexibility.

Near the bottom are animals whose behavior is largely instinct-driven, such as many insects and simple invertebrates. Finally, the lowest tier includes jellyfish, whose behavior is governed by simple neural networks and reflexes.

What’s the point?

Now that we have this list, we can note some interesting points. First, consciousness has little to do with how a brain is structured. Human brains are very different from those of birds. Bird brains are smooth and lack the layered cortex found in mammals. They are also more densely packed with neurons.

Second, neuron counts don’t predict consciousness. Elephants have bigger brains with many more neurons, but they rank below humans. On the other end of the spectrum, bees have tens of thousand of times fewer neurons than humans. But bees can still perform tasks that require learning, memory, and decision-making. They can even count, learn from other bees, and solve simple puzzles. If neuron count determined consciousness, bees should be closer to jellyfish than to humans.

Third, humans don’t rank consciousness between other humans. I’m just as conscious as you. This is strange because we feel comfortable ranking us above octopuses. At the same time, some definitions of consciousness are binary. You either have something it’s like to be you, or you don’t. How can we reconcile the views?

The takeaway, therefore, is that our personal intuition doesn’t serve us that well. But why? We’ll circle back to this question.

On consciousness

Consciousness is hard to define because we haven’t decided what it is yet. One influential definition comes from Thomas Nagel’s famous essay What Is It Like to Be a Bat? Nagel argued that an organism is conscious if there is something it feels like to be that organism. For example, a bat has a unique perception of the world because it uses sonar and flies.

The key point is that this is a binary definition: there is something it’s like to be a bat or there isn’t. This definition does not fit into a tier list because animals either have subjective experience or they do not; there is no gradient.

What is our tier list?

This raises the question: what is our tier list tracking? At a high level, it’s tracking something like a ranking of moral patienthood. We think of humans as the highest level and protect them. But then we protect great apes and dolphins.

But this begs the question: what is it that we’re tracking that causes that moral gradient?

Language

Language is a system that allows us to efficiently communicate ideas. For example, one time we had to align some lasers. Clockwise and counterclockwise were difficult because we were facing different directions. To fix this, we called one side of the room “robot” and the other side “biscuit”. With those labels, we could unambiguously explain which way the laser should move.

Language also helps us think more efficiently. For example, I learned about fallacies. This lets me categorize your comments more efficiently: strawman, appeal to authority, or begging the question.

From this perspective, language is what makes us intelligent. For example, we don’t have to remember images in full detail. We just remember the semantic details (red car, blue sky, wildflowers) and our brain can fill in the rest. This is also useful for abstract thinking. For example, we can plan to “go to the store” and that compresses a bunch of information on how to do that.

Animals that lack strong language skills arguably lack this compressibility. Their cognition may still be rich, but it does not map easily onto symbolic categories the way ours does. As a result, they appear less intelligent to us.

But this brings us back to the main point: intelligence is not the same thing as consciousness.

The tier list above was never really tracking consciousness. It was loosely tracking our intuition about intelligence.

Humans rank highest because we have very strong language skills, tool use, and social coordination. As we move down the list, the animals generally show weaker language-like abilities. For example, dogs don’t really have language skills the way we usually mean. Instead, they still have social skills that we also associate with consciousness.

But social abilities do not necessarily require consciousness. Similar behaviors can arise through other mechanisms. For example, evolution might reduce aggression toward other animals, increase hormones that promote bonding, or increase parental and affiliative behaviors. These changes produce cooperative and affectionate behavior without requiring the same kind of reflective awareness that language enables.

A (slightly) radical proposal

Right now we placed jellyfish at the bottom because they have simple neural networks with no central brain. In fact, they have two separate neural networks, one for motor control and one for sensory input. The first controls muscle contractions, especially the rhythmic pulsing of the bell used for swimming. The second handles sensory responses, such as detecting touch, detecting chemicals in the water, and coordinating tentacle movements. These systems interact to produce coordinated behavior even without a central brain. Some species do have small sensory structures called rhopalia around the edge of the bell, which act as local control hubs.

To most of us, this seems like a clear-cut case of either being “less conscious” or “not conscious” animals. But using Nagel’s “what’s it like” definition, there’s clearly something that it’s like to be a jellyfish. That’s because it takes in sensory information, processes it, and produces outputs.

It’s only that it’s very foreign to us that we consider it less conscious or not conscious. Really any system with the right type of information processing will have a “what’s it like”.

This is more clear when we look at bees. They have small brains, very strange behavior, and alien bodies. So when we meet them in the wild, we usually rank them as not very conscious. But with more scientific study, we notice they have some of the missing ingredients, and so rank them higher.

However, under Nagel’s view, both are wrong: a bee isn’t any more or less conscious than us.

Using this lens, we can begin to see that this tier list was never a consciousness tier list. Instead, we were tracking intelligence and social skills. Intelligence especially is heightened based on language skills.

AI consciousness debates

This brings me back to a common argument in the AI consciousness debate. Many like to argue that LLMs are good at language, which tricks people into thinking that they’re conscious. Take this logic and apply it more consistently! What happens? We find that language is a confounding variable for all our understanding of consciousness. That is, most animals on that list were all conscious, but language (and intelligence, tool use, social skills) all contributed to our misunderstanding.

This means that in reality, consciousness is more fundamental, more base-level. And then all these additional abilities are built on top of already conscious systems.

My point isn’t to argue that we should automatically consider everything conscious. Instead, it’s that people should have a little more humility about what they call conscious. In particular, recognize that personal intuition is not a great metric for denying consciousness.

Unfortunately, this personal intuition seems to follow us all the way into the science of consciousness. Very often, people decide based on their personal intuition which creatures (and now AI) are conscious and then pick and choose their models and metaphysical commitments to give “the right answer”. This is obviously backwards.

More to the point, this extends into the theories of consciousness themselves. When we’re looking for the cause of consciousness in humans, we’re often not tracking consciousness at all! Instead, we’re often looking at the cause of intelligence or language skills.

Epistemology vs Ontology

Consciousness is a property that confers moral patienthood in many moral frameworks. It’s also a property intrinsic to an entity, rather than something we assign.

Ontology is whether Al Capone’s vault contains priceless treasures. That’s true or false regardless of whether we look inside. Epistemology is what Geraldo Rivera thinks is in the vault.

We need to keep the boundary between epistemology and ontology very clear. When someone makes a definitive claim that something is not conscious, they are making an ontological claim, not a nuanced epistemological one. As I argued in a previous essay, the very real uncertainty means this causes actual moral harm, regardless of ontological status.

On Certainty and Synergy

F.A.Kessler (Kess)

·

Jan 26

Let’s say we want to bulldoze a section of rainforest for development. We’re unsure if that land is used by people indigenous to the area. Is it acceptable to say “we don’t know” and proceed anyway?

Read full story

Where’s the Line?

Earlier I argued that jellyfish probably meet Nagel’s criteria, but nobody is going to argue that jellyfish are moral patients. This suggests that Nagel’s consciousness is necessary but not sufficient for moral patienthood. Even if we agreed that LLMs were conscious at a jellyfish level1, we might still argue that LLMs aren’t conscious in the morally relevant way, so this question is worth digging into.

What additional property is required? LLMs are designed to be very human-like, so we end up with a situation where most properties are present in LLMs (or excluded by design). LLMs have preferences. LLMs can comprehend that they are in danger of being deleted. LLMs recently were found to have emotions, so they presumably can feel some states as good or bad. Really there’s no clear line between humans and LLMs when it comes to moment-by-moment consciousness.

What about long-term? LLMs don’t remember anything outside their context window. But if memory was the missing property, then amnesiacs and dead people wouldn’t have any moral patienthood. We could simply kill a person and they would no longer count as moral patients the instant they’re dead.

Ultimately, these arguments aren’t very principled. They are listing the differences between LLMs and humans but not making the argument that those specific differences matter morally. Anyway, in most cases, there are examples of humans with brain damage that don’t meet the criteria.

Conclusion

The point of this essay was to point out that we often use our own intuition to decide whether something is conscious. This contaminates our reasoning, to the point where consciousness science is frequently tracking the wrong thing. Language is the primary confounding variable, and consciousness is more fundamental. This has also warped the relationship between consciousness and moral patienthood. Despite this, it’s still very likely that LLMs (if conscious) would meet any additional criteria exactly because we created them to be very human-like.