In early December 2019, HSE University welcomed Onur Güntürkün from Ruhr-University Bochum, a leading expert in studies of thinking, to deliver a lecture entitled ‘Cognition without a Cortex.’ Among other things, Onur Güntürkün is known for demonstrating that magpies pass the mirror test, i.e., recognize themselves in a mirror. IQ.HSE attended Prof. Güntürkün’ lecture, which was organized with the support of the German Research Foundation, DFG, and asked him several questions about the problem of thinking, and why cleaner fish turned out to be ‘smarter’ than octopi.
— We have a ‘thinking machinery’, which does the thinking, and at some point, the thinking produces the language, i.e., a method of symbolic information encoding, a kind of ‘swap file’ for thinking, and this is how we get the intellect as we understand it. Do we need a brain for all of this, or would any symbol-operating machinery work?
Corals do not have brains, although they do have nervous systems. They have to move, and to move, nerve cells are needed. Nerve cells don't really make a brain, because they can be distributed like a network. But in the moment, you coordinate your input with an output, and you have to have some form of coming together, and this is the primitive version of the brain. They are also extremely simple reflexive machineries: there is light coming —hide; there is no light —go out. But in the moment, you want to go beyond reflex machineries, so you need more neurons, more aggregations. And then you are in the game of having a brain.
Onur Güntürkün / Ksenia DoroninaThe question is, is the brain the only thing that helps us to think, and the answer is, no, there is more to it. Let me give you an example. I have this bottle in my hand. And I'm looking into your eyes, and I know it, and put it on my knee. I do this without looking at it. Previously we thought that this is a very complicated calculation. I have to know the distance, the power in my muscles, and the weight. I calculate all this to make this movement. But the reality is, I don't do it. I simply go down slowly, and when my pinky finger tells me that I touch the knee, I stop doing it.
— Are you talking about embodied cognition?
Yes. The presence of a finger with a tactile surface helps me to reduce the complexity of the calculation. Still, a calculation is needed, but it's a much simpler calculation than we had thought before.
— Does this mean that to be intelligent, you don't need a brain, but a very complex body?
No, if the body is complex, the brain is complex, too.
— Let me ask you a couple of questions about the mirror test. Your group has proven that magpies pass it, and there are also results with primates, elephants and dolphins. All of them are animals that are commonly seen as ‘smart’ ones. But earlier this year, Japanese researchers demonstrated that cleaner wrasse, a fish, also passes this test. This result was puzzling to the academic community. What do you think of it?
I think this is a great experiment, and I'm confident, not 100% sure, but confident that the conclusion is right. I'm very enthusiastic about this publication, because I do not believe that self-recognition is the privilege of a very small number of animals. I think that self-recognition is much more widespread.
When we walk, we see our shadow following us. And we know it is our shadow, not a strange other animal following us. All animals have shadows. All animals have similar aspects of their everyday life. They have to understand what belongs to them, and what belongs to others. I see a shadow coming –there must be another animal coming. I see my shadow walking – it is not an animal.
It’s possible that self-recognition is much more widespread. I think that there are three levels of mirror self-recognition. The most primitive version is 'I think this is another animal'. The highest is 'I think it is me'. But there is possibly a very large number of animals between these two points: 'I know it is not another animal, because its movements are one-to-one correlated with my movements, but I'm not clever enough to understand that it has dependence'. We know this from experiments with different animals, such as chickens, pigeons, and different monkeys that are not able to recognize themselves in the mirror by classic tests but behave very strangely in front of the mirror. They do not behave as if there is another monkey, they behave as if there is a very uncanny, a very weird individual in front of them. An individual that does identical things, but I don't know what that is.
— Does this mean that it's something not present in the environment I'm accustomed to? And that’s why I'm scared, right?
When you look at children at about 18 months, they learn that the mirror image is themselves. And we learn it by detecting the correlation between the movement in the mirror and our own movement. We detect this correlation earlier. But only at about 18 months do we have this 'a-ha' moment: it's me! This 'a-ha' moment does not occur at 12 months, or at 9 months: it occurs at about 18 months of age. You need ontogenetic of your mental life to be able to understand that this correlation is due to ‘me being there’.
I think this is also important to understanding the mental life of other animals. In these three groups of animals, I believe that many more animals probably recognize 'it's me', but they don't show it to us, because they don't pass the classic mirror test, or we haven't tested them. There is a sizeable group of animals that detect the correlation and are anxious of this strange animal, but they are not clever enough to understand 'it's me'. And then, there are animals, the simple ones that just think 'that's another guy'.
— You have partly answered my next question about the mirror test, but I’m going to ask it anyway. Cleaner fish vs. octopi. The fish passes, but octopi don’t. If I ask you, which of them looks more intelligent, I think you will still respond that it’s the octopi, right?
I could, because the movements of the octopi are so much more complex. And we are used to complexity being correlated with cognition. But if we look at the social life of these two animals, the cleaner wrasse vs. the octopi, the cleaner wrasse does unbelievable things.
They clean other animals from parasites. They have a cleaning shop, a spot in the ocean, where they live, usually with their partner (they are monogamous) and wait for customers.
These customers are usually much larger than they are themselves, and potentially dangerous. They remember who is in line: first, second, third, and fourth. They remember the sequence of customers arriving. And they serve them by sequence. They only make two exceptions: if a customer is very dangerous, it is served first. Or if a customer comes every day, it is also served first.
They also observe their partner. If their partner is not cleaning well, they have a fight with their partner. Think of the complexity of this social communication.
— And octopi live alone.
They are solitary animals. At that level, you may begin to believe that this little cleaner wrasse fish is a more complex cognitive-enabled animal.
— So, you weren't surprised by these researchers’ results?
I was surprised, but I was positively surprised. I know these people, and I know that they are very hard-working and very good scientists. I believe in their data, and I tend to believe in their interpretation.
— We got the mirror tests, the tools test, and so on. All these tests check for some cognitive ability, is that right? And by summing all the tests into one, we hope to get the idea that someone could be intelligent. And as far as I understand, there is a blurred line between the idea of intelligence and all these abilities we can check for. And we don't have good enough experiments and good enough tests to strive through this zone. Is that right?
My worry is that our tests are not the bad side of it, but rather the lack of good theories. We lack good theories to combine the different test results into cognitive theories that are possibly correct, testable and deep. This, I would say, is our most severe problem.
I tend to believe that we have enough tests. There are never enough tests, I'll put it like that, it's always good to have more tests to test different aspects that we haven't tested yet. But our main problem is not the development of new tests, but the development of good theories. Because when I say 'cognition', what does that mean exactly? It's information processing, but this is not an answer; it is just another word.
For example, you can concentrate on what we call 'executive functions'. Then you can boil it down to some core mental processes. But we lack a good theory that combines the core aspects of cognition into a machinery that could give us the ability to test it with causalities. I would say that this is the most important challenge.
— This looks like something parallel to what is called the 'hard problem of consciousness'. Would it be correct to say that the problem of understanding cognition is somewhat similar to the problem of understanding consciousness? As far as I remember, Chalmers, who invented the ‘hard problem’, said that cognition is the easy part.
I think it's the other way around. I think the ‘hard problem’ only looks so difficult because we can only test this by first-person experience. It is me who perceives red, and you cannot know how I perceive the colour red. The principle, the scientific problem – I don't think that it is so complicated. But it is so complicated because we don’t have a test to really look into one’s conscious experience.
The problem of cognition is more complex, because it is made of such different sub-components. Each sub-component has different functionalities and possibly different brain structures that are responsible for it. It's a more diverse and a more multi-faceted problem. Therefore, I think that the problem with coming up with a good theory of cognition is more complicated. But maybe I only say this because I'm not interested in consciousness.
— Right now, what in your view is the best approach to this general area of cognition? Now we have the behavioural tests and the theories that we can formulate based on our results and our understanding of what is what. And we've got neuroscientists who see the brain and model the connectors and try to connect these data with behavioural data. Which angle of attack seems the most fruitful to you right now, and how should we proceed?
It is a difficult question, because it is a difficult question for everybody to ask. First of all, we are living in a time when we think that deep learning has solved all that. But it didn't solve anything. It's just a system that is so powerful because it can iterate by computers within milliseconds all kinds of learning procedures with unbelievable speed. We do not have the luxury in our real life. We are superior to these computers in many ways.
So, when I see a bottle falling down, like this bottle, I only have to experience this a few times to first have top-down expectation of the next bottle falling. The Google machine needs 10,000 iterations until it gets it. But since its iteration happens within a split second, it looks so smart. It is a way of deep learning and deep mind that lacks the predictive power that develops so fast in biological systems.
Still, it is incredibly interesting and, therefore, we need these artificial intelligence experts in the club for the future development of the theory of cognition. We need them very much. But we should not think that they have the answer. They are part of the answer that we are seeking.
We need experimental psychologists and biologists to do the critical experiments. We need neuroscientists and neurobiologists, and we also need theoretical neuroscientists who work with brain-like structures, but on a computer basis. They model neuronal networks that aren't biologically realistic brain simulations, but they are not inspired by biology like the networks of Google. Rather, they are biologically realistic, so we can say that the properties running there are very similar to the properties running in the animals. We need all that. And, obviously, we also need philosophers who specialize in the philosophy of mind. This is simply an accumulation of bright-minded people. There is no guarantee that this accumulation will result in anything. But what else can we do?
— If we manage to build an artificial brain that is a total copy of the natural brain, and have it pass some cognitive task, will we perform this experiment as a demonstration that we are powerful enough to do it? Maybe you have thought about the main threshold we need to pass in the direction of creating this theory of cognition.
What you describe is the core idea of the Human Brain Project of the European Union, which is expected to cost half a billion euro. It's huge. It is the core idea of Henry Markram, a neurobiologist who once said, 'let's build a cortical column of a rat one-to-one in computer, exactly as it is, and then multiply by ten million, and then, we have consciousness.’
It failed because this is a pure bottom-up approach. But you need both: you need the pure bottom-up approach and combine it with a top-down approach. To give you a very simple example, in working memory, when we make a mistake, it's usually a phonological mistake: we confuse Kiev with kiwi, even though kiwi is a fruit and Kiev is a city. But in the long-term memory, we confuse baron with king. Both of them are aristocratic titles. So, we confuse semantic categories, while in working memory we confuse phonological categories.
You need top-down knowledge, and you need experimental psychology to know that. And then you have to say, look, Henry Markram, will you craft this? This is a future of a real brain. Do you find that in your cortical columns? Perhaps not. Because we need a wiring, a top-down topology that creates these functions that are not easy to achieve from the bottom up, we need to combine both approaches. And this is now happening in the Human Brain Project more than what happened previously. It's a bold endeavour that the European scientists are undertaking. I don't want to be pessimistic, but I think we need more time than what given by the European Union to achieve this goal.
— If not Human Brain Project, which project looks more promising to you right now?
I think the Human Brain Project is a promising one, but I think the whole world of cognitive neuroscience is accumulating knowledge at such a rapid speed that we will then be possibly able to come up with better theories. This is our hope since better theories emerge in the brains of humans. The more data there is, the more difficult it is to have a theory. But it's not impossible. I think that over time we will come up with that. But it will not come from a single project, the Human Brain Project or some other for that matter since the United States has similar projects, China has similar projects, and both of them have different designs. I am certain that despite not having a similar project of its own, Russia has great scientists also working on cognition. And this is true for many more countries. We need all that. We need science. That's simple, and it's the bottom line. We need science.