A classification of the global catastrophic risks of AI is presented, along with a comprehensive list of previously identified risks. This classification allows the identification of several new risks. We show that at each level of AI’s intelligence power, separate types of possible catastrophes dominate. Our classification demonstrates that the field of AI risks is diverse, and includes many scenarios beyond the commonly discussed cases of a paperclip maximizer or robot-caused unemployment. Global catastrophic failure could happen at various levels of (...) AI development, namely, before it starts self-improvement, during its takeoff, when it uses various instruments to escape its initial confinement, or after it successfully takes over the world and starts to implement its goal system, which could be plainly unaligned, or feature-flawed friendliness. AI could also halt at later stages of its development either due to technical glitches or ontological problems. Overall, we identified around several dozen scenarios of AI-driven global catastrophe. The extent of this list illustrates that there is no one simple solution to the problem of AI safety, and that AI safety theory is complex and must be customized for each AI development level. (shrink)

The structure-mapping theory has become the de-facto standard account of analogies in cognitive science and philosophy of science. In this paper I propose a distinction between two kinds of domains and I show how the account of analogies based on structure-preserving mappings fails in certain (object-rich) domains, which are very common in mathematics, and how the axiomatic approach to analogies, which is based on a common linguistic description of the analogs in terms of laws or axioms, can be used successfully (...) to explicate analogies of this kind. Thus, the two accounts of analogies should be regarded as complementary, since each of them is adequate for explicating analogies that are drawn between different kinds of domains. In addition, I illustrate how the account of analogies based on axioms has also considerable practical advantages, e. g., for the discovery of new analogies. (shrink)

The objective of research in the foundations of Al is to explore such basic questions as: What is a theory in Al? What are the most abstract assumptions underlying the competing visions of intelligence? What are the basic arguments for and against each assumption? In this essay I discuss five foundational issues: (1) Core Al is the study of conceptualization and should begin with knowledge level theories. (2) Cognition can be studied as a disembodied process without solving the symbol grounding (...) problem. (3) Cognition is nicely described in propositional terms. (4) We can study cognition separately from learning. (5) There is a single architecture underlying virtually all cognition. I explain what each of these implies and present arguments from both outside and inside Al why each has been seen as right or wrong. (shrink)

This piece is a commentary on a precis of Maggie Boden's book "The creative mind" published in Behavioral and Brain Sciences.

Connectionist models provide a promising alternative to the traditional computational approach that has for several decades dominated cognitive science and artificial intelligence, although the nature of connectionist models and their relation to symbol processing remains controversial. Connectionist models can be characterized by three general computational features: distinct layers of interconnected units, recursive rules for updating the strengths of the connections during learning, and “simple” homogeneous computing elements. Using just these three features one can construct surprisingly elegant and powerful models of (...) memory, perception, motor control, categorization, and reasoning. What makes the connectionist approach unique is not its variety of representational possibilities or its departure from explicit rule-based models, or even its preoccupation with the brain metaphor. Rather, it is that connectionist models can be used to explore systematically the complex interaction between learning and representation, as we try to demonstrate through the analysis of several large networks. (shrink)

What is creativity? One new idea may be creative, whereas another is merely new: What's the difference? And how is creativity possible? These questions about human creativity can be answered, at least in outline, using computational concepts. There are two broad types of creativity, improbabilist and impossibilist. Improbabilist creativity involves novel combinations of familiar ideas. A deeper type involves METCS: the mapping, exploration, and transformation of conceptual spaces. It is impossibilist, in that ideas may be generated which – with respect (...) to the particular conceptual space concerned – could not have been generated before. The more clearly conceptual spaces can be defined, the better we can identify creative ideas. Defining conceptual spaces is done by musicologists, literary critics, and historians of art and science. Humanist studies, rich in intuitive subtleties, can be complemented by the comparative rigour of a computational approach. Computational modelling can help to define a space, and to show how it may be mapped, explored, and transformed. Impossibilist creativity can be thought of in “classical” Al terms, whereas connectionism illuminates improbabilist creativity. Most Al models of creativity can only explore spaces, not transform them, because they have no self-reflexive maps enabling them to change their own rules. A few, however, can do so. A scientific understanding of creativity does not destroy our wonder at it, nor does it make creative ideas predictable. Demystification does not imply dehumanization. (shrink)

Many kinds of creativity result from combination of mental representations. This paper provides a computational account of how creative thinking can arise from combining neural patterns into ones that are potentially novel and useful. We defend the hypothesis that such combinations arise from mechanisms that bind together neural activity by a process of convolution, a mathematical operation that interweaves structures. We describe computer simulations that show the feasibility of using convolution to produce emergent patterns of neural activity that can support (...) cognitive and emotional processes underlying human creativity. (shrink)

My aim here is to raise a few questions concerning the problem of representation in scientific discovery computer programs. Representation, as Simon says in his paper, "imposes constraints upon the phenomena that allow the mechanisms to be inferred from the data". The issue is obviously barely outlined by Simon in his paper, while it is addressed in detail in the book by Langley, Simon, Bradshaw and Zytkow (1987), to which I shall refer in this note. Nevertheless, their analysis would appear (...) to leave open certain issues related to the nature of the "constraints" imposed by representation on problem solving strategies. (shrink)

The paper explores the creative thinking process and throws light on creativity enhancement. From the perspective of possible creativity enhancement both the characteristics of creativity and the creative thinking process are discussed, together with an analysis of the process and its common factors. Constraints on innovation (as a special type of creativity), innovation management and the acceptance of change are discussed; creativity between cooperating individuals is also examined. Some possible computer-based tools to enhance creativity, including innovation, are discussed. A framework (...) of facilities to help promote and support the use of creativity enhancement techniques and tools (SOLI) is proposed. The paper is the result of a study undertaken by the author into creativity, innovation and cooperation. (shrink)

Self-improvement was one of the aspects of AI proposed for study in the 1956 Dartmouth conference. Turing proposed a “child machine” which could be taught in the human manner to attain adult human-level intelligence. In latter days, the contention that an AI system could be built to learn and improve itself indefinitely has acquired the label of the bootstrap fallacy. Attempts in AI to implement such a system have met with consistent failure for half a century. Technological optimists, however, have (...) maintained that a such system is possible, producing, if implemented, a feedback loop that would lead to a rapid exponential increase in intelligence. We examine the arguments for both positions and draw some conclusions. (shrink)

This article describes an implemented architecture for intermediate vision. By integrating a variety of Intermediate visual mechanisms and putting them to use in support of concrete activity, the implementation demonstrates their utility. The sytem, SIVS, models psychophysical discoveries about visual attention and search. It is designed to be efficiently implementable in slow, massively parallel, locally connected hardware, such as that of the brain.SIVS addresses five fundamental problems. Visual attention is required to restrict processing to task-relevant locations in the image. Visual (...) search finds such locations. Visual routines are a means for nonuniform processing based on task demands. Intermediate objects keep track of intermediate results of this processing. Visual operators are a set of relatively abstract, general-purpose primitives for spatial analysis, out of which visual routines are assembled. (shrink)