Genes, Brain, and Mind: the Evolution of Cognition
1998; Cell Press; Volume: 20; Issue: 6 Linguagem: Inglês
10.1016/s0896-6273(00)80489-4
ISSN1097-4199
Autores Tópico(s)Plant and Biological Electrophysiology Studies
ResumoNeuroscience, with its multiple levels of investigation from gene to behavior and mentation, has long searched for a conceptual framework that integrates mechanisms across functional levels. Not surprisingly, no satisfactory large-scale theory of brain and mind has yet been formulated. At a recent conference, “Evolution of Brain and Mind” (Human Frontiers Science Program, Strasbourg, France, November, 1997), the elaboration of an evolutionary context challenged many traditional views of brain function and suggested novel research strategies. Participants reexamined evidence for innate versus learned function, emphasized the unique computational designs of different brain subsystems, explored the nature of learning, and described ongoing efforts to model adaptive cognitive systems. Cognition and brain function were framed in the context of evolutionary constraints. Characterization of complex cognitive capacities helped clarify the organization of cognition, the roles of innate and learned function, and the role of natural language in reasoning. Brain mechanisms underlying cognition were discussed, including the cellular and molecular mechanisms of cortical evolution, the potential origins of plasticity and learning in evolved gene structure, the emergence of a primitive theory of mind in primate cortical neurons and networks that may contribute to consciousness. Historically, cognitive psychology has been heavily influenced by the intellectual current of the moment, whether turn-of-the century nativist, stressing innate capabilities, mid–twentieth century behaviorist, emphasizing environmental determination, or late twentieth century connectionist, stressing the primacy of synaptic plasticity. Eschewing efforts to identify a single process underlying cognition, conference discussions emphasized that the brain is not an all-purpose cognitive machine, employing a single, common computational strategy. Reports converged to indicate that multiple, discrete brain systems employing unique, domain-specific computations have evolved to solve different evolutionary problems. The search for a set of purely associational mechanisms for learning, motivated by the British empiricist position, may be ill-framed. An association is simply a signal-conducting connection between elements, whether notions, neurons, or nodes in a computer network. The association occurs, for example, because two units have been active simultaneously (temporal pairing), or through feedback mechanisms that reduce output errors by adjusting associative strengths. Associational theories of learning assume that altered strength of associations is the basic learning mechanism, and that the association itself is the memory. As a paradigmatic example of the heterogeneity of learning mechanisms, Gallistel contrasted the computational demands of spatial learning with those required for classical conditioning (14Gallistel C.R The Organization of Learning. Bradford Books/MIT Press, Cambridge, MA1990Google Scholar, 15Gallistel C.R Cramer A.E Computations on metric maps in mammals getting oriented and choosing a multi-destination route.J. Exp. Biol. 1996; 199: 211-217PubMed Google Scholar). Spatial memory gets us from train to home and bedroom to bathroom. It also gets the foraging ant or bee to a food source via exploratory, circuitous routes but back home by astoundingly direct paths. Spatial learning involves mapping of the environment onto a common geocentric framework. Spatial learning requires computation of egocentric (personal) velocities, distances, and angles, internalization of the geocentric directional framework (map), transformation of egocentric vectors to geocentric vectors, positional updating, and course setting. All of these operations require the ant's nervous system to store the value of a variable over time. For the ant, the deduced calculations appear to be accomplished in part by dead-reckoning, which involves summing the successive small changes in position to get the final net change in position. When the decision is made to head home, the positional values stored in the “dead-reckoning integrator” are used to navigate the straight run home. While this extended procedure comprises the obligatory minimal operation set for ant navigation, actual mediating mechanisms in the ant nervous system have yet to be identified. The foraging bee provides even more direct evidence that it has remembered the positional coordinates of a food source for subsequent use. On returning to the hive, the bee performs the famous waggle dance, indicating the direction of the food source relative to the sun and the approximate distance of the source from the hive (47Von Frisch K Lindauer M Schmiedler Wie erkennt die Biene den Sonnenstand bei geschlossener Wolkendecke?.Naturwissenschäftliche Rundschau. 1960; 13: 169-172Google Scholar). Bee and ant thus exhibit highly specialized computational mechanisms specifically adapted to solve spatial learning and memory problems. Classical conditioning, by contrast, is computationally distinct. How does the brain decide which cues predict an event? Conditioning involves a nonstationary, multivariate time series analysis to determine which stimuli predict which rates of reward (14Gallistel C.R The Organization of Learning. Bradford Books/MIT Press, Cambridge, MA1990Google Scholar). The analysis is multivariate because there are many possible predictors, and experiments indicate that the predictive power of one stimulus profoundly affects conditioning to other stimuli. The nonstationary aspect complicates the problem: relationships in the real world change, as in extinction, for example. Conditioning involves measuring temporal intervals, counting events, converting counts and intervals from different contexts into rate estimates, and testing for changes in these rate estimates—computations strikingly different from those required for spatial learning. Associative theories assumed that the temporal interval between CS onset (conditioned stimulus, such as a tone) and US onset (unconditioned stimulus, such as food or foot shock) determines the rate of conditioning. However, it has been shown that the CS–US interval has no effect on rate, if the US–US interval is adjusted to the change in CS–US interval (16Gibbon J Baldock M.D Locurto C.M Gold L Terrace H.S Trial and intertrial durations in autoshaping.J. Exp. Psychol. Anim. Behav. Process. 1977; 3: 264-284Crossref Scopus (198) Google Scholar). This is a clear violation of the associationist account but a straightforward prediction of models in which conditioning depends on comparing the expected interreward interval during the CS to the expected interval in its absence (14Gallistel C.R The Organization of Learning. Bradford Books/MIT Press, Cambridge, MA1990Google Scholar). New successful, simple models of conditioning are based on mechanisms that record, store, and manipulate durations of temporal intervals. Gallistel's extensive analysis of classical conditioning studies indicates that conditioned stimulus–reward relationships are timescale invariant. When a US is given in the presence of background stimuli alone, the animal associates the US with the background, and learning that would have occurred on pairing with a conditioned stimulus is blocked (10Fanselow M.S Pavlovian conditioning, negative feedback, and blocking mechanisms that regulate association formation.Neuron. 1998; 20: 625-627Abstract Full Text Full Text PDF PubMed Scopus (116) Google Scholar). The associationist account of this critical phenomenon requires an internal trial clock (38Rescorla, R.A., and Wagner, A.R. (1972). A theory of Pavlovian conditioning: variations in the effectiveness of reinforcement and nonreinforcement. In Classical Conditioning II, A.H. Black and W.F. Prokasy, eds. (New York: Appleton-Century-Crofts), pp. 64–99.Google Scholar), for which evidence is lacking. The interval timing explanation simply assumes that the animal learns the inter-US interval that occurs with background stimuli alone, and exhibits no response to another CS, if the inter-US interval in the presence of CS and background is the same as that for background alone. Thus, the absolute interval from stimulus onset to reinforcement is irrelevant. It is the ratios between expected intervals that are relevant, and these ratios are unaffected by temporal rescaling (lengthening or shortening of all intervals by a common scale factor). Timescale invariance is fatal for a simple associative temporal pairing of stimulus and reward as a solution to classical conditioning. The computations required to “solve” spatial learning and conditioning are different because the mathematics of the problem domains are different. Each has a unique problem structure necessitating different computations. Classical conditioning effectively requires the solution of simultaneous equations. In contrast, spatial learning requires vector transformation in the spatial domain. Rather than searching for a single associational mechanism that underlies all learning, Gallistel argues for characterizing the mathematics of the problem domain to allow definition of the neural computations that were selected to solve the evolutionary task. The clear implication is that very different brain systems, incorporating different computational design features, underlie spatial learning and classical conditioning. This raises the question of whether the extensively studied phenomenon of long-term potentiation (LTP) may model only certain types of learning and memory. In LTP, and by implication, many forms of learning, the altered association is a change in synaptic strength, and the underlying molecular cascade is hypothesized to be the mechanism mediating multiple memory types. The characteristic synaptic associativity, however, may underlie only certain types of memory but not participate in other forms of learning, according to many conferees. What are the implications of domain specificity for traditional neuroscientific approaches to learning? One set of questions concerns the “levels of analysis” problem. For example, can the computational alphabet provided by the associative, Hebbian synapse constitute a basis for the higher order computational combinations and permutations required for radically different types of learning? Is the contradiction in computational strategies, therefore, more apparent than real? Is it more appropriate instead to envisage a hierarchy of computational complexity, in which molecular mechanisms at the synapse, for example LTP and LTD, underlie the increasingly complex computations identifiable at the task domain-specific level? In this view, the adaptations that mediate specialized information acquisition are built upon computational mechanisms common to all learning. Identification of these elemental computations and the molecular and cellular mechanisms mediating them constitutes a critical research program. Delineation of combinatorially integrated specialized learning at the cell and systems levels becomes a tractable, though formidable, problem. More generally, does the nervous system actually perform the idealized foregoing computations—are there actually physical instantiations of the deduced computations that are theoretically capable of solving the foregoing learning problems? These questions may now be solvable; functional neuroanatomy, whole cell and single channel physiology, targeted gene deletion, and behavioral studies may begin to explicitly define mechanisms governing the putative computations involved in many domain-specific tasks. The evolutionary, domain-specific, cognitive framework places emphasis on how systems develop, the roles of innate versus learned function, the role of natural language in cognition, and the unique computational capacities of different systems. Analysis of mathematical reasoning provided a paradigm for examining these issues explicitly and implicitly. Carey presented persuasive evidence that mathematical reasoning is present in preverbal infants and primates and may be innate and not learned (49Xu F Carey S Infants' metaphysics the case of numerical identity.Cogn. Psychol. 1996; 30: 111-153Crossref PubMed Scopus (486) Google Scholar). To examine primates and prelinguistic infants, Carey used the violation-of-expectancy looking-time method (42Spelke, E.S. (1985). Preferential looking methods as tools for the study of cognition in infancy. In Measurement of Audition and Vision in the First Year of Postnatal Life, G. Gottlieb and N. Krasnegor, eds. (Hillsdale, NJ: Lawrence Erlbaum Associates), pp. 323–364.Google Scholar): subjects reliably look longer at events that are physically or conceptually impossible than at events that do not violate natural laws. In one experiment with infants as young as 5 months, an object was retrieved from behind one screen and replaced, and a second object was retrieved and replaced behind a second screen. Nothing passed between the screens. When the screens were removed, babies reproducibly looked longer at the unexpected result of only a single object than at the expected result of two objects. With appropriate controls and permutations, it is apparent that infants “know” that objects trace spatiotemporally continuous paths; the infants possess criteria for individuation of objects and numerical identity. Moreover, they distinguish “one” from “another” and represent “same one.” These numerical representations are available prelinguistically; they do not depend on language. This numerosity is apparently not simply based on distinctions between “more” and “less,” since infants do not distinguish between nonindividuated masses, such as large and small piles of sand or quantities of water. Infants completely fail numerosity tests with sand piles but exhibit numerical competence when the sand is individuated into discrete objects. Analogous experiments in adult macaques indicate distinctions between “one” and “another” and representation of “same one.” Apparently, numerosity did not emerge evolutionarily as part of the human specialization for language. Adult primates and 5-month-old infants represent the numbers of small collections of objects and the numerical relations among them, such as two objects plus one object equals 3 objects, or 2 objects minus 1 object equals 1 object, or 3 objects are more than 2 objects. These observations leave open, however, the format that nonlinguistic creatures use to represent such numerical facts. Carey argued that it is unlikely that human infants' representations of small sets of objects contain explicit symbols for integers. Rather, they are more likely to resemble the representations of number in first order logic. Nonetheless, by 5 months of age, the rudiments of mathematical reasoning are evident. These observations contradict the view of 35Quine W.V.O Word and Object. MIT Press, Cambridge, MA1960Google Scholar and 33Piaget J The Child's Construction of Reality. Routledge and Kegan Paul, London1955Google Scholar that ontological constructs such as object and number are cultural and depend upon mastery of the syntax of natural language. The robust violation-of-expectation method may be applicable to investigations of the evolution of a number of cognitive capacities in both humans and primates and may be used to examine the issue of domain specificity itself. In principle, the method can be used to compare the developmental profiles of the capacity for spatial learning, classical conditioning, and numerosity. Differences and/or similarities in capability onset and change with age may help define the nature of domain specificity versus generality of these cognitive abilities. Combination with newer imaging techniques, including functional MRI, may help localize participating brain areas. The results of Carey and others indicate that systems exhibit computational capacities during early infancy, raising the possibility that they are innate. Experience presumably builds upon these endowments. The work also raises questions about appropriate computer-based models of cognition and its development. If different neural systems express entirely different intrinsic computational abilities during development, the use of totally open, neutral, unbiased networks to model brain system development may be misguided. Rather, attention should be focused on how a network mimicking language may differ from that subserving spatial memory or conditioning. Spatial learning, conditioning, and mathematical reasoning were presented as evidence for the domain-specific organization of cognition. Singer used another domain-specific function, visual perception, to approach the neurobiology of consciousness from an evolutionary perspective. How is visual percept coherence achieved? Do the same cortical mechanisms underly consciousness? Consciousness and related cognitive abilities are widely regarded as requisites for socialization and the emergence of culture. Consideration of the elusive subject of consciousness necessitates rigorous delineation of specific phenomena to be examined and equally strict exclusion of inappropriate areas. Discussion focused on phenomenal awareness of visual sensation, the ability to report about internal brain states, and the ability to signal others about these internal states. In contrast, evaluations such as self-awareness and individuality were regarded as culturally derived and not presently amenable to neuroscientific analysis. According to Singer's hypothesis, consciousness requires the re-representation of items, such as primary sensory features, to attain the level of phenomenal awareness. In Singer's formulation, consciousness requires the dynamic association of neurons, each encoding a small subset of features, which as a whole represent a full feature constellation. The temporary binding of neurons into an ensemble is the representation that reaches awareness through selective attention (41Singer W Engel A.K Kreiter A.K Munk M.H.J Neuenschwander S Roelfsema P.R Neuronal assemblies necessity, signature, and detectability.Trends Cogn. Sci. 1997; 1: 252-261Abstract Full Text PDF PubMed Scopus (195) Google Scholar). The brain, of course, is already known to combine features through an entirely different strategy, the progressive, serial combination of features in a hierarchical fashion, leading to increasingly complex representations in a small group of neurons. The extreme version of this strategy is exemplified by the famous “grandmother cell” model, in which the neuron at the top of the hierarchy presumably recognizes our grandmothers. In fact, cells that recognize whole faces have been identified, and hierarchical organization is a well-defined feature of visual percept formation (e.g.,19Hubel D.H Wiesel T.N Functional architecture of macaque monkey visual cortex.Proc. R. Soc. Lond. B Biol. Sci. 1977; 198: 1-59Crossref PubMed Google Scholar). The role of ensemble formation, while theoretically attractive, has yet to be documented. Singer's work suggests that ensemble formation occurs through synchronous discharge of a neuronal population. Using the visual system as an example, Singer presented a cogent description of the occurrence and advantages of a synchronous discharge strategy. Synchronization is extremely rapid, allowing an individual neuron to participate in one assembly for several milliseconds and then switch to another. Second, transmission of synchronous discharge does not require temporal summation. Finally, synchronous input yields output with little temporal dispersion, permitting a high fidelity signature of relatedness among neural processing levels. Examination of the roles, if any, of attentional systems in synchrony have been initiated only recently. One early study indicates that stimulation of the attentional mesencephalic reticular formation enhances stimulus-specific synchronization in cat visual cortex (31Munk M.H.J Roelfsema P.R König P Engel A.K Singer W Role of reticular activation in the modulation of intracortical synchronization.Science. 1996; 272: 271-274Crossref PubMed Scopus (384) Google Scholar). It is unclear, however, whether separate attentional systems are required to be aware of representation in an ensemble. As well recognized by Singer, the crux requirement for the synchronous binding model has yet to be fulfilled: evidence for a functional role. The existence of response synchronization is well documented (41Singer W Engel A.K Kreiter A.K Munk M.H.J Neuenschwander S Roelfsema P.R Neuronal assemblies necessity, signature, and detectability.Trends Cogn. Sci. 1997; 1: 252-261Abstract Full Text PDF PubMed Scopus (195) Google Scholar), although oscillations are frequently absent in appropriate areas during perception (23Kiper D.C Gegenfurtner K.R Movshon J.A Cortical oscillatory responses do not affect visual segmentation.Vision Res. 1996; 36: 539-544Crossref PubMed Scopus (89) Google Scholar). On the other hand, synchronization in cat visual cortex, for example, correlates with perception in interocular rivalry (12Fries P Roelfsema P.R Engel A.K König P Singer W Synchronization of oscillatory responses in visual cortex correlates with perception in interocular rivalry.Proc. Natl. Acad. Sci. USA. 1997; 94: 12699-12704Crossref PubMed Scopus (394) Google Scholar). Presently, however, there is scant evidence, if any, that synchrony actually does bind features into percepts. Simply posed, does perturbation of correlated discharge actually interfere with perception? In one of the few such studies performed, picrotoxin, a GABAA antagonist, was used in honeybees to disrupt oscillatory synchrony, while leaving firing rates intact (43Stopfer M Bhagavan S Smith B.H Laurent G Impaired odour discrimination on desynchronization of odour-encoding neural assemblies.Nature. 1997; 390: 70-74Crossref PubMed Scopus (729) Google Scholar). Olfactory perception was deranged: similar odors could no longer be distinguished. While subject to several interpretations, and although relying on a single pharmacological agent, the results suggest that perception was impaired in the absence of synchrony. A number of additional issues bear on the synchrony-binding-perception hypothesis. Gilbert and colleagues have shown that neurons of similar visual orientation in different cortical columns exhibit correlated firing (46Ts'o D.Y Gilbert C.D Wiesel T.N Relationships between horizontal interactions and functional architecture in cat striate cortex as revealed by cross-correlation analysis.J. Neurosci. 1986; 6: 1160-1170Crossref PubMed Google Scholar). Consequently, the principles of synchrony may be defined by architecture and connectivity leading to binding and not by external stimuli per se. At the very least, it has proven difficult to disentangle the contributions of circuits and stimuli. In other experiments, this group has demonstrated that distributed visual cells respond to salient discontinuous contours by increasing firing rates, not synchrony, and that firing rate, not correlated firing, can account for perception (46Ts'o D.Y Gilbert C.D Wiesel T.N Relationships between horizontal interactions and functional architecture in cat striate cortex as revealed by cross-correlation analysis.J. Neurosci. 1986; 6: 1160-1170Crossref PubMed Google Scholar, 45Ts'o D.Y Gilbert C.D The organization of chromatic and spatial interactions in the primate striate cortex.J. Neurosci. 1988; 8: 1712-1727Crossref PubMed Google Scholar). Finally, a number of experiments have failed to associate synchrony with appropriate perceptual tasks (9Fahle M Koch C Spatial displacement, but not temporal asynchrony, destroys figural binding.Vision Res. 1995; 35: 491-494Crossref PubMed Scopus (81) Google Scholar, 23Kiper D.C Gegenfurtner K.R Movshon J.A Cortical oscillatory responses do not affect visual segmentation.Vision Res. 1996; 36: 539-544Crossref PubMed Scopus (89) Google Scholar). The ongoing debates in this fertile area are part of a larger context concerning the role of temporal codes in cognition. How much information is actually encoded in the temporal pattern of firing of individual neurons and their populations? It has been estimated, for example, that only 20% of the useful information is incorporated in temporal patterns and 80% in rate codes (1Abbott L.F Rolls E.T Tovee M.J Representational capacity of face coding in monkeys.Cereb. Cortex. 1996; 6: 498-505Crossref PubMed Scopus (82) Google Scholar). In this sense, Singer's work bears upon larger, well-recognized issues concerning how information is represented in the nervous system. The evolution of phenomenal awareness allowed the evolution of another critical cognitive capacity, the appreciation that others also possess consciousness. A “theory of mind” constitutes the seminal insight that others are also endowed with a psychology of motivations, plans, actions, and reactions. This knowledge shapes primate socialization, alters the nature of communication with others, and forms a basis for the emergence of culture, including the competitive dynamics of social interactions, that characterize human affairs. What systems underlie a “theory of mind”? Rizzolatti and colleagues have been performing single unit recordings in a region of area F5 in the ventral premotor cortex of macaques. Based on common gross architectural features, cytoarchitectonics, connectivity to common cortical areas, and roles in voluntary motor actions, the region is a potential homolog of Broca's language area in humans (39Rizzolatti G Arbib M.A Language within our grasp.Trends Neurosci. 1998; 21: 188-194Abstract Full Text Full Text PDF PubMed Scopus (2040) Google Scholar). Individual neurons in this area respond only when the monkey is grasping with the hand. The responses are reproducible, do not habituate, and depend on grasping itself, not on the anticipation of food or other artifacts. Remarkably, the monkey's neurons also respond when the human keeper is observed to grasp an object with his hand. Consequently, these “mirror neurons” represent both motor actions of the monkey and actions of other individuals. What ongoing functions do the mirror neurons serve for the monkey, and how does this lead to a theory of mind? Rizzolatti suggests that the neurons play roles in “understanding motor events” (40Rizzolatti G Fadiga L Gallese V Fogassi L Premotor cortex and the recognition of motor actions.Cogn. Brain Res. 1996; 3: 131-141Crossref PubMed Scopus (3241) Google Scholar). Mirror neurons recognize specific actions of others and map them onto self-generated actions. Motor activity alters the individual's relation to the outside world, and these consequences are monitored by the senses and remembered. Execution of movements are routinely associated with predictions of consequences: movements have meanings. The meaning of an observed action arises from matching it with a self-executed action. By similarly representing others' movement and one's own, mirror neurons can attribute meaning to motor activity of others. This attribution of a psychology to another's motor behavior represents a primitive theory of mind. These observations indicate that the neuronal apparatus potentially underlying a theory of mind is available for analysis of mechanism and for manipulation. Although the primate system functions in the absence of language, the proximity to, or even identity with, Broca's speech area in humans raises provocative questions regarding the relationship of the observation-matching system to the evolution of communication and language itself (39Rizzolatti G Arbib M.A Language within our grasp.Trends Neurosci. 1998; 21: 188-194Abstract Full Text Full Text PDF PubMed Scopus (2040) Google Scholar). One of the goals of the meeting was to explore the structural substrates for the evolution of cognitive capacities. How did the brain evolve to subserve these specialized functions? Krubitzer focused on mammalian cortex, noting commonalities and the evolved differences that may underlie cognitive functions (25Krubitzer L The organization of neocortex in mammals are species differences really so different?.Trends Neurosci. 1995; 18: 408-417Abstract Full Text PDF PubMed Scopus (243) Google Scholar). All mammalian cortices are divided into multiple functional subunits exhibiting common sensory stimulus preferences; the subunits are characterized by unique connections and distinct cytoarchitectonics. Organizational differences among functionally related cortices are also apparent. For example, the location, size, and shape of cortical fields change across species. Adaptation to specialized lifestyles, often consisting of modified peripheral receptor/effector morphology, is accompanied by striking alterations in cortical organization. The arboreal squirrel, for example, which relies heavily on vision, devotes half the cortical surface to visual processing; visual subdivisions are both larger and more numerous, and somatosensory and auditory cortex is reduced (21Kaas J.H Krubitzer L.A Johanson K.L Cortical connections of areas 17 (V-I) and 18 (V-II) of squirrels.J. Comp. Neurol. 1989; 281: 426-446Crossref PubMed Scopus (77) Google Scholar). In contrast, terrestrial hamsters, mice, and rats exhibit marked reduction in their relatively simple visual cortices (48Wagor E Mangini N.J Pearlman A.L Retinotopic organization of striate and extrastriate visual cortex in the mouse.J. Comp. Neurol. 1980; 193: 187-202Crossref PubMed Scopus (194) Google Scholar), with markedly expanded somatosensory cortex (5Chapin J.K Lin C.S Mapping the body representation in the SI cortex of anesthetized and awake rats.J. Comp. Neurol. 1984; 229: 199-213Crossref PubMed Scopus (494) Google Scholar). Approximately half of this large field is dedicated to the vibrissae barrel system, which is central
Referência(s)