Less Wrong is a community blog devoted to refining the art of human rationality. Please visit our About page for more information.

Mini-Review: 'The Oxford Handbook of Philosophy of Cognitive Science'

6 Post author: lukeprog 01 February 2012 09:48PM

Today I had the pleasure of wading through Oxford's new Handbook of Philosophy of Cognitive Science. The book consists of review chapters on the following topics:

  • Cognition (computationalism vs. embodied cognition, mental representation)
  • Consciousness
  • The nature of thought (concepts, language and thought)
  • Specific mental phenomena (perception, attention, emotions)
  • Meta-theoretic issues (the assumptions of cognitive science, relationships between disciplines)
  • Conceptual issues (the relation between concepts used by psychology, neuroscience, and philosophy)
  • First-order empirical issues (theory of mind, language, culture and cognition)
  • Traditional philosophical issues (rationality, metaphilosophy)

Below, I'll summarize a few of the chapters I found most useful and interesting.

In "Consciousness and Cognition," Robert van Gulick begins by listing 10 different things that is often meant by the term "consciousness." He then explains that some regard consciousness as more basic than cognition, while others see consciousness as dependent on cognition. (I take the latter view.) Van Gulick then surveys three main categories of theories of consciousness: philosophical theories (including Dennett's "multiple drafts" theory), cognitive theories (including Baars' "global workspace" theory and Tononi's information integration theory), and neurobiological theories (including Dehaene & Naccache's neuronal version of global workspace theory and Lamme's "local recurrence" model). Van Gulick concludes by surveying some of the methods used in consciousness studies.

In "Embodied Cognition," Lawrence Shaprio attempts to explain the difference between computational and embodied theories of cognition. Computational theorists treat the mind as a computational system, and seek to infer the algorithms by which the mind transforms inputs into outputs. (I'll add that at some level of organization this must be true, for physics appears to be computational and the mind runs on physics.) And what do embodied theories claim?

After reading the chapter, I remain confused as to how embodied approaches are supposed to be non-computational. For example, Shapiro quotes Ester Thelen explaining embodied cognition this way: "...to say that cognition is embodied means that it arises from bodily interactions with the world [and] depends on the kinds of experiences that come from having a body with particular perceptual and motor capabilities that are inseparably linked and that together form the matrix within which reasoning, memory, emotion, language, and all other aspects of mental life are meshed." But I don't see anything non-computational about that!

As far as I can tell, embodied theories are motivated by a rejection of early theories which naively proposed that the human mind is entirely a symbol-manipulating computation system ala the General Problem Solver and that the agent's body had little importance for how the the agent's cognitive algorithms manipulated those symbols. Embodied theories are correct to reject these theses, but this makes embodied theories incompatible only with the most naive forms of computationalism. The way I would put it is that the mind is computational, and we must be careful to remember that human cognition is embodied, situational, and dynamical. But this is not how Shapiro prefers to describe things.

In "Computationalism," Gualtiero Piccinini outlines the three research traditions of computationalism: classicism, connectionism, and computational neuroscience. He goes on to describe different kinds of computation:

On page 237, Piccinini says something similar to my thoughts above on Shapiro's chapter:

While it is safe to say that cognition involves computation in the generic sense, and that nervous systems perform computations in the generic sense, it is much harder to establish that cognition involves a more specific kind of computation.

The rest of the chapter offers a preliminary look at how the evidence might weigh for and against different kinds of computationalism about the mind.

In "Representationalism," Frances Egan describes a specific kind of computationalism. Representationalism is "the view that the human mind is an information-using system, and that human cognitive capacities are to be understood as representational capacities." The chapter discusses several kinds of representationalism, and the arguments given for and against them.

In "Artificial Intelligence," Diane Proudfoot and Jack Copeland focus on the quest for human-level AI. Their opening paragraph quotes Turing on the obvious achievability of AI, the whole brain emulation route ("One way [of making AI] would be to take a man as a whole and to try to replace all the parts of him by machinery"), and the danger of "runaway AI." Section 1 discusses the Turing Test. Section 2 discusses and rejects the Chinese Room argument against Strong AI, and section 3 discusses an "a priori" argument for Strong AI:

Given that the mind is scientifically explicable rather than a mystery, the mind is a mechanism, an information-processing machine; since the set of possible operations that can be carried out by information-processing machines is identical to the set of operations that can be carried out by the universal Turing machine... the mind must ultimately be explicable in terms of the computational properties of the UTM.

Section 4 discusses Moore's law and "the furistists" (Moravec, Joy, Kurzweil, Bostrom, the Singularity Institute). They quote SI as saying that "...at the very least it should be physically possible to achieve a million-to-one speedup in thinking." Proudfoot & Copeland reply:

But... the appeal to Moore's (or other similar) projections is fallacious. These provide no reason to think that relative increases in computer speed will be matched by increases is speed of thought.

And of course I agree. Faster computer speed only results in faster thought under certain ways of building a thought-machine. Faster computer speed only establishes the possibility for faster thought. I think faster thought is highly likely, but the argument for this conclusion is not given in the quoted article.

Section 5 concerns Singularitarianism, and unfortunately associates the word only with Ray Kurzweil, and is thus rather dismissive.

Comments (4)

Comment author: scientism 02 February 2012 02:17:16AM 2 points [-]

With embodiment different researchers use the concept at cross purposes. There are "embodied cognition" aficionados who simply claim that embodiment in some sense informs or shapes cognition (like Lakoff and Johnson). That's not particularly interesting. The more interesting claim is that the "bodily interactions with the world" that Thelen describes can replace an algorithmic account of some computational process taking place in the mind. It's the claim that there are no computations, no symbols manipulated, no information processed, there's simply the situated embodied organism reacting dynamically to its environment. (Sometimes researchers hedge on whether computation also has a role to play but it's easier to understand as a straightforward alternative.)

Comment author: fubarobfusco 02 February 2012 07:53:01AM 0 points [-]

Section 4 discusses Moore's law and "the furistists"

Is this a typo for "futurists" or a new term?

Comment author: ChrisHallquist 01 February 2012 11:31:18PM 0 points [-]

The details of this review make it sound kind of mixed--is there a better resource out there for getting up to speed on cognitive science?

Comment author: lukeprog 02 February 2012 12:48:27AM 2 points [-]

If you want to focus on the "science" end of cognitive science, I recommend Bermudez. The Frankish and Ramsey volume coming out in April may also be good.