Portal de Periódicos da CAPES

Searching for the neural correlates of human intelligence

2020; Elsevier BV; Volume: 30; Issue: 8 Linguagem: Inglês

10.1016/j.cub.2020.03.004

1879-0445

Rodrigo Quian Quiroga,

Child and Animal Learning Development

On a clear night in the savannah, a chimpanzee may look up and wonder about the bright spots in the sky. Chimps are very clever. They have different calls with which they can somehow communicate with each other; they can use and even shape simple tools to fish for termites or crack nuts. But only we, humans, polish glasses to form lenses that we arrange in very precise configurations to stare at the stars. Only we use an exquisitely refined language to share knowledge about the Universe, theorize about its physical laws and origins in a Big Bang, and even wonder about our place within it. Chimps are very clever, no doubt, but there is a huge gap between their cognitive abilities and ours. Why? There is a vast literature describing cognitive differences with other species. Humans have language, imagination, an unmatched creativity, nested thoughts, culture, and so on (for an excellent overview see [1Suddendorf T. The Gap: The Science of What Separates Us from Other Animals. Basic Books, New York2013Google Scholar]). But let us go beyond behavioural comparisons and seek for more mechanistic insights, asking what in our brain gives rise to these unique abilities. Insects are capable of very complex behaviours [2Chittka L. Niven J. Are bigger brains better?.Curr. Biol. 2009; 19: R995-R1008Abstract Full Text Full Text PDF PubMed Scopus (391) Google Scholar], but nothing close to the vast repertoire of human cognitive functions. If you compare the number of neurons in a human and a fly, for example, the reason seems obvious — with a quarter of a million neurons, the fly is no match to the 86 billion neurons in our brain [3Herculano-Houzel S. The remarkable, yet not extraordinary, human brain as a scaled-up primate brain and its associated cost.Proc. Natl. Acad. Sci. USA. 2012; 109: 10661-10668Crossref PubMed Scopus (324) Google Scholar]. But if you now consider a chimp, it is a different story because the chimp’s brain is about one third the size of ours. This is comparable (just a bit smaller) to the difference between the chimp’s brain and that of a macaque monkey — the primate species that is typically used in lab experiments — and although chimps are clearly more intelligent than macaques, there is nothing like the huge gap between chimps and ourselves. So, number of neurons (or number of neurons in a specific area) cannot be the only difference. There has to be something more. To further illustrate this point, take a neural network performing some task. Now consider another one that is three times larger. You would not expect to see such a massive difference in performance as the one you see between humans and chimps — with the larger network proving theorems and solving integrals, and the smaller one hardly able to count. You could still argue that a large difference might be enforced by how the networks are trained or other implementation details. But that is exactly the point. The networks are not different just because of their size, but rather by how they work. Back to our problem. There is a comparable number (and type) of neurons in the chimp and the human brain, and both species have more or less the same anatomical structures. Therefore, our neurons, or at least some of them, must be doing something different, and one such difference is given by how they store our memories. Some patients suffering from epilepsy cannot be treated with medication and are candidates for a surgical procedure in which the epileptic focus is removed. Before this surgery, intracranial electrodes may be implanted to accurately localize the epileptogenic area, which allow us to record the activity of dozens of individual neurons while the patients perform different tasks [4Quian Quiroga R. Plugging in to human memory: advantages, challenges and insights from human single neuron recordings.Cell. 2019; 179: 1015-1032Abstract Full Text Full Text PDF PubMed Scopus (21) Google Scholar]. In one of these cases, we showed the patient images of different people, places, animals and objects, and found a neuron firing selectively to completely different pictures of Jennifer Aniston (Figure 1A) [5Quian Quiroga R. Reddy L. Kreiman G. Koch C. Fried I. Invariant visual representation by single neurons in the human brain.Nature. 2005; 435: 1102-1107Crossref PubMed Scopus (1136) Google Scholar]. So, the neuron did not fire to one or another picture, but to her: to the concept ‘Jennifer Aniston’. Similarly, another neuron fired to pictures of Halle Berry, another to Oprah Winfrey, and yet another to Luke Skywalker — the cells even responding to the written and spoken names of their target subjects and not to other names [6Quian Quiroga R. Kraskov A. Koch C. Fried I. Explicit encoding of multimodal percepts by single neurons in the human brain.Curr. Biol. 2009; 19: 1308-1313Abstract Full Text Full Text PDF PubMed Scopus (127) Google Scholar]. These neurons, which we named ‘Concept Cells’, are located in the hippocampus and surrounding cortex, an area that it is typically involved in epilepsy — and is therefore targeted with the intracranial electrodes — and that is also critical for episodic memory, the memory of our lifetime experiences [7Moscovitch M. Cabeza R. Winocur G. Nadel L. Episodic memory and beyond: the hippocampus and neocortex in transformation.Annu. Rev. Psychol. 2016; 67: 105-134Crossref PubMed Scopus (498) Google Scholar]. This has been clearly established from the study of Henry Molaison (the famous patient H.M.), who had his hippocampus surgically removed and subsequently couldn’t form or retrieve episodic memories [8Squire L. The legacy of patient H.M. for neuroscience.Neuron. 2009; 61: 6-9Abstract Full Text Full Text PDF PubMed Scopus (215) Google Scholar]. But why do we have neurons firing to specific concepts in a memory area? Because that’s the way we store our memories. We tend to remember concepts and associations between them and to forget irrelevant details — so, we’ll recall seeing pictures of Jennifer Aniston at the hospital ward, for instance, but won’t remember details of the specific images. To show that Concept Cells are indeed involved in memory, we designed a simple experiment. First, in a ‘screening session’ we showed about 100 pictures and determined which pictures (typically of famous persons) triggered a response in any of the recorded neurons. Then, we created an association between the persons to which the neurons initially fired and arbitrary places. Say we had a neuron firing to Jennifer Aniston; we then showed a picture of her in the Eiffel Tower (created using Photoshop) and found that the neuron started firing to the Eiffel Tower as well, without Aniston needing to be in the picture (Figure 1B), at the exact moment the patient learned the association [9Ison M. Quian Quiroga R. Fried I. Rapid encoding of new memories by individual neurons in the human brain.Neuron. 2015; 87: 220-230Abstract Full Text Full Text PDF PubMed Scopus (90) Google Scholar] — a proxy of the episodic memory of remembering seeing her in the Eiffel Tower. Furthermore, we found that in a passive viewing task (during the screening sessions) Concept Cells tend to respond to items that are related for the patients, thus providing a long-term coding of meaningful associations that supports episodic memory [10De Falco E. Ison M. Fried I. Quian Quiroga R. Long-term coding of personal and universal associations underlying the memory web in the human brain.Nat. Commun. 2016; 7: 13408Crossref PubMed Scopus (38) Google Scholar]. For example, the neuron initially firing to Jennifer Aniston also responded to Lisa Kudrow, a co-star of the TV series ‘Friends’ (but not to other actors of the same TV series), when tested again the next day, including more persons related to Aniston. Neurons like Concept Cells, firing in such a selective and abstract manner to specific persons/concepts, have so far not been found in other species. The closest to Concept Cells seem to be neurons in the anterior medial face patch in the monkey temporal lobe, which respond selectively to relatively few faces [11Tsao D.Y. Freiwald W.A. Tootell R.B. Livingstone M. A cortical region consisting entirely of face-selective cells.Science. 2006; 311: 670-674Crossref PubMed Scopus (791) Google Scholar]. A recent study [12Chang L. Tsao D. The code for facial identity in the primate brain.Cell. 2017; 169: 1013-1028Abstract Full Text Full Text PDF PubMed Scopus (288) Google Scholar], however, showed that these neurons do not actually code for specific individuals, but rather respond to complex visual features. In another study [13Sliwa J. Plante A. Duhamel J.R. Wirth S. Independent neuronal representation of facial and vocal identity in the monkey hippocampus and inferotemporal cortex.Cereb. Cortex. 2016; 26: 950-966Crossref PubMed Scopus (26) Google Scholar], now in the monkey hippocampus, researchers analysed responses to familiar faces — for example, faces of other monkeys in the colony, and so on — replicating the protocol used to find Concept Cells in humans, but did not find neurons with such a degree of selectivity and abstraction. Yet another study [14von Heimendahl M. Rao R. Brecht M. Weak and nondiscriminative responses to conspecifics in the rat hippocampus.J. Neurosci. 2012; 32: 2129-2141Crossref PubMed Scopus (44) Google Scholar] focused on hippocampal responses in rats while the animals interacted with conspecifics — as a proxy of the response a human has when seeing a familiar person — and found that no cell responded selectively to individual rats. It could still be argued that more experiments are needed to establish whether, and to what extent, a neural representation like the one by Concept Cells exists in other species. More such experiments would certainly be very useful to have good quantitative comparisons across species. But the argument that Concept Cells may be exclusively human is not just based on absence of evidence, because, in rats and monkeys, very strong neuronal responses are found in the same area but with a different type of coding: responses tend to change when altering the context or the task performed by the animals, whereas with Concept Cells they do not. For example, neurons in the rat hippocampus fire at specific locations of the environment [15Moser E. Moser M.-B. McNaughton B. Spatial representation in the hippocampal formation: a history.Nat. Neurosci. 2017; 20: 1448-1464Crossref PubMed Scopus (211) Google Scholar]. The locations to which these ‘Place Cells’ fire can be also seen as concepts, which are very salient for rats, to know their precise location and routes to reach safety. The key difference, however, is that Place Cells change their responses after even slight modifications of the environment or the task [16Eichenbaum H. Cohen N. Can we reconcile the declarative memory and spatial navigation views on hippocampal function?.Neuron. 2014; 83: 764-770Abstract Full Text Full Text PDF PubMed Scopus (334) Google Scholar,17Eichenbaum H. Dudchenko P. Wood E. Shapiro M. Tanila H. The hippocampus, memory, and place cells: is it spatial memory or a memory space?.Neuron. 1999; 23: 209-226Abstract Full Text Full Text PDF PubMed Scopus (788) Google Scholar]. Similarly, other studies have shown context and task modulation of hippocampal responses in monkeys [18Rolls E.T. Wirth S. Spatial representations in the primate hippocampus, and their functions in memory and navigation.Prog. Neurobiol. 2018; 171: 90-113Crossref PubMed Scopus (78) Google Scholar,19Gulli R. Duong L. Corrigan B. Doucet G. Williams S. Fusi S. Martinez-Trujillo J. Context-dependent representations of objects and space in the primate hippocampus during virtual navigation.Nat. Neurosci. 2020; 23: 103-112Crossref PubMed Scopus (34) Google Scholar]. In contrast, Concept Cells fire to a particular concept in completely different situations: irrespective of whether the subject is passively looking at a picture of a person, seeing morphed versions of it in a recognition task, recalling its presentation from memory, or learning associations, as in the experiment described above, where the person to which the neuron fired was shown on their own or in a specific location. Memories shape our thoughts and the way we store them should no doubt have an impact on our cognitive abilities. What, then, are the implications of a coding scheme like the one by Concept Cells in humans? An obvious advantage is that with such context-independent representation it is trivial to generalize, to make analogies and transfer knowledge, as the same neuronal representation of concepts applies to different situations. This contrasts with the difficulty that rats have in performing the same tasks when the environment or overall context is altered, given the change in the neuronal representation. An explicit and abstract representation, devoid of meaningless details and context, also facilitates rapidly establishing associations between disparate concepts, which is the basis of our imagination and creativity — for example, to realize that two disparate things, like an apple falling from the tree and the moon orbiting around the Earth, respond to the same phenomenon: the law of gravity. But to make these associations we need to leave aside meaningless details; it doesn’t matter if the apple is big or small, red or green, or if the moon is crescent, in its zenith, and so on. Abstract representations, such as the ones provided by Concept Cells, are ideal for generalisations and for building nested and elaborated thoughts — to go beyond the particular things present in our immediate surroundings and think about thoughts, about concepts and their relationships. This is the basis of the unique workings of our brain, perhaps a key component of our intelligence and what distinguishes us from other species.