Sometimes it happens that reading a book ignites a seemingly unstoppable whirlpool of ideas. The book in question is “Surfing Uncertainty: Prediction, Action, and the Embodied Mind” by Andy Clark.
Why is this a partisan review? Because Clark himself had already convinced me that the general idea is worth pursuing, well before writing the book. To use a famous expression: I want to believe. However, since I keep obsessing about my own biases, I also want to be as critical as possible – call it overcompensation, if you must.
In this post I will briefly review the book in general terms, the whirlpool of ideas mentioned above is mostly about the criticism I have to offer, which will come later (with the vague hope it might be useful).
As the title suggests, Clark’s book is about a subject I’ve touched before: the brain as a prediction engine (for an introduction, see previous posts: The Predictive Brain, part 1 and part 2). Clark himself summarised the arguments developed in the book via a series of posts published at the brains blog (starts here); for a review that also describes how the book is organised, see Andrew Buskell.
For my part, my previous discussion (links above) did not even mention a fundamental concept: confidence. When issuing a prediction, or even when detecting a signal, one important element that should never be overlooked is the evaluation of how much confidence can be attributed to the result. I’m using the word “confidence” here in order to keep the idea fairly general, for the time being. Clark does an exceptionally good job at explaining why the concept is indeed foundational, and how introducing it allows to move the generic predictive approach from the level of being a nice idea, all the way up to a research framework in the making. I mention this because, after appreciating the crucial role that confidence estimations play, my older presentation of the matter starts looking incomplete to the point of being misleading.
An important feature of the book is the emphasis it places on embodiment. However, it’s worth noting that the book starts by adopting mainstream assumptions of the computational kind – indeed, an overarching aim of the book is to bridge the gap between computational and embodied approaches. I was a little surprised to find that the computational assumptions are not really discussed: the fact that brains and neurons process information is taken as self-evident (Note: I agree!). The book proposes the “Predictive Processing” label (PP) as an overarching definition, able to point to a whole family of separate approaches. The “processing” word is telling: we are dealing with a solidly computational outlook. As the book proceeds, however, the ’embodiment’ promise in the title gets gradually fulfilled: PP isn’t merely applied to perception, instead, the book shows how the same framework can be used to model action and action control. The result is a continuum, from perception to action, which cannot possibly brush aside the fact that behaviours are physical: actual body parts move. Yes, brains have a controlling role over action, but, alas, much of vanilla cognitive science has been historically happy to study brain function without giving much attention to the role that bodies have in shaping what the brain does and, crucially, in defining what count as successful strategies.
Clark’s treatment has the refreshing quality of restoring the due balance in a field that has been characterised by unjustified prevalence of opposing extremes. Once upon a time, uncompromising behaviourism of the Skinner kind was accepted as the default assumption, only to get superseded by the opposite (and symmetrically wrong) stance of computational cognitivism.
Needless to say, in my view, this is one reason why this book was necessary (there are more!), and finds me in full agreement. [For those interested in this general debate, leaving aside the specific view offered by PP, I have explored (also) the relation between computationalism and embodiment in two posts at Conscious Entities: part 1, part 2. Moreover, I have recently produced my own Twitter-storm, commenting on Krakauer et. al. (2016) [Highly recommended reading!]. From the other side of the fence, my discussions (see also) with Golonka and Wilson might provide some insight on why and how classic cognitive approaches are being challenged by the school(s) of (more or less) Radical Embodiment.]
So far, so good: the assumptions on which the book rests look more than reasonable to my eyes, which made my reading sympathetic from the start. Additionally, one of the overarching aims of the book, reconciling computational and embodied approaches fits my own agenda perfectly. What’s not to like? Well, I must also report that some my own expectations did not find satisfaction.
Clark is a philosopher, so I was hoping to find a book that revolved around known philosophical issues, and showed how PP helps surpassing them. Pretty much like Hohwy’s “The Predictive Mind“, Clark’s book isn’t taking this approach. Instead, it begins from general considerations or observable phenomena, and methodically ties them to existing scientific literature. The result is a book that organises and summarises an astonishing amount of scientific (empirical, theoretical and frequently pretty arduous) work. Not what I was hoping for, but it turns out that sometimes you do get what you need. The works cited by Clark are usually scientific papers: as such, the vast majority isn’t suited to discuss the big picture at length, and is even less able to provide an overview of how different pieces fit together. Thus, we needed Clark to do this hard work, and indeed, it’s hard to imagine how it could have been done better.
If you are trying (like me) to gain a better understanding of where PP-centric research is heading, what it assumes, and what are the main conceptual pillars on which it rests, this book will satisfy your hunger and might even leave you feeling bloated. [Personal note: while reading the second half of the book I found that I was studiously slowing down. In part, this was to avoid overload and to make sure I was assimilating as much content as possible. In part, I just didn’t want the book to finish, as reading it was a genuine pleasure throughout.]
What of my aversion to standard academic publishing? [I.e.: my claim that peer-reviewed monographs frequently spoil the joy of reading by being overly cautious and pedantic.] Once again, my expectations proved to be misplaced (I love to be surprised! 😉 ).
On caution: the need to be fairly uncontroversial does transpire in many ways, you could say that it is ubiquitous. For example, Clark keeps adding caveats like “if the story I’ve been constructing is on the right track“, which went duly noticed and appreciated. We are, after all, dealing with an emerging, still to be established, research programme. The field is solidifying quickly, but we are far away from having a fairly complete and coherent jigsaw. On the other hand, treatments of thorny issues such as consciousness itself (why do we perceive some predictions?) are cautious to the point of being disappointingly sketchy, if not overlooked.
The flip side is how Clark leverages support from existing literature, which did surprise me in a good way. While reading, a recurring pattern characterised my reactions: judging on the primary text itself, I was frequently inclined to conclude “yes, the argument is promising, but I’m not convinced that it is strong/watertight enough to abandon due scepticism”. However, when the book relied on a body of evidence that I did happen to be familiar with, my initial reaction was regularly overridden (sometimes after checking the references): I ended up realising that Clark’s arguments are backed by wide and deep evidence (empirical and/or theoretical), to a point that wasn’t immediately evident by reading the text itself. Thus, my recommendation for future readers is obvious, but important: if you find yourself unconvinced, do read the relevant references. It is likely that you’ll find plenty of reasons to be convinced in the supporting bibliography. Needless to say: despite my bias against academic (peer-reviewed) monographs, Clark’s book struck me as an example that it is possible to get it right – chapeau.
Another pleasing consequence of how the book is organised is that, by offering a clear overview of the research field, it helped me in identifying the areas that failed to convince me in full. Since I know that I want to believe, I am very determined to use my resources to find and explore the reasons why the PP framework might not hold water. I found some, technical, which I plan to discuss separately. I also found what might be summarised as “gaps”, both at the beginning of the story (in the form of philosophical foundations as well as the lack of a convincing evolutionary perspective), and at the end (mostly in the lack of explanation of what we experience as our mental life).
The first consideration is a little worrisome; it seems to me that the research field is at risk of doing the usual mistake: oversimplifying. As for what I perceived as gaps, I see no reason for concern: if I’m right, these gaps should be treated as opportunities, to be seized by theorists coming from both philosophical and scientific perspectives.
In follow-up posts, I will try my own luck, and see where my criticism might lead – with apologies for being so openly and unjustifiably haughty.
Clark, A (2016). Surfing Uncertainty: Prediction, Action, and the Embodied Mind Oxford Scholarship DOI: 10.1093/acprof:oso/9780190217013.003.0011
Hohwy, J (2013). The Predictive Mind Oxford University Press DOI: 10.1093/acprof:oso/9780199682737.001.0001
Krakauer, J., Ghazanfar, A., Gomez-Marin, A., MacIver, M., & Poeppel, D. (2017). Neuroscience Needs Behavior: Correcting a Reductionist Bias Neuron, 93 (3), 480-490 DOI: 10.1016/j.neuron.2016.12.041