Guessing Games and The Power of Prediction

The CogPhi reading group resumes next week.  CogPhi offers the chance to read through and discuss recent literature in the Philosophy of Artifical Intelligence and Cognitive51zmr2bn5hhl-_sx329_bo1204203200_ Science.  Each week a different member of the group leads the others through the chosen reading for that week. This term we’ll be working through Andy Clark’s new book on predictive processing, Surfing Uncertainty: Prediction, Action and the Embodied Mind.

CogPhi meets fortnightly, sharing the same time slot and room as E-Intentionality, which meets fortnightly in the alternate weeks. Although CogPhi announcements will be made on the E-Int mailing list, attendance at one  seminar series is not required for attendance at the other.  CogPhi announcements will also be made here.

Next week, October 20th, from 13:00-13:50 in Freeman G31, Jonny Lee will lead the discussion of the Introduction (“Guessing Games”) and Chapter 1 (“Prediction Machines”).  Have your comments and questions ready beforehand.  In fact, feel free to post them in advance, here, as comments on this post.

EDIT:  Jonny sent out the following message yesterday, the 19th:

It’s been brought to my attention that covering both the introduction and chapter 1 might be too much material for one meeting. As such, let’s say we’ll just stick to the introduction. If you’ve already read chapter 1, apologies, but you’ll be ahead of the game. On the other hand, if the amount of reading was putting you off, you’ve now only got 10 pages to get through!


2 thoughts on “Guessing Games and The Power of Prediction

  1. Here are some thoughts of mine on this week’s reading. It’s only the introduction, so I understand that some or all of these questions might be dealt with later in the book. I’m asking them now only as a way to guide how I read the book proper.

    1 Clark is ambitious, in that he is proposing PP not (just) as an explanation of how human brains enable human cognition, but as a constitutive account of, e.g., what it is “to encounter the world as a locus of meaning”. Will he really be able to argue for such a strong claim? It would have to be a very different book from he ones he has written before, which draw mainly on empirical results. Cf his claim: “To match the given picture… by twiddling knobs… just is to understand quite a lot about geology and geological causes.”

    2 The notion of “probabilistic model” introduced on page 5 is rather weak: is only a matter of choosing one’s models (which may be deterministic?) according to which is the most probable. I’m happy with that, but will he shift to a more substantial notion of probabilistic later on?

    3 Despite his distinction between the personal and sub-personal, Clark frequently shifts between those levels without comment, referring to things that the brain does (predict, model build) as what “we” do, and referring to things that we do (make sense of, perceive, experience) as things the brain does. Does this weaken his arguments?

    4 “Notice that the real-world perceptual matching task targets not a single static outcome (as in SLICE) but rather an evolving real-world scene. Matching the incoming signal, in the kinds of cases we will be considering, thus requires knowing how the elements of the scene will evolve and interact across multiple spatial and temporal scales.” Why should the second sentence follow from the first? If was are not meant to see how now, why not say something like “I will show that this requires…”?

    5 SLICE is not a good example, not just for the reasons Clark gives, but because it also is not really predictive.

    6 I was expecting the robe-SLICE elaboration to pave the way for Friston’s notion of active inference. But Clark stops short of that, talking only of acting “in ways appropriate to the combinations of bodily and environmental causes that (it estimates) make the current sensory data most likely”, and in ways that “alter the long-term structure of its own social and material environment, so as to inhabit a world in which the ‘energetic inputs that matter’ are more reliably served up as and when required.” Neither of these require acting in ways that make one’s model more likely.


  2. Pingback: Prediction Machines | PAICS

Leave a Reply

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

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

Google photo

You are commenting using your Google 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 )

Connecting to %s