Sumários
Connectionism
11 Maio 2026, 11:00 • David Yates
This week and next we will examine the main alternative to the classical computational theory of cognition, connectionism. We will begin with several related problems for classical computational accounts of mind, which have nothing to do with how the symbols get their meaning. These problems focus instead on whether it is plausible that central cognitive systems, i.e. those that are not modular (specialised for a particular task in the way that e.g. vision is) are classical symbol processors. In the Monday class we will perform a simple psychological test that seems to show that cognition is sensitive to content in a way that classical computationalism rules out. Then, as we will see, Fodor himself gave some powerful arguments that central cognition cannot be classically computational, based on the claim that belief-fixation is Quinean and Isotropic. These two properties together can be referred to as cognitive holism and can be seen to motivate connectionist accounts of the mind. We will focus on the central differences between classical and connectionist cognitive architectures by studying a simple connectionist network, and the ways in which both cognition and representation differ between CCTM and connectionism. We will then examine the way in which modern deep learning models of cognition, the currently dominant paradigm, developed from connectionism.
Seminar discussion on teleosemantics, artificial intelligence, and "Swampman"
6 Maio 2026, 11:00 • David Yates
Suppose the real reason why you want to defend a causal correlation theory of meaning is Searle's Chinese Room argument. You decide to defend the robot reply, according to which computations performed in the context of a robot's brain, as it interacts with the world perceptually, can give rise to meaning, because computations in such a context can give rise to causal correlations between the robot's symbols and the things they represent. Now you must face the problem of error: if meaning is causal correlation, how is it possible for a symbol to be incorrectly applied? Where do truth and falsity come from? At this point you might appeal to teleosemantics to solve the problem of error, but now there's a problem: the symbols in a robot's brain were not selected by evolution to covary with things in the world, in order to increase the robot's chances of survival and reproduction! Teleosemantics seems to entail that no computational system could ever think, just as Searle claimed, because such systems are not the product of natural selection, whatever the context in which they compute. To put the point differently: Searle himself could appeal to teleosemantics to explain where original (human) intentionality comes from, without undermining the Chinese Room.
Is the above argument sound? Does teleosemantics entail that strong AI is false?
Reading:
Jerry Fodor, Psychosemantics ch. 4 'Meaning and the World Order'.
You can focus on pp. 97-106 of this chapter for a survey of different version of naturalistic semantics in terms of how they respond to the problem of error, and read on for Fodor's own theory.
See this paper if you are interested in evolutionary approaches to meaning:
Karen Neander, 'Can Darwin Solve the Problem of Intentionality?'
Causal theories of meaning
4 Maio 2026, 11:00 • David Yates
This week we will focus on denotational theories of meaning, i.e. theories that explain the semantic properties of symbols in terms of direct relationships between individual symbols and what they denote. We will contrast such views with functional role semantics, which explain meanings in terms of symbol-symbol relationships. We will begin by considering the crude causal theory of meaning and the problem or error. We will then consider several solutions to the problem of error, including Dretske's theory positing a special learning period in which meanings are fixed, Fodor's asymmetric dependence theory, and the most popular theory, teleosemantics. Finally, we will consider the Swampman objection to teleosemantics.
Seminar discussion on Fodor and the problem of holism for FRS
29 Abril 2026, 11:00 • David Yates
Reading:
Jerry Fodor, Psychosemantics ch. 3 'Meaning Holism'.
Much of this chapter is difficult and it presupposes knowledge of certain other areas of philosophy of mind, language and psychology that we have not covered in class. However, you can focus on pp. 55-60 for a summary of meaning holism and its problems. Meaning holism is a consequence of functional role semantics, which we will discuss in class, and this section of Fodor's book explains what meaning holism is and why (according to Fodor) it is problematic.
The more ambitious among you might wish to read further. You can ignore the bit about confirmation holism; pp. 67-71 contain Fodor's ideas about the relationship between functionalism and the meanings of thoughts; pp. 71ff contain Fodor's views on functional role sematics, including why it seems like a promising semantic theory, contrasting it with its main rival (denotational or Millian semantics, which is Fodor's own position).
Functional Role Semantics
27 Abril 2026, 11:00 • David Yates
This week we will focus on the Chinese Room argument and the problem of meaning, or intentionality. Following on from Searle's argument, we will study this problem as a problem for CCTM: how might the symbols in a classical computational AI get their meaning? Is syntax sufficient for semantics after all, or is something else required for symbols to be meaningful? If so, what is it, and can it be stated in naturalistic terms? To begin with we will focus on the systems reply to Searle, and examine how it is connected with functional role semantics. We will connect this reply with functionalism and see that, from a functionalist perspective, syntax may after all be sufficient for semantics.