It is often indeterminate what function a given computational system computes. This phenomenon has been referred to as “computational indeterminacy” or “multiplicity of computations.” In this paper, we argue that what has typically been considered and referred to as the challenge of computational indeterminacy in fact subsumes two distinct phenomena, which are typically bundled together and should be teased apart. One kind of indeterminacy concerns a functional characterization of the system’s relevant behavior. Another kind concerns the manner in which the (...) abstract states are interpreted. We discuss the similarities and differences between the two kinds of computational indeterminacy, their implications for certain accounts of “computational individuation” in the literature, and their relevance to different levels of description within the computational system. We also examine the inter-relationships between our proposed accounts of the two kinds of indeterminacy and the main accounts of “computational implementation.”. (shrink)

For decades AI researchers have built agents that are capable of carrying out tasks that require human-level or human-like intelligence. During this time, questions of how these programs compared in kind to humans have surfaced and led to beneficial interdisciplinary discussions, but conceptual progress has been slower than technological progress. Within the past decade, the term agency has taken on new import as intelligent agents have become a noticeable part of our everyday lives. Research on autonomous vehicles and personal assistants (...) has expanded into private industry with new and increasingly capable products surfacing as a matter of routine. This wider use of AI technologies has raised questions about legal and moral agency at the highest levels of government (National Science and Technology Council 2016) and drawn the interest of other academic disciplines and the general public. Within this context, the notion of an intelligent agent in AI is too coarse and in need of refinement. We suggest that the space of AI agents can be subdivided into classes, where each class is defined by an associated degree of control. (shrink)

The ethics of emerging forms of artificial intelligence has become a prolific subject in both academic and public spheres. A great deal of these concerns flow from the need to ensure that these technologies do not cause harm—physical, emotional or otherwise—to the human agents with which they will interact. In the literature, this challenge has been met with the creation of artificial moral agents: embodied or virtual forms of artificial intelligence whose decision procedures are constrained by explicit normative principles, requiring (...) the implementation of what is commonly called artificial morality into these agents. To date, the types of reasoning structures and principles which inform artificial morality have been of two kinds: first, an ethically maximal vision of artificial morality which relies on the strict implementation of traditional moral theories such as Kantian deontology or Utilitarianism, and second, a more minimalist vision which applies stochastic AI techniques to large data sets of human moral preferences so as to illicit or intuit general principles and preferences for the design of artificial morality. Taken individually, each approach is unable to fully answer the challenge of producing inoffensive behavior in artificial moral agents, most especially since both forms are unable to strike a balance between the ideal set of constraints which morality imposes on one hand, and the types of constraints public acceptability imposes, on the other. We provide an alternative approach to the design of artificial morality, the Ethical Valence Theory, whose purpose is to accommodate this balance, and apply this approach to the case of autonomous vehicles. (shrink)

The use of "levels of abstraction" in philosophical analysis (levellism) has recently come under attack. In this paper, we argue that a refined version of epistemological levellism should be retained as a fundamental method, which we call the method of abstraction. After a brief introduction, in section two we make clear the nature and applicability of the (epistemological) method of levels of abstraction. In section three, we show the fruitfulness of the new method by applying it to five case studies: (...) the concept of agenthood, the Turing test, the definition of emergence, quantum observation and decidable observation. In section four, we further characterise and support the method by distinguishing it from three other forms of "levellism": (i) levels of organisation; (ii) levels of explanation and (iii) conceptual schemes. In this context, we also briefly address the problems of relativism and antirealism. In the conclusion, we indicate some of the work that lies ahead, two potential limitations of the method and some results that have already been obtained by applying the method to some long-standing philosophical problems. (shrink)

I love information upon all subjects that come in my way, and especially upon those that are most important. Thus boldly declares Euphranor, one of the defenders of Christian faith in Berkley’s Alciphron (Berkeley, (1732), Dialogue 1, Section 5, Paragraph 6/10). Evidently, information has been an object of philosophical desire for some time, well before the computer revolution, Internet or the dotcompandemonium (see for example Dunn (2001) and Adams (2003)). Yet what does Euphranor love, exactly? What is information? The question (...) has received many answers in different fields. Unsurprisingly, several surveys do not even converge on a single, unified definition of information (see for example Braman 1989, Losee (1997), Machlup and Mansfield (1983), Debons and Cameron (1975), Larson and Debons (1983)). (shrink)

A. Newell and H. A. Simon were two of the most influential scientists in the emerging field of artificial intelligence (AI) in the late 1950s through to the early 1990s. This paper reviews their crucial contribution to this field, namely to symbolic AI. This contribution was constituted mostly by their quest for the implementation of general intelligence and (commonsense) knowledge in artificial thinking or reasoning artifacts, a project they shared with many other scientists but that in their case was theoretically (...) based on the idiosyncratic notions of symbol systems and the representational abilities they give rise to, in particular with respect to knowledge. While focusing on the period 1956-1982, this review cites both earlier and later literature and it attempts to make visible their potential relevance to today's greatest unifying AI challenge, to wit, the design of wholly autonomous artificial agents (a.k.a. robots) that are not only rational and ethical, but also self-conscious. (shrink)

For millennia, knowledge has eluded a precise definition. The industrialization of knowledge (IoK) and the associated proliferation of the so-called knowledge communities in the last few decades caused this state of affairs to deteriorate, namely by creating a trio composed of data, knowledge, and information (DIK) that is not unlike the aporia of the trinity in philosophy. This calls for a general theory of knowledge (ToK) that can work as a foundation for a science of knowledge (SoK) and additionally distinguishes (...) knowledge from both data and information. In this paper, I attempt to sketch this generality via the establishing of both knowledge structures and knowledge systems that can then be adopted/adapted by the diverse communities for the respective knowledge technologies and practices. This is achieved by means of a formal–indeed, mathematical–approach to epistemological matters a.k.a. formal epistemology. The corresponding application focus is on knowledge systems implementable as computer programs. (shrink)

In computational complexity theory, decision problems are divided into complexity classes based on the amount of computational resources it takes for algorithms to solve them. In theoretical computer science, it is commonly accepted that only functions for solving problems in the complexity class P, solvable by a deterministic Turing machine in polynomial time, are considered to be tractable. In cognitive science and philosophy, this tractability result has been used to argue that only functions in P can feasibly work as computational (...) models of human cognitive capacities. One interesting area of computational complexity theory is descriptive complexity, which connects the expressive strength of systems of logic with the computational complexity classes. In descriptive complexity theory, it is established that only first-order systems are connected to P, or one of its subclasses. Consequently, second-order systems of logic are considered to be computationally intractable, and may therefore seem to be unfit to model human cognitive capacities. This would be problematic when we think of the role of logic as the foundations of mathematics. In order to express many important mathematical concepts and systematically prove theorems involving them, we need to have a system of logic stronger than classical first-order logic. But if such a system is considered to be intractable, it means that the logical foundation of mathematics can be prohibitively complex for human cognition. In this paper I will argue, however, that this problem is the result of an unjustified direct use of computational complexity classes in cognitive modelling. Placing my account in the recent literature on the topic, I argue that the problem can be solved by considering computational complexity for humanly relevant problem solving algorithms and input sizes. (shrink)

Logic has been a—disputed—ingredient in the emergence and development of the now very large field known as knowledge representation and reasoning. In this book (in progress), I select some central topics in this highly fruitful, albeit controversial, association (e.g., non-monotonic reasoning, implicit belief, logical omniscience, closed world assumption), identifying their sources and analyzing/explaining their elaboration in highly influential published work.

The Gricean paradigm in pragmatics has recently been attacked for its alleged lack of explanatory import, based on the claim that it does not seek accounts of how utterance interpretation actually works, but merely of how it might work. This article rebuts this line of attack by offering a clear and detailed account of the explanatory project of Gricean pragmatics according to which the latter aims for rationalizing explanations of utterance interpretation. It is shown that, on this view, Gricean pragmatics (...) seeks psychological explanations of utterance interpretation that are “cognitively real” in a perfectly clear and robust sense. (shrink)

Following Marr’s famous three-level distinction between explanations in cognitive science, it is often accepted that focus on modeling cognitive tasks should be on the computational level rather than the algorithmic level. When it comes to mathematical problem solving, this approach suggests that the complexity of the task of solving a problem can be characterized by the computational complexity of that problem. In this paper, I argue that human cognizers use heuristic and didactic tools and thus engage in cognitive processes that (...) make their problem solving algorithms computationally suboptimal, in contrast with the optimal algorithms studied in the computational approach. Therefore, in order to accurately model the human cognitive tasks involved in mathematical problem solving, we need to expand our methodology to also include aspects relevant to the algorithmic level. This allows us to study algorithms that are cognitively optimal for human problem solvers. Since problem solving methods are not universal, I propose that they should be studied in the framework of enculturation, which can explain the expected cultural variance in the humanly optimal algorithms. While mathematical problem solving is used as the case study, the considerations in this paper concern modeling of cognitive tasks in general. (shrink)

I present a novel, collaborative, methodology for linguistics: what I call the ‘explanatory economy’. According to this picture, multiple models/theories are evaluated based on the extent to which they complement one another with respect to data coverage. I show how this model can resolve a long-standing worry about the methodology of generative linguistics: that by creating too much distance between data and theory, the empirical credentials of this research program are tarnished. I provide justifications of such methodologically central distinctions as (...) the competence/performance and core/periphery distinction, and then show how we can understand the push for simplicity in the history of generative grammar in this light. (shrink)

Marr’s seminal distinction between computational, algorithmic, and implementational levels of analysis has inspired research in cognitive science for more than 30 years. According to a widely-used paradigm, the modelling of cognitive processes should mainly operate on the computational level and be targeted at the idealised competence, rather than the actual performance of cognisers in a specific domain. In this paper, we explore how this paradigm can be adopted and revised to understand mathematical problem solving. The computational-level approach applies methods from (...) computational complexity theory and focuses on optimal strategies for completing cognitive tasks. However, human cognitive capacities in mathematical problem solving are essentially characterised by processes that are computationally sub-optimal, because they initially add to the computational complexity of the solutions. Yet, these solutions can be optimal for human cognisers given the acquisition and enactment of mathematical practices. Here we present diagrams and the spatial manipulation of symbols as two examples of problem solving strategies that can be computationally sub-optimal but humanly optimal. These aspects need to be taken into account when analysing competence in mathematical problem solving. Empirically informed considerations on enculturation can help identify, explore, and model the cognitive processes involved in problem solving tasks. The enculturation account of mathematical problem solving strongly suggests that computational-level analyses need to be complemented by considerations on the algorithmic and implementational levels. The emerging research strategy can help develop algorithms that model what we call enculturated cognitive optimality in an empirically plausible and ecologically valid way. (shrink)

Knowledge representation is one of the central concepts in Artificial Intelligence. It is very common that knowledge about a field is expressed in natural language. Therefore, most of the times, knowledge representation using a logic programming language derives into a translation problem. This translation consists in the formalization of the statements, belonging to the knowledge level, which are converted into formulas of the so called symbolic level. Knowledge may be imprecise or vague and, in order to deal with vagueness using (...) declarative techniques, fuzzy logic programming amalgamates classical logic programming and fuzzy logic. Fuzzy logic programming has mainly led to programming languages that use annotations to represent vagueness. But vagueness is a linguistic phenomenon which is implicit in the statements of the knowledge level. Hence, the natural connection existing between these two levels is broken when annotations are employed, since they introduce weights in a symbolic level which are not present in the knowledge level and converts knowledge representation in a more complex, counterintuitive task. In order to overcome this problem, we propose a fuzzy linguistic logic framework which allows the treatment of imprecision through linguistic resources. This framework makes a clean separation between precise knowledge and vague knowledge. In this paper, we argue that this separation is more declarative than the one dispensed by the approach based on annotations, and can be beneficial for modeling a problem. (shrink)

In his famous 1982 paper, Allen Newell [22, 23] introduced the notion of knowledge level to indicate a level of analysis, and prediction, of the rational behavior of a cognitive arti cial agent. This analysis concerns the investigation about the availability of the agent knowledge, in order to pursue its own goals, and is based on the so-called Rationality Principle (an assumption according to which "an agent will use the knowledge it has of its environment to achieve its goals" [22, (...) p. 17]. By using the Newell's own words: "To treat a system at the knowledge level is to treat it as having some knowledge, some goals, and believing it will do whatever is within its power to attain its goals, in so far as its knowledge indicates" [22, p. 13]. In the last decades, the importance of the knowledge level has been historically and system- atically downsized by the research area in cognitive architectures (CAs), whose interests have been mainly focused on the analysis and the development of mechanisms and the processes governing human and (arti cial) cognition. The knowledge level in CAs, however, represents a crucial level of analysis for the development of such arti cial general systems and therefore deserves greater research attention [17]. In the following, we will discuss areas of broad agree- ment and outline the main problematic aspects that should be faced within a Common Model of Cognition [12]. Such aspects, departing from an analysis at the knowledge level, also clearly impact both lower (e.g. representational) and higher (e.g. social) levels. (shrink)

David Haig propounds and illustrates the unity of a radically revised set of definitions of the family of terms at the heart of philosophy of cognitive science and mind: information, meaning, interpretation, text, choice, possibility, cause. This biological re-grounding of much-debated concepts yields a bounty of insights into the nature of meaning and life. An interpreter is a mechanism that uses information in choice. The capabilities of the interpreter couple an entropy of inputs to an entropy of outputs is dispelled (...) by observation. The second entropy is dispelled. I propose that an interpreter’s response to inputs meaning of the information for the interpreter. In this conceptual framework, the mechanisms of interpreters provide the much-debated link between Shannon information and semantics. (shrink)

The ‘received view’ about computation is that all computations must involve representational content. Egan and Piccinini argue against the received view. In this paper, I focus on Egan’s arguments, claiming that they fall short of establishing that computations do not involve representational content. I provide positive arguments explaining why computation has to involve representational content, and how that representational content may be of any type. I also argue that there is no need for computational psychology to be individualistic. Finally, I (...) draw out a number of consequences for computational individuation, proposing necessary conditions on computational identity and necessary and sufficient conditions on computational I/O equivalence of physical systems.Keywords: Computation; Representation; Computational identity; Explanation; Narrow content; Physical computation. (shrink)

Artificial intelligence has often been seen as an attempt to reduce the natural mind to informational processes and, consequently, to naturalize philosophy. The many criticisms that were addressed to the so-called “old-fashioned AI” do not concern this attempt itself, but the methods it used, especially the reduction of the mind to a symbolic level of abstraction, which has often appeared to be inadequate to capture the richness of our mental activity. As a consequence, there were many efforts to evacuate the (...) semantical models in favor of elementary physiological mechanisms simulated by information processes. However, these views, and the subsequent criticisms against artificial intelligence that they contain, miss the very nature of artificial intelligence, which is not reducible to a “science of the nature”, but which directly impacts our culture. More precisely, they lead to evacuate the role of the semantic information. In other words, they tend to throw the baby out with the bath-water. This paper tries to revisit the epistemology of artificial intelligence in the light of the opposition between the “sciences of nature” and the “sciences of culture”, which has been introduced by German neo-Kantian philosophers. It then shows how this epistemological view opens on the many contemporary applications of artificial intelligence that have already transformed—and will continue to transform—all our cultural activities and our world. Lastly, it places those perspectives in the context of the philosophy of information and more particularly it emphasizes the role played by the notions of context and level of abstraction in artificial intelligence. (shrink)

This chapter describes the conceptual foundations of cognitive science during its establishment as a science in the 20th century. It is organized around the core ideas of individual agency as its basic explanans and information-processing as its basic explanandum. The latter consists of a package of ideas that provide a mathematico-engineering framework for the philosophical theory of materialism.

While the research programs in early cognitive science and artificial intelligence aimed to articulate what cognition was in ideal terms, much research in contemporary computational neuroscience looks at how and why brains fail to function as they should ideally. This focus on impairment affects how we understand David Marr's hypothesized three levels of understanding. In this essay, we suggest some refinements to Marr's distinctions using a population activity model of cortico-striatal circuitry exploring impulsivity and behavioral inhibition as a case study. (...) In particular, we urge that Marr's computational level should be redefined to include a description of how systems break down. We also underscore that feed-forward processing, cognition disconnected from behavioral context, and representations do not always drive cognition in the way that Marr originally assumed. (shrink)

Marr's levels of analysis—computational, algorithmic, and implementation—have served cognitive science well over the last 30 years. But the recent increase in the popularity of the computational level raises a new challenge: How do we begin to relate models at different levels of analysis? We propose that it is possible to define levels of analysis that lie between the computational and the algorithmic, providing a way to build a bridge between computational- and algorithmic-level models. The key idea is to push the (...) notion of rationality, often used in defining computational-level models, deeper toward the algorithmic level. We offer a simple recipe for reverse-engineering the mind's cognitive strategies by deriving optimal algorithms for a series of increasingly more realistic abstract computational architectures, which we call “resource-rational analysis.”. (shrink)

Thirty years after the publication of Marr's seminal book Vision the papers in this topic consider the contemporary status of his influential conception of three distinct levels of analysis for information-processing systems, and in particular the role of the algorithmic and representational level with its cognitive-level concepts. This level has been downplayed or eliminated both by reductionist neuroscience approaches from below that seek to account for behavior from the implementation level and by Bayesian approaches from above that seek to account (...) for behavior in purely computational-level terms. (shrink)

We consider approaches to explanation within the cognitive sciences that begin with Marr's computational level or Marr's implementational level and argue that each is subject to fundamental limitations which impair their ability to provide adequate explanations of cognitive phenomena. For this reason, it is argued, explanation cannot proceed at either level without tight coupling to the algorithmic and representation level. Even at this level, however, we argue that additional constraints relating to the decomposition of the cognitive system into a set (...) of interacting subfunctions are required. Integrated cognitive architectures that permit abstract specification of the functions of components and that make contact with the neural level provide a powerful bridge for linking the algorithmic and representational level to both the computational level and the implementational level. (shrink)

Daniel C. Dennett ha dedicado una parte considerable de su obra a concebir una aplicación de la ingeniería inversa y el adaptacionismo para explicar la evolución de la mente humana. Dennet considera esta perspectiva como una posibilidad prometedora en el desarrollo de una psicología científica, en contraposición al "materialismo eliminacionista" de la neurociencia. En este artículo se expone una aproximación conceptual y se examina un antecedente filosófico en las discusiones sobre el adaptacionismo en biología y psicología evolutiva: la intencionalidad o (...) teoría de los sistemas intencionales. Daniel C. Dennett has dedicated a great part of his work to the conception of an application of reverse engineering and adaptationism in order to explain the evolution of the human mind. Dennett sees this perspective as a promising possibility in the development of a scientific psychology, in contrast to the "eliminative materialism" of neuroscience. The article provides a conceptual approach and examines a philosophical precedent in the discussions regarding adaptationism in biology and evolutionary psychology: intentionality or the theory of intentional systems. Daniel C. Dennett dedicou uma parte considerável de sua obra a conceber uma aplicação da engenharia inversa e do adaptacionismo para explicar a evolução da mente humana. Dennett considera essa perspectiva como uma possibilidade prometedora no desenvolvimento de uma psicologia científica, em contraposição ao "materialismo eliminacionista" da neurociência. Neste artigo, expõe-se uma aproximação conceitual e examina-se um antecedente filosófico nas discussões sobre o adaptacionismo em biologia e psicologia evolutiva: a intencionalidade ou teoria dos sistemas intencionais. (shrink)

We propose a framework for including information‐processing bounds in rational analyses. It is an application of bounded optimality (Russell & Subramanian, 1995) to the challenges of developing theories of mechanism and behavior. The framework is based on the idea that behaviors are generated by cognitive mechanisms that are adapted to the structure of not only the environment but also the mind and brain itself. We call the framework computational rationality to emphasize the incorporation of computational mechanism into the definition of (...) rational action. Theories are specified as optimal program problems, defined by an adaptation environment, a bounded machine, and a utility function. Such theories yield different classes of explanation, depending on the extent to which they emphasize adaptation to bounds, and adaptation to some ecology that differs from the immediate local environment. We illustrate this variation with examples from three domains: visual attention in a linguistic task, manual response ordering, and reasoning. We explore the relation of this framework to existing “levels” approaches to explanation, and to other optimality‐based modeling approaches. (shrink)