Alessio Plebe

The chapter explores the challenges posed by the proliferation of robots in society, particularly humanoid robots, focusing on cognitive constraints, ergonomic concerns, cross-cultural issues and the emergence of an artificial morality in human–robot interactions. It also delves into the distinctions between physical and intentional trust, highlighting the role of deference, both biologically grounded and epistemic, in human–robot relationships. The concept of selective deference is introduced as a means of navigating interactions with robots, emphasizing the importance of balancing trust and deference based on the context and the technology involved. Overall, the analysis underscores the need for a nuanced understanding of trust and deference in the evolving landscape of human–robot interactions.

This chapter attempts to move from the fundamental computational properties of the brain, previously described, into a sketch of how the brain builds a representation of the world. The next part of the book will deal more specifically with the linguistic portrayal we humans have of the world, a topic which has also served as a tentative approach to explaining the neural mechanisms that allow animals to build knowledge. In a sense, the core question of epistemology. Several philosophers, like Jerry Fodor (1983, 1990), have denied that an explanation of representations can be given in terms of neural biophysical properties. Today, such a radical refusal of the neurocomputational approach has become more and more marginal, in any case its thorough defense is beyond the scope of this book, and left to better advocates (Churchland 2002). On the other hand, any endeavor toward explaining mental representations by neural mechanisms, has first to acknowledge that the notion of “representation” itself is problematic, and at the heart of current philsophical controversies. In addition, there are also positions entirely within a neuroscientific perspective, that deny the concept of representations completely. The title of this chapter leaves no doubt that we, instead, appeal to the notion of representations. It is out of the scope of this book the attempt to settle the philsophical debate on representations, and to lay down any new theory, our approach is to explore the emergence of semantic phenomena through the use of neurocomputational models based on a set of plausible mechanisms, that will be described in this chapter. Before that, we will offer a short overview of the philsophical issues posed by mental representation in general, with more detail on neural representation, and on neural computation over representations.We will offer a selection of a small number of mechanisms, deemed to be at the core of the bridge between electrochemical activity and world representation. The ability of detecting coincidences, at different scale levels of neural circuits, is regarded as the most general, and probably most effective mechanism. Other more specific strategies will be discussed such as receptive fields, topological organizations, and selective processing pathways.

Humans, like several other primates, are visual creatures, and almost half of our neurons are devoted to the processing of visual signals. The excellence found in our ability to do so, is not just due to our ophthalmological capabilities, which are outperformed by other species, such as birds, it is instead on the semantic side, in our ability to classify hundreds of object categories on the basis of their visual appearance only. Vision has historically been the earliest and most investigated function in the brain, thanks to its unique correspondence between the two dimensional organization of the distal stimulus and cortical processing units. Taken together, these two factors have led us to investigating the semantics of objects whose essential features are captured by their visual appearance. The first model presented in this chapter is a sort of prelude to a full blown semantics, with a simulation of the full visual pathway that brings light signals into recognition of object categories, together with the auditory pathway, in a simulation of the emergence of a first lexicon, that in infants begins exactly with visual objects. Most of the components of this model, and the methods used for its development and subsequent analyses, will be shared by the models that follow. The second model presented in this chapter, taps into a range of semantic phenomena typically observed in the early stages of language development in children, such as the change in the speed of learning, and the so called “fast-mapping” phenomenon.

Semantics, in the sense used in this book, seeks to understand the meaning of words and sentences, explaining the relations between expressions in a natural language and the world. This chapter intends to give a short account on semantics as has been developed before neurosemantics, and trace the path that naturally lead to its neuro form. The link that, in our view, connects the milestones of semantic theories to neurosemantics, is the aim of constructing precise mathematical models of the relations between linguistic entities and their referents. Therefore, much attention will be given to the transition from a descriptive semantics to its mathematical foundation in modern logic. It will be argued that the unsatisfactory aspect of this project was to have neglected the mind, which became on the contrary, the main object of investigation during the cognitive turn. We will describe the vicissitudes of cognitive semantics, its merits, and the counterside of a serious weakness in the level and robustness of its mathematical modeling.

This study investigates the potential use of current neural language models in psychological practice, particularly in diagnosing patients through the analysis of textual content. The growing interest in the capabilities of neural language models, especially in areas where natural language serves as a primary information source, and specifically by psychology, is the motivation behind this study. As a case study, we tackled the assessment of coherence in autobiographical narratives, a diagnostic methodology widely recognized in psychology for its pivotal role in understanding personal identity. This is a particularly intriguing case study for two reasons. On the one hand, it demonstrates how these tools can be a valid aid in psychological practice, in a task notoriously burdensome and time-consuming for human operators. On the other hand, it tests the ability of neural language models to grasp deep, global conceptual aspects from a text, a task not easily accomplished even by professional psychologists. Coherence is evaluated here quantitatively using the standard procedure known as the Narrative Coherence Coding Scheme. The results obtained are particularly promising, especially when using models subjected to fine-tuning with some examples of stories and their evaluations. However, even without fine-tuning, models prompted with appropriate prompts also perform acceptably.

This paper provides an epistemological and methodological analysis of the recent practice of using neural language models to simulate brain language processing. It is argued that, on the one hand, this practice can be understood as an instance of the traditional simulative method in artificial intelligence, following a mechanistic understanding of the mind; on the other hand, that it modifies the simulative method significantly. Firstly, neural language models are introduced; a study case showing how neural language models are being applied in cognitive neuroscience for simulative purposes is then presented; after recalling the main epistemological features of the simulative method in artificial intelligence, it is finally highlighted how the epistemic opacity of neural language models is tackled by using the brain itself to simulate the neural language model and to test hypotheses about it, in what is called here a co-simulation.

A long-standing question in language development research concerns the contrast between early word learning and the learning of adjectives. The difficulty children experience early on in the acquisition of color terms, for example, led Darwin to speculate that children are initially color blind. Adjectives, in fact, are almost entirely missing in early productive vocabularies across languages. Despite the accounts proposed to explain the phenomena the debate continues to be far from being resolved. We believe adjective learning requires the development of two basic abilities. The first one is at a syntactic level: the comprehension of utterances with adjectives entails a proto-syntactic ability in discriminating the different roles of two words, by virtue of their sequence. The second is at a semantic level, which entails understanding the predicative function of the adjective, against the more basic mapping of sounds with whole objects. Both abilities develop as a result of the synergy between learning from progressive exposure to a language rich environment and the maturation of neural structures. The “visual diet” also influences the type of adjectives learned, this being particularly pertinent to what color terms are learned. This chapter will discuss neurocomputational models that have in part simulated neural processes behind the learning of adjectives in linguistic development; how an initial sensitivity to word-order (leading to early syntactic learning or what we call proto-syntactic ability) might develop through linguistic exposure and brain maturational processes; and how both exposure to language as well as the range of colors dominant in a particular natural environment might influence not only how color terms are learned, but how colors are perceived, a topic that is right at the center of the historic linguistic relativity debate.

This paper sheds light on the shift that is taking place from the practice of ‘coding’, namely developing programs as conventional in the software community, to the practice of ‘curing’, an activity that has emerged in the last few years in Deep Learning (DL) and that amounts to curing the data regime to which a DL model is exposed during training. Initially, the curing paradigm is illustrated by means of a study-case on autonomous vehicles. Subsequently, the shift from coding to curing is analysed taking into consideration the epistemological notions, central in the philosophy of computer science, of function, implementation, and correctness. First, it is illustrated how, in the curing paradigm, the functions performed by the trained model depend much more on dataset curation rather than on the model algorithms which, in contrast with the coding paradigm, do not comply with requested specifications. Second, it is highlighted how DL models cannot be considered implementations according to any of the available definitions of implementation that follow an intentional theory of functions. Finally, it is argued that DL models cannot be evaluated in terms of their correctness but rather in their experimental computational validity.

: Autonomous driving systems promise important changes for future of transport, primarily through the reduction of road accidents. However, ethical concerns, in particular, two central issues, will be key to their successful development. First, situations of risk that involve inevitable harm to passengers and/or bystanders, in which some individuals must be sacrificed for the benefit of others. Secondly, and identification responsible parties and liabilities in the event of an accident. Our work addresses the first of these ethical problems. We are interested in investigating how humans respond to critical situations and what reactions they consider to be morally right or at least preferable to others. Our experimental approach relies on the trolley dilemma and knowledge gained from previous research on this. More specifically, our main purpose was to test the difference between what human drivers actually decide to do in an emergency situations whilst driving a realistic simulator and the moral choices they make when they pause to consider what they would do in the same situation and to better understand why these choices may differs. Keywords: Self-driving Cars; Trolley Problem; Moral Choices; Moral Responsibility; Virtual Reality Dilemmi morali nelle automobili a guida autonoma Riassunto: I sistemi di guida autonomi promettono importanti cambiamenti per il futuro dei trasporti, principalmente attraverso la riduzione degli incidenti stradali. Tuttavia, vi sono preoccupazioni etiche, in particolare due questioni centrali, fondamentali per il loro sviluppo. In primo luogo, le situazioni di rischio che comportano inevitabili danni ai passeggeri e/o ai pedoni, ovvero situazioni in cui alcune persone devono essere sacrificate a beneficio di altri. In secondo luogo, l’identificazione delle parti responsabili in caso di incidente. Il nostro lavoro affronta il primo di questi problemi etici. Siamo interessati a studiare come gli umani rispondono a situazioni critiche e quali reazioni considerano moralmente giuste o almeno preferibili. Il nostro approccio sperimentale si basa sul trolley problem e sulle conoscenze acquisite da precedenti ricerche su questo ambito. Più specificamente, il nostro scopo principale è quello di testare la differenza tra ciò che i conducenti umani decidono effettivamente di fare in una situazione di emergenza, mentre guidano un simulatore realistico, e le scelte morali che compiono se posti nella stessa situazione e hanno la possibilità di decidere senza limiti di tempo. Lo scopo è inoltre comprendere come e perché queste scelte possono differire. Parole chiave: Automobili a guida autonoma; Trolley problem; Scelte morali; Responsabilità morale, Realtà virtuale

In recent years, the family of algorithms collected under the term ``deep learning'' has revolutionized artificial intelligence, enabling machines to reach human-like performances in many complex cognitive tasks. Although deep learning models are grounded in the connectionist paradigm, their recent advances were basically developed with engineering goals in mind. Despite of their applied focus, deep learning models eventually seem fruitful for cognitive purposes. This can be thought as a kind of biological exaptation, where a physiological structure becomes applicable for a function different from that for which it was selected. In this paper, it will be argued that it is time for cognitive science to seriously come to terms with deep learning, and we try to spell out the reasons why this is the case. First, the path of the evolution of deep learning from the connectionist project is traced, demonstrating the remarkable continuity, and the differences as well. Then, it will be considered how deep learning models can be useful for many cognitive topics, especially those where it has achieved performance comparable to humans, from perception to language. It will be maintained that deep learning poses questions that cognitive sciences should try to answer. One of such questions is the reasons why deep convolutional models that are disembodied, inactive, unaware of context, and static, are by far the closest to the patterns of activation in the brain visual system.

In recent years, a number of different disciplines have begun to investigate the fundamental role context appears to play in a number of cognitive phenomena. Traditionally, linguistics, and the fields of communication and pragmatics in particular, have been the areas that have focused the most on contextual effects. Context has increasingly been studied for its role in influencing mental concepts, for some scholars being considered constitutive for most – if not all – concepts. Cognitive neuroscience is now starting to consider in a systematic way how context interacts with neural responses, although this research is still scattered and concentrated in a small number of specific cases only. In this chapter, we attempt to tie these three levels together, since only from their integration can a comprehensive explanation of how context affects cognition be constructed. The way context drives language comprehension depends on the effects of context on the conceptual scaffolding of the listener, which in turn, is the result of his neural responses in combination to context. These neural responses derive from learning throughout the history of experiences of the individual, and the association between possible contexts and heard utterances. The road we take to accomplishing the multi-level integration between what appear to be distant domains, is a computational one. This approach meets with the mechanistic framework of explanation, which is currently held as the most appropriate way of approaching cognitive phenomena that is often characterized by a multiplicity of levels, as is the case with context. The core underlying concept of the neurocomputational framework here proposed, is an account of neural representation, based on structural similarity. Structural representations are still the best option on the market in cognitive science, but in their traditional form, derived from classical measurement theory, are affected by a number of serious drawbacks, including not being able to account for context. We suggest a different account of structural similarity, one informed by current neuroscience, where the homomorphic relations required for structural similarity are derived from neural population coding. In a preliminary mathematical sketch, we indicate how this approach can construct neural aggegations that are sensitive to context.

This book examines the concept of “ Neurosemantics”, a term currently used in two different senses: the informational meaning of the physical processes in the neural circuits, and semantics in its classical sense, as the meaning of language, explained in terms of neural processes. The book explores this second sense of neurosemantics, yet in doing so, it addresses much of the first meaning as well. Divided into two parts, the book starts with a description and analysis of the mathematics of the brain, including computational units, representational mechanisms and algorithmic principles. This first part pays special attention to the neural architecture which has been used in developing models of neurosemantics. The second part of the book presents a collection of models, and describes each model reproducing specific aspects of the semantics of language. Some of these models target one of the core problems of semantics, the reference of nouns, and in particular of nouns with a strong perceptual characterization. Others address the semantics of predicates, with a detailed analysis of colour attributes. While this book represents a radical shift from traditional semantics, it still pursues a line of continuity that is based on the idea that meaning can be captured, and explained, by a sort of computation.

This paper analyzes the rapid and unexpected rise of deep learning within Artificial Intelligence and its applications. It tackles the possible reasons for this remarkable success, providing candidate paths towards a satisfactory explanation of why it works so well, at least in some domains. A historical account is given for the ups and downs, which have characterized neural networks research and its evolution from “shallow” to “deep” learning architectures. A precise account of “success” is given, in order to sieve out aspects pertaining to marketing or sociology of research, and the remaining aspects seem to certify a genuine value of deep learning, calling for explanation. The alleged two main propelling factors for deep learning, namely computing hardware performance and neuroscience findings, are scrutinized, and evaluated as relevant but insufficient for a comprehensive explanation. We review various attempts that have been made to provide mathematical foundations able to justify the efficiency of deep learning, and we deem this is the most promising road to follow, even if the current achievements are too scattered and relevant for very limited classes of deep neural models. The authors’ take is that most of what can explain the very nature of why deep learning works at all and even very well across so many domains of application is still to be understood and further research, which addresses the theoretical foundation of artificial learning, is still very much needed.

Neurosemantics is not yet a common term and in current neuroscience and philosophy it is used with two different sorts of objectives. One deals with the meaning of the electrical and the chemical activities going on in neural circuits. This way of using the term regards the project of explaining linguistic meaning in terms of the computations done by the brain. This book explores this second sense of neurosemantics, but in doing so, it will address much of the first as well, for we believe that the capacity of neural circuits to support linguistic meaning, hinges on their peculiar role in coding entities and facts of the world. It is an enterprise at the edge of the available state-of-the-art knowledge in neuroscience and specifically, in the growing understanding of brain computational mechanisms. We conceive neurosemantics, however, as the natural evolution of a long standing project that began in the early days of Boole’s logic, the idea that semantics can be construed and explained in mathematical terms. Classical formal semantics, for a very long time, excluded from the analysis of language any account of mental processes, which on the contrary, became the central focus during the cognitive turn. Cognitive semantics, however, failed to provide a rigorous mathematical framework for semantic processes. Today, it is possible to begin explaining language by way of a new mathematical foundation, one that is empirically grounded in how the brain computes: neurosemantics. The way this book intends to contribute is twofold. One is to present a series of existing examples of neurosemantics in practice: early models addressing aspects of linguistic semantics purely in neurocomputational terms. The other is to try to identify the principles upon which models of this kind can be constructed, and their corresponding neural bases.

The previous chapters have given samples of neurosemantics addressing specific semantic phenomena, using a unified neurocomputational approach. Much of the complexity of real language has been neglected, and in this chapter other developments will be presented, which fill in some of gaps that remain. The models presented in this chapter are not original developments of the authors, their selection is due to their theoretical grounds and their motivations, which are perfectly in line with the neurosemantics enterprise, as we have defined it here. In particular, Friedemann Pulvermüller and his associates have attempted to answer questions in semantics by developing neurocomputational models, based on brain representational mechanisms, as those here described in Chap. 3, and compatible with the organization of brain areas involved in language processing. Section 7.1 has shown how naturally the first instance of syntax emerges from sequences of adjectives and nouns. From that first step into a complete management of syntax, the brain needs to organize circuits to handle higher order combinatorial information, something simulated by Pulvermüller and called discrete combinatorial neuronal assemblies. With this computational architecture it is possible to explain main syntactic structure, such as that of verbal phrases. Much more is needed, but what has been achieved so far, the mathematical frameworks laid down, the definition of the research project, strongly suggest that neurosemantics today is feasible, and is one of the deeper and most appropriate efforts in explaining linguistic meaning.

As the final part of the semantics of neurons, and as a prelude to the second part, the neurosemantics of language, this chapter seeks mathematical formulations for the mechanisms that enable the construction of representations in the brain. It is not a general review of the rich variety of mathematical solutions proposed so far for simulating neural circuits, currently available on the market. It is the introduction to the mathematical framework adopted in all the neurosemantic models that will be described in the second part.One of the main challenges any endeavor of mathematical formalization of neural activities must face, is their impressive abundance. The number of neurons involved in almost all cognitive functions is so large that it is impossible to give an overall sense of their activity by means of individual descriptions. Mathematics only offers the great advantage of synthesis, the possibility of capturing in a concise formulation the principles ruling the behavior of millions of interacting elements. Mathematical formulations can be implemented in a software, and the simulated results can be analyzed in detail. In the cortex, neurons are characterized not only by their large number, but also by properties such as local cooperative and competitive interactions, which fit well within an established mathematical framework, that of self-organization. The adopted neural architecture derives from this general framework. In the interpretation of the activities of many neurons in the same cortical area, resulting from a simulation, a well established neurocomputational concept will be used, that of population coding, discussed in the last section of this chapter.

In all the aspects of semantics explored in the previous chapters, neural computation is today a fairly well established approach, with the models discussed not being the only ones. On the other hand, a neurocomputational approach has not been yet established for terms related to morality. The main reason is that empirical brain information on moral processing is still in its early infancy. As has been found with other aspects of word meaning, neuroscientific investigations such as those undertaken by Greene et al. (2001), have shown that there is a relatively consistent set of brain areas that are engaged during moral reasoning, as well as during emotional reactions and decision making in general (Casebeer and Churchland 2003; Moll et al. 2005). The models presented in this chapter can, therefore, be considered as pioneering works, or first steps in the enterprise of the neurosemantics of morality. The first model lacks linguistic capabilities and is a collection of a series of brain areas that learn the emotional component that contributes to the evaluation of potential actions, and then makes decisions. The second model combines the morally relevant areas of the first, with the auditory pathway that processes linguistic utterances, and simulates the emergence of the meaning of moral terms.

Every mathematical framework is developed around some basic computational item, for example, sets, numbers, and vectors. This is the starting point for a computational view of the brain as well, the basic units have to be specified. While in the ethereal world of mathematics the basic components can be arbitrarily assumed, even invented from scratch, the case of the brain is constrained by its biophysical structure. The current view today is dominated by the paradigm constructed by Ramón y Cajal, where the neuronal cell is the basic computational device of the brain. Enormous progress has been achieved in characterizing the computational properties of the brain under this paradigm, which will partly be reviewed in this chapter, limiting ourselves to those aspects that are useful for understanding the representational power of the brain. However, like any scientific paradigm, the so called “neuron dogma” might possibly change in the future, there are scholars for example (London and Häusser, Annu Rev Neurosci 28:503–5032, 2005) that argue that dendrites display autonomous computational capabilities, and might be a better candidate for the title of basic computational unit. Despite many attempts, it is still difficult to spell out what the most specific function of the neuron as a computational device is, and how that makes it so different from other man-made computational devices. We favor an idea advanced by Turing (1948, Intelligent machinery. Technical report, National Physical Laboratory, London, reprinted in Ince DC (ed) Collected Works of A. M. Turing: Mechanical Intelligence, Edinburgh University Press, 1969) long ago: neurons have no special built in function, but that of being able to learn virtually any function, by experience. Plasticity is the term in neuroscience that includes the biological mechanisms that explain how neurons work as extraordinary learning machines.The last section of this chapter deals with a special organization of neurons, that has long been held to deserve a specific computational description (Stevens, What form should a cortical theory take? In: Koch C, Davis J (eds) Large-scale neuronal theories of the brain. MIT, Cambridge, pp 239–255, 1994), and seems to be the privileged site of semantic representations: the cortex.

There is a diffuse sentiment that to anthropomorphize is a mild vice that people tend to do easily and pleasingly, but that an adult well educated person should avoid. In this paper it will be provided an elucidation of ``anthropomorphism'' in the field of common sense knowledge, the issue of animal rights, and about the use of humans as a model in the scientific explanation. It will be argued for a ``constructive anthropomorphism'', i.e., the idea that anthropomorphism is a natural attutude to attribute human psychological features to other individuals, no matter they are actually rational agents, or not. If we know the ``grammar'' of this attutude, we can avoid the risks in overestimating the environmental inputs towards anthropomor­phism and, at the same time, take the heuristic advantages of anthropomor­phism in the use of human mind as a model for both everyday circumstances and scientific enterprise.

This paper addresses a fundamental line of research in neuroscience: the identification of a putative neural processing core of the cerebral cortex, often claimed to be “canonical”. This “canonical” core would be shared by the entire cortex, and would explain why it is so powerful and diversified in tasks and functions, yet so uniform in architecture. The purpose of this paper is to analyze the search for canonical explanations over the past 40 years, discussing the theoretical frameworks informing this research. It will highlight a bias that, in my opinion, has limited the success of this research project, that of overlooking the dimension of cortical development. The earliest explanation of the cerebral cortex as canonical was attempted by David Marr, deriving putative cortical circuits from general mathematical laws, loosely following a deductive-nomological account. Although Marr’s theory turned out to be incorrect, one of its merits was to have put the issue of cortical circuit development at the top of his agenda. This aspect has been largely neglected in much of the research on canonical models that has followed. Models proposed in the 1980s were conceived as mechanistic. They identified a small number of components that interacted as a basic circuit, with each component defined as a function. More recent models have been presented as idealized canonical computations, distinct from mechanistic explanations, due to the lack of identifiable cortical components. Currently, the entire enterprise of coming up with a single canonical explanation has been criticized as being misguided, and the premise of the uniformity of the cortex has been strongly challenged. This debate is analyzed here. The legacy of the canonical circuit concept is reflected in both positive and negative ways in recent large-scale brain projects, such as the Human Brain Project. One positive aspect is that these projects might achieve the aim of producing detailed simulations of cortical electrical activity, a negative one regards whether they will be able to find ways of simulating how circuits actually develop.

In this paper we defend structural representations, more specifically neural structural representation. We are not alone in this, many are currently engaged in this endeavor. The direction we take, however, diverges from the main road, a road paved by the mathematical theory of measure that, in the 1970s, established homomorphism as the way to map empirical domains of things in the world to the codomain of numbers. By adopting the mind as codomain, this mapping became a boon for all those convinced that a representation system should bear similarities with what was being represented, but struggled to find a precise account of what such similarities mean. The euforia was brief, however, and soon homomorphism revealed itself to be affected by serious weaknesses, the primary one being that it included systems embarrassingly alien to representations. We find that the defense attempts that have followed, adopt strategies that share a common format: valid structural representations come as “homomorphism plus X”, with various “X”, provided in descriptive format only. Our alternative direction stems from the observation of the overlooked departure from homomorphism as used in the theory of measure and its later use in mental representations. In the former case, the codomain or the realm of numbers, is the most suited for developing theorems detailing the existence and uniqueness of homomorphism for a wide range of empirical domains. In the latter case, the codomain is the realm of the mind, possibly more vague and more ill-defined than the empirical domain itself. The time is ripe for articulating the mapping between represented domains and the mind in formal terms, by exploiting what is currently known about coding mechanisms in the brain. We provide a sketch of a possible development in this direction, one that adopts the theory of neural population coding as codomain. We will show that our framework is not only not in disagreement with the “plus X” proposals, but can lead to natural derivation of several of the “X”.

Neural plasticity has been invoked as a powerful argument against nativism. However, there is a line of argument, which is well exemplified by Pinker and more recently by Laurence and Margolis The conceptual mind: new directions in the study of concepts, MIT, Cambridge, 2015) with respect to concept nativism, according to which even extreme cases of plasticity show important innate constraints, so that one should rather speak of “constrained plasticity”. According to this view, cortical areas are not really equipotential, they perform instead different kinds of computation, follow essentially different learning rules, or have a fixed internal structure acting as a filter for specific categories of inputs. We intend to analyze this argument, in the light of a review of current neuroscientific literature on plasticity. Our conclusion is that Laurence and Margolis are right in their appeal to innate constraints on connectivity—a thesis that is nowadays welcome to both nativists and non-nativists —but there is little support for their claim of further innate differentiation between and within cortical areas. As we will show, there is instead strong evidence that the cortex is characterized by the indefinite repetition of substantially identical computational units, giving rise in any of its portions to Hebbian, input-dependent plasticity. Although this is entirely compatible with the existence of innate constraints on the brain’s connectivity, the cerebral cortex architecture based on a multiplicity of maps correlating with one another has important computational consequences, a point that has been underestimated by traditional connectionist approaches.