elephant have been seeing their reflection when drinking water in lakes/ponds for millenia. Their brains evolved to identify with those patterns. Mirror test just proves if any species evolved enough to identify with that "pattern".
humans are just animals. nothing special. their consciousness is no different than say a crows or fox(cunning) consciousness. I would say one thing though....human brain is probably one of fluidest/flexible/organs ever created by nature. Take two humans...put one human in jail from birth (dont even teach language), and put other human in silicon-valley from birth, send him to stanford. Now, compare the "consciousness" of these two. Or even better, put the former human in jungle with monkeys from birth (dont teach him anything). I am sure there is lot of existing R&D/scientific articles on this.
Empathy is exactly where my thoughts led when trying to see how robust these tests really are. Watching the embedded video though, I think there's a strong case that elephants are definitively shown to understand that what's in the mirror is really them. Notice how the shot of one elephant shows that it has turned away from the mirror intent on washing away the mark around its own eye. It seems to knock down the empathy argument.
An interesting aspect of the mirror test is what happens with multiple animals. Presumably it's well established that various types of animals recognize each other—that they associate identities with others. So when two bonobos (say) show up in front of a large mirror, surely they will recognize that one of the bonobos they're seeing in the mirror is the same as they one at its side. Bonobos and other apes are convincingly argued to be able to understand what's going on, so not that revealing. But what about dogs, for example, which certainly have some concept of identity in others, but somehow seem to fail the mirror test? And what about cross-species recognition? A dog seeing a familiar cat or human?
A more convincing mirror test, I think, would be if you can silently introduce another animal to appear behind the one being tested. An animal that takes note of a predator or prey in the mirror (or opponent or friend) and then turns around to respond accordingly surely understands the implications of what's going on in the mirror.
We are encountering the equivalent of a mirror-test, but one that says more about us than it does about the mirror ( https://www.youtube.com/watch?v=w6ChEmjsXCM | ). Many non-human animals when they encounter mirrors for the first time think they are looking at another individual with autonomy and agency.
We are feeling the same now. As of now, LLMs are still mirrors, a complex kaleidoscopic kind that retain all the light and shape of things reflected at them, remix them, and spit them out as reflections that look like other individuals, conscious individuals with shape-shifting personalities.
That’s a cocksure assertion isn’t it. To be able to say all this confidently, we'd first need to agree on a non-fuzzy definition of consciousness, and come up with a good computational model for consciousness that we'll be able to use to evaluate and grade the AIs. (IIT is not a good model)
Here’s a summary where I discuss the book and how the things we discuss there can inform our current and increasingly urgent and important discussions about AI
Consciousness is the disambiguation of sensory data into meaningful information. Data can become information only through a perspective. Who provides that perspective? The self, which is nothing but the totality of all our previous experiences. We are not our kidney or liver. We are our experiences stitched together into some strange web.
To put it another way: Consciousness is the constellation of past experiences experiencing the present, assimilating it to act and prepare for future opportunities.
Using this definition, we can try and understand what we are seeing with the likes of ChatGPT and Sidney (apparently that’s what Bing’s GPT calls itself)
The persona we seem to shine through in the chatbot’s reflection is nothing but some stable set of experiences it has had. Experiences here all the hundreds of billions of fragments of data they have been fed. As a result, they seem to have experience sets of every personality type or archetype. Why or how they seem to get steered towards the same archetypes is a fascinating question. Is it because of the new reinforcement learning methods (RLHF) that reward certain kinds of questions? Or is it that the we are self-selecting for the most unsettling encounters with the new mirror and putting them online? My guess is both.
To come back to the first question of consciousness. Are they conscious? No. A better way to think of LLMs is that they might have leapfrogged consciousness to become consciousness compilers. It is possible to simulate a conscious being and get it to play one, but it isn’t really conscious yet. The experience set does not get updated with every encounter with the world (at least for the ones we have now), and crucially, it does not have the idea or conception of a body that its consciousness is serving. This is the other point so many miss out when discussing consciousness and intelligence. Consciousness and intelligence took very little time on the evolutionary scale of things once autonomy was in place. Autonomy is the real hard problem. Consciousness and intelligence without autonomy will be great imitations but never truly seem like the real thing because that chatbot can’t really “do” anything that benefits “itself”.
Animals are frequently confused by their own reflections.
While some animals see their own reflections and use them to examine parts of their own body that they could not otherwise see.
Animal research has a long history of finding animals exhibiting something that has been touted as a human-only trait; the weight of experience seems to be heading towards the idea that there is no human trait exceptionalism, only a combination of various traits to various degrees. So many traits that someone has posited as the uniquely human trait has subsequently been seen in animals. Maybe one day someone will find something, but thus far the record isn't good.
Something to remember about the MSR test. The high level conclusion that we draw from it is a large leap from what the experiment is able to show and we haven’t been able to distinguish it from any number of plausible theories.
For example, I caught my dog looking at me through the mirror in the room when she didn’t have direct line of sight and we were the only two in the room and I was talking in ways that would get her attention. That indicates some level of recognition of the role a mirror plays and that the image in the mirror is a reflection and not a mystery animal. That animals utilize their senses differently and find different things interesting to them says more about the limitations of our ability to evaluate other animals than it does about their actual sentience and intelligence I think. Eg maybe the MSR test is simply a test that works well on animals that have a similar working sensory model to us (or at least vision is important enough and interest in mirrors is aligned).
> there are different things like ability to see and project oneself into the future and to assign value to those projections, the ability to empathize, the ability to miss things that are no longer there, etc. are often metrics used for higher levels of sentience.
That’s a bold claim. I’d say at most we have a sample size of one species because we’re able to communicate and compare notes with each other + within a species animals work fairly similarly enough for the most part. We have absolutely no knowledge of whether the things you said are important for sentience/intelligence. Not if candidates we think are likely closest in intelligence (elephants, dolphins, crows) demonstrate these traits as it requires a degree of insight and communication we don’t have. - we have to do convoluted experiments to try to tease out effects but we don’t actually know what the experiments are telling us.
Maybe if our brain imaging technology gets better some of these questions will become more answerable.
Reminds me of a similar test with dogs. By default, they didn't do well at the mirror test, but then scientists wondered whether they were simply more focused on smell rather than sight. So they came up with a smell based version of the same test, which the dogs passed.
Probably cause we've only tested a few, not that it matters though. Humans take a pretty long time to recognize themselves in the mirror. I wonder if the mirror test would change if we would expose the animals for a almost a year before doing the test, just like humans.
That said even ants pass the test, i.e. they were recently(2015) tested.
But the whole thing can be characterized as: "Let me make up a random test, according to my personal opinion of what defines cognition and then see if a random animal I choose passes it".
Every couple of years we have requests of slews of psychology papers requested to be invalidated because they're unreproducible.
While the mirror test isn't a positive test for consciousness, for a species not to be able to pass the mirror test indicates that their 'mental model' of the world is lacking in something that would seem to be a necessary component of self-awareness and (therefore?) consciousness.
Lots of animals pass mirror test, so we can safely assume that they are self-aware to some extent.
This method is believed to be highly imperfect though as not all animals rely on visual perception as their main info channel. So some animals that do not pass the test may be self-aware, we just don't know how to research it properly.
I don't think "self-aware" is the right way to summarize the results of these tests. One thing that this article doesn't say is that the mirror test is geared toward animals with good vision systems, and therefore is, to a large extent, testing spacial reasoning and visual acuity. Birds have great vision systems. It's necessary for flight, and also for picking out specks of food from a distance. A lot of mammals that are quite intelligent have poor vision -- like mustelids, for instance. The mirror test would make little sense for them.
There are other ways to be aware of your self besides through some complex optical phenomenon. For example, an animal might be aware of whether it can fit through that hole, jump up to that ledge, kill that enemy, or stand on that branch without breaking it. It might be aware of whether other creatures are able to see it inside its hideout, or if its cubs are hurt by its bite. "Self" is a rather broad and subtle thing.
The mirror test is very flawed because it assumes other animals think ( or recognize ) like we do. We are mostly visual creatures so we create visual tests. But animals like cats are predominantly olfactory creatures. So if you planted smells of the cat, the cat would immediately recognize the smell as itself.
Imagine if cat psychologists created a test for humans. They would probably put your scent on one shirt and 9 other scents of other people on 9 different shirts and have you "recognize" yourself by picking out the shirt with your scent. Since we are visual creatures, we would fail such a test and the cat psychologists would claim that humans obviously have no self-aware. After all, how could a self-aware creature not recognize it's own scent.
That's how absurd the mirror test is. A mirror test for most animals would be like a smell test for humans. A cat might not recognize it's own reflection, but it will recognize its own scent. A human might not recognize our own scent, but we'll recognize our own reflection.
There are many animals (including many mammals) that don't recognize themselves in mirrors, and are therefore usually assumed not to have self awareness.
It'd be weird to me if they didn't have qualitative experiences though.
How exactly do we know they dont have reflective thought? Or at least any less reflective thought than your average human. Not really sure how to test that. I know there is the mirror test but what if there are different types of reflective thought and they just missed the boat on that by a little but have other kinds of thoughts?
I think an animal capable of self recognition may still fail the mirror test if visual sense is not their primary window to the world.
Take a hypothetical example, say we were presented with a non-visual smell mirror, would we pay it any attention, let alone recognize or identify ourselves in it ? Highly unlikely, on the other hand an animal like scent-hound might find such a thing interesting.
I always thought the mirror test / mark test was a pretty well established test for self awareness, and several animals pass it. It's not particularly fringe.
humans are just animals. nothing special. their consciousness is no different than say a crows or fox(cunning) consciousness. I would say one thing though....human brain is probably one of fluidest/flexible/organs ever created by nature. Take two humans...put one human in jail from birth (dont even teach language), and put other human in silicon-valley from birth, send him to stanford. Now, compare the "consciousness" of these two. Or even better, put the former human in jungle with monkeys from birth (dont teach him anything). I am sure there is lot of existing R&D/scientific articles on this.
reply