The Life & Mind Seminar Network

“Radical post-cognitivism: new approaches to intelligence and the mind” the inaugural lecture of Professor Mark Bishop

Posted in General by Andrew Martin on June 26, 2012

What follows is the review I wrote for the AISBQ of Mark Bishop’s inaugural lecture, which addressed the past, present and future of Cognitive Science, Enaction and Cognitive Computing.  The article was written for a slightly more general audience than I imagine necessary for this blog, so I’d be very happy to go into further detail or discussion in the comments of this page. – Andrew

On the 13th March, Professor Mark Bishop gave his inaugural lecture to a full lecture theatre at Goldsmiths, University of London. Entitled “Radical post-cognitivism: new approaches to intelligence and the mind”; described as the “integration of a lifetime’s work”, the theme was how the underlying assumptions of Cognitive Science have changed during Bishop’s career, and how he sees it developing in the future. The core argument was that much contemporary Cognitive Science research tacitly assumes intelligence is the result of computations upon conceptual representations; a philosophical stance that is at least questionable given many longstanding critiques [2].

Bishop identified three avenues by which Computationalism came to pervade Cognitive Science: (i) explicitly, that cognition was taken to be defined as computation upon representations;(ii) implicitly, that cognition could be defined as computation upon vectors of real numbers and(iii) descriptively, through confusion of accurate computational models of neurons with “an ontological claim about the reality of what neurons do”. “We can describe the operation of brain neurons mathematically, computationally, but that’s no reason to believe that brain neurons really do compute.”

The current state of the art of AI was addressed with a video of IBM’s Watson and a live demonstration of Apple’s Siri, which both seamlessly integrate natural language processing, voice recognition and information retrieval to a degree that many would have considered unfeasible only a few years ago, especially without extensive user calibration. Bishop noted that these undeniable successes utilised an approach that was not inspired by, or attempting to recreate, human intelligence. As such, contemporary artificial intelligence was shown to have successful applications, but was limited in its explanatory power.

Bishop continued by identifying weaknesses in a computational account of mind, firstly the human mind was suggested to be capable of insights unreachable through logical inference. This argument, citing John Lucas, Roger Penrose, and Kurt Gödel, showed there exists logical statements that a human can see to be true but a computational process could never prove to be true. Bishop subsequently questioned the very notion of computation, claiming that the criteria for assigning computational properties to a process are “observer relative”. This claim was fortified with reference to his earlier work on John Searle’s Chinese Room argument which criticises the notion that computation could ever lead to understanding and Bishop’s own “Dancing with Pixies” argument which aims to demonstrate that a strong computational theory of mind implies panpsychism  [5].

The other branch of computationalism, that mental processes manipulate representations, was considered next. An entertaining demonstration of inattentional blindness (including a few extra surprises for anyone who had previously “seen the gorilla”) the success of which questions whether the human mind actually processes a camera-like representation of the visual scene. Subsequently, the homuncular argument which claims that explaining vision with representations begs the question as the representations themselves require a observer. Bishop cited Dennett’s “content/vehicle distinction” clarifying he was not denying the existence of patterns of neural activity that appeared to represent the outside world, but that their existence was not sufficient evidence that they were being exploited as such by the mind.

As an introduction to an alternative to computationalism, Bishop described the operation of a centrifugal (or “Watt”) Governor, as an example of adaptive real-time behaviour, without objective representations. This led to a discussion of swarm intelligence, where intelligent behaviour can appear to emerge without the existence of a central executive controller or any encoding of a global goal. The application of this approach was further demonstrated with a discussion of the success of Bishop’s implementations of Stochastic Diffusion Search [1].

This led to Bishop’s concluding claim, that artificial intelligence and cognitive science are finally parting ways, artificial intelligence applying computational techniques on “big data” to real problem solving, but at the expense of providing insights on big questions about mind. Cognitive science can continue to address these questions, but to do so requires a change of tack to align itself more with philosophers such as the phenomenologists Maurice Merleau-Ponty, Humberto Maturana and Francisco Varela who emphasise the role of the body, environment and action. Bishop encapsulated this emphasis in the phrase “My brain, in my body, in our world.”

On these grounds Bishop identified the “four E’s” defining characteristics for a new era of cognitive science, that research should recognise the extent to which they are: Ecological, accounting for the environment; Embodied, concerning the physical presence of a system; Embedded, concerning the system’s relation to the environment; Enactive, concerning the role of action.

References

[1] M. Bishop. Stochastic Searching Networks. In First IEE Conference on Artifical Neural Networks, pages 329–331, 1989.

[2] M. Bishop. A cognitive computation fallacy? cognition, computations and panpsychism. Cognitive Computation, 1(3):221–233,  2009.

[3] M. Bishop. Radical post-cognitivism: new approaches to intelligence and the mind. March 2012. Inaugural lecture given at Goldsmiths, University of London.

[4] S.J. Nasuto, M. Bishop, and K. De Meyer. Communicating neurons: A connectionist spiking neuron implementation of stochastic diffusion search. Neurocomputing, 72(4-6):704–712, 2009.
[5] J.Preston and M.Bishop. Views Into the Chinese Room: New Essays on Searle and Artificial Intelligence. Clarendon Press, 2002.

Advertisements

10 Responses

Subscribe to comments with RSS.

  1. Tom Froese said, on June 28, 2012 at 7:16 am

    Very nice! I couldn’t agree more.

    How was the talk received by the audience? Was it considered as utterly outrageous by everyone except for a small minority? Or are more people coming around to similar conclusions?

  2. Andrew Martin said, on June 28, 2012 at 2:20 pm

    The talk was received very well, but as it was an inaugural that’s not really a fair assessment. Any questions other than the simplest of clarifications (of which there were none) weren’t allowed, and the audience included University administrators and family.

    Having said that, it must have hit home for many people, in the following reception I didn’t hear a single anti-Chinese Room argument, which is something of a first for me.

  3. Mike Beaton said, on June 29, 2012 at 3:51 pm

    Good review! Thanks.

    I wonder if, in saying that Cog. Sci. has assumed “computations upon conceptual representations”, that you (or Mark Bishop?) are (were?) implicitly treating conceptualism and computationalism as the same (or, at least, as positions which stand or fall together).

    I just wanted to point out that this might well be wrong. Conceptualism (in its barest form) says that that mind and rationality are ‘the same’ (co-extensive). And rationality, in its barest form, is just ‘action for reasons’. So there’s a version of conceptualism, which I support, which says that having a mind is the same thing as being able to act for (your own) reasons. I would say that this claim is completely separable from computationalism (I’m not a computationalist), or from any type of claim that having a mind involves manipulating representations.

    Other than that (or including that, if I’ve read you wrong) great review, thanks!

    Mike

    • Tom Froese said, on July 8, 2012 at 1:51 am

      Hahaha… Mike, but of course! It’s been a while since we have had this debate! ;-)

      Personally, I find the term ‘reasons’ still too cognitivist (or, before that time, they would have said intellectualist). In fact, it is intimately tied up with the whole modernist project of Descartes and co, which we must overcome as fast as possible if we are to survive on this planet.

      Having said that, I can sympathize with what you are aiming for, I think. What about saying that mind is about acting for meaning? I feel that this may sidestep the whole conceptual/non-conceptual debate. In other words, mind is sense-making.

      What do you think?

      • Mike Beaton said, on July 9, 2012 at 9:58 am

        Ah, Mr. Froese, good day to you, sir!

        So, Andrew, sorry if I am hijacking the debate, or at least this sub-thread! Hopefully Tom’s comments bring it back to the key issues!

        Tom, I agree with what you say about Cartesianism, and I think in the end we agree with each other about this stuff. For whatever reasons (in my own life), I am discussing the limits of intellectualism from ‘within the tradition’, and you’re coming at it more from outside. But I think there is room for rapprochement (see, e.g., the McDowell-Dreyfus debate, and some work by David Lauer; and closest to my own themes, I guess, Noë who is both a conceptualist and an enactivist).

        Is there any ‘point’ in my position? I hope so, I hope I can convince a few more people who are like I was, by being able to start where they start from, but we will see! Also, there is the whole issue of the ineliminable subjectivity (in the genuine, sense-making sense) in concepts like information, and even entropy, which is not understand by most philosophers and even scientists who appeal to them (http://bit.ly/NjbU1q). (Again, limits of intellectualism which ought to be recognised from within.)

        Mike

    • Andrew Martin said, on July 8, 2012 at 8:12 pm

      My use of the word “conceptual” may have been a poor choice. The point I wanted to get across was that there are some well known arguments against orthodox representationalism.

  4. christophemenant said, on July 8, 2012 at 2:18 pm

    “Mind is sense making”. Mind as “acting for meaning”. Great wording Tom!
    Interestingly, this is part of a presentation I made last wk at AISB/IACAP 2012 conference in Birmingham where it is argued that TT, CRA & SGP could be understood as “can AAs generate human-like meanings?” (presentation available at http://crmenant.free.fr/IACAP-AISB2012-C.Menant-050712.pdf)
    “Action” (mental or physical, or …) is part of the story, even if not always explicited.

    Would you agree ?

    • Tom Froese said, on July 10, 2012 at 3:04 am

      Christophe: Yes, I think that the problem is indeed about the generation of meaning. But I’m not sure whether I would say that the problem of meaning is limited to the human-like; I think science would already have achieved a lot if we could fully understand the generation of animal-like meaning.

      Mike: If you could take on such concepts as information and entropy from within the tradition, as you say, that would be awesome! I recently spoke with Dan Hutto about his forthcoming enaction book, and my impression was that he too easily assumed that there is ‘information’ in the world independently of sense-making. I know that this was a strategic move on his part, but I wonder if we can do better…

      • Mike Beaton said, on July 10, 2012 at 9:06 am

        Many philosophers and scientists (wrongly) assume that there is information in the world independently of sense making.

        A key point is that some of the deepest thinkers on this issue from within science realise that this is not so.

        Have a look at Section 4 of http://bit.ly/NjbU1q (which is now forthcoming in the International Journal of Machine Consciousness), for what I have managed to say on this so far.

      • christophemenant said, on July 10, 2012 at 9:31 am

        Tom: Agreed that meaning is not limited to human-like.
        Meaning generation exists within animals, humans & AAs. There is a tool modeling that: the Meaning Generator System. It is in the IACAP-AISB2012 presentation (see p 4). The generated meaning is system dependent and links the agent to its environment. The meaning generation process is constraint satisfaction based. Constraint=> Meaning => Action for constraint satisfaction. The MGS is simple. Just need to position the constraints (original-intrinsic for animals & humans, derived-transposed for AAs). You will find a paragraph addressing enaction & MGS in the book chapter “computation on information, meaning and representations. An evolutionary approach” (http://crmenant.free.fr/2009BookChapter/C.Menant.211009) published at http://www.worldscibooks.com/compsci/7637.html.
        As you will see, meaning generation is kept separated from autonomy or adaptivity. These performances can be part of the agent containing the MGS, keeping the latter as simple as possible.


Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out / Change )

Twitter picture

You are commenting using your Twitter account. Log Out / Change )

Facebook photo

You are commenting using your Facebook account. Log Out / Change )

Google+ photo

You are commenting using your Google+ account. Log Out / Change )

Connecting to %s

%d bloggers like this: