Back on the anti-representation train

I have the wonderful opportunity at the moment to teach methods of Ecological Psychology and Dynamic Systems Theory, their philosophical basis, theoretical concepts, how they require certain analyses, and what kinds of explanations this/these perspective/s give. I am lucky to be in a department that is (as of yet) wholly representation-dominant, yet are curious, interested, and promotes theoretical plurality. I just personally keep running into the wall of not being able to believe in representations as an ontological foundation for psychology.

Perhaps someone can convince me otherwise. Four different ways of defining representations are as follows. The ‘literal-structural’ version, which amounts to old-school phrenology, i.e., it is literally neurons/physical structures in the brain that are/store representations. I do not meet many researchers that hold this belief. Mostly, I find this belief in folk psychology, because I think it is an easily envisionable version that has surface validity through movies and tv-series that talk about brain function in this way (the British “knowledge”-panel program QI is an example of this). Second, you have a ‘literal-activational’ version, which amounts to what I would call modern phrenology, i.e., it is the electrical activity in the brain that are representations. This and the literal-structural version are sometimes unhelpfully combined. fMRI, MRI, PET and similar techniques try to get at this by measuring blood flow to different parts of the brain, although it is an indirect technique since the assumptions go 1) thoughts, feelings, reactions, cognitions, etc, are produced in the brain, 2) in the brain there is electrical activity presumed to indicate usage of a particular brain area, 3) when a brain area is used, increased blood flow is seen to the area, 4) blood flow can indicate brain-part-usage and therefore thoughts, feelings, reactions, cognitions, etc. Third, we leave the literal kinds and hop the fence to ‘symbolic-mathematical’ accounts, that representations aren’t literal parts in the brain, but a more abstract version, instantiated by mathematics. This version is often combined with the methods of the ‘literal-activational’ version, and I’ve listened to several, prominent cognition scholars that have expressed beliefs in how mathematical equations are running in/produced by the brain… somehow. And several of them, upon explaining how the math got there to begin with, used various kinds of nativism as explanations. I’ve also found network-explanations, or ‘patterns of activation across groups of neurons’ and similar, both here and in the fourth grouping. Fourth, the ‘symbolic-abstract’ version, which often amounts to more hand-waving than the third, where representations are not mathematics, they can be groups of patterned activity, sometimes explained as dynamic clouds of activation of different kinds.

I just don’t find myself believing in any of them. With the literal-structural versions, there’s actually been attempts at finding them. Or, they were called Engrams at the time, and could simply not be found. There is little wiggle room in this description, either you can show empirically that a literal neuron/structure changes when you memorize something new or change a memory, or you are more or less forced to accept that this story isn’t the best one. Which is why I think so few ascribe to it beyond non-experts (e.g. folk psychology and AI researchers, the latter of which are almost exclusively dependent on this version).

The literal-activational version is far more popular however. In part, a lot of support comes from the medical sciences, where, if you poke a brain with electricity while talking or playing an instrument, are often disrupted in their activities. Or, if you have particular cognitive issues, a neurosurgeon can often quite easily pick out where a tumor is pushing up against (if you have a tumor, which isn’t always the case). This evidence is a muddle to me, since you can also completely remove brain parts, even half the brain(!), and still maintain functioning -which seems to me to break down the validity of this narrative. It is usually explained away with brain plasticity, but if brain plasticity is true (in reference to literal-activational representations), then the reliability across time when looking at the same brain, or when looking at different brains, break down (Anderson’s book After Phrenology is an interesting read). In fact, if either of the literal accounts are true, we should have specific structures and/or patterns of activation found to be stable across time for a single person, or across people. And if this were substantiated by research and industry, where the f* are our mind-reading helmets? Often when I bring this point up in discussions, neurocognitivists retreat back to neurons and how we don’t have the technical expertise to measure individual neurons in the entire brain simultaneously to answer that question. There are a couple of practices that seem to support the view however, sending a whole bunch of electricity down the spine seems to help people with motor-function constraints to regain/less disruptive movements. Which is great. But it is a far cry from being specific in the sense that’s needed to support the theoretical position. There are also toy versions like moving cubes on a screen using EEG, and beyond taking a surprising amount of time to train, once the person leaves the room and comes back the next day, the process restarts and cannot be continued the next day. I understand this position though, I do, I can deal with living alongside it (although I am extremely frustrated by grant-decisions heavily favoring neuroscience). However, the evidence surrounding it promises a specificity, a specificness, of identifying recurring activation/al patterns, that very clearly both empiricism and practical implementation does not live up to.

We get to the symbolic version of brain activity, particularly mathematical accounts, the story about activation standing in for representations seems to me to either give up the ontological foundation of representations (hand-wavy “it’s maths” explanations), or simply add another step of assumptions to the literal-activational account. Now, not only do you have to solve the above problems, but you also have to explain why a conglomerate of biology-chemical-electrical activity would instantiate an accounting tool that humans created to begin with. A good tool, mind you, but human-created nonetheless. From this point, of course, you get all kinds of half-to-non-scientific abstractions about how everything in the world is made up of mathematics, a tall-tale version of ruthless reductionism, and you of course lose the ontology of the phenomena you are trying to explain. You’d be surprised who I have heard literally say, in very public circumstances, that babies run physics equations in their brain that they are born with. And they are given so much research funding. My personal grievances aside, we have the symbolic-abstract version, which I find the most convincing, perhaps surprisingly. In some ways it can be seen as a less specific version of the literal-activational/structural versions of representations. Often, the explanation begins from the point that larger structures in the brain are not specific specific, but rather, there may be particular patterns of activation across structures that are recruited in a kind of online fashion. The patterns can thus “move around” the brain in part-deterministic-part-stochastic ways, meaning that, if you think about a cat, you come into that time-period from a different point than if you had been asked to think about it the next day. So, a similar pattern would repeat, but, not necessarily the same neurons and not necessarily all the same structures. This account would additionally fit the empirical findings detailed in After Phrenology. It would also explain how fMRI and MRI studies find general trends (averages over both spatial structures, and averages over time -how much time of course depends on imaging technique) across people. However. If this account is true, then I will most likely never get my mind-reading helmet, because it would be near-impossible to know how a particular pattern of activation would only stand for one object, and, what that pattern would look like the next day. Of course, an objection could be that it is not objects that are represented, but everything that we are exposed to continuously at the same time (plus memories, plus …). But then it would be nigh impossible to sort out which components of input lead to what activational pattern, particularly if that pattern changes over each instance. I do have some sympathy for this position though, as it seems to me to better fit more of the empirical data, but it gives a version of representations that is practically unusable except as a theoretical description.

To sum up. The more literal narratives of representations gives promises of specificity (particularly the medically inspired accounts), but this just hasn’t materialized on the practical-functional end, and, there is plenty of contradictory empirical evidence. The symbolic-mathematical perspective seems to explain some of the contradictions due to the shift of ontological basis to mathematics, but this step seems fantastical as it requires another set of beliefs to accept an ontological reality of maths. The world isn’t maths. The world is the world. Although it can be described in detail by maths. Lastly, we have the more symbolic-abstract-network version, which seems to me to cover most of the empirical literature and dispel the contradictions of the literal account. However, this perspective seems to me to not live up to the definition of a representation to begin with, losing the ‘specificness’ of representations that at the outset make them attractive to AI-researchers. In a recent interview with a prominent Ecological Psychological researcher, they were reacted to with a “I don’t see how anything could be anything else than computation”, and I have the exact opposite view, I have a very hard time seeing how anything could really be computational (that says something of value about psychological phenomena beyond simple, mechanical, surface level generalisations).So what is the brain up to? Biology does not preserve components that are not used.

Well, finally, we have the Raja-ian version of brain-activity, but we have left the realm of representations. Where the brain and central nervous system is seen more as a tuning fork, a resonance device, than something harboring ‘the real world’ in one way or another. As a non-content explanation of the brain I have high hopes for this perspective, and perhaps it is compatible with some versions of the abstract-variational-network version of representations (which do not live up to the demands of what a representation would be in the first place). But, just as more literal proponents of representations wait for the technological solutions to their theoretical problems, I’ll have to wait out the empirical evidence and theory-building required for a fuller account of the resonance narrative.

(2/4) Cognitive Psychology in Crisis: Ameliorating the Shortcomings of Representationalism. Representationalism.

I have yet to hear back from my examinators, so the following has not been officially critiqued. I did want to receive this before I posted it here, but alas, here we are.

This chapter is devoted to critiquing the principles of representations and clarifies and exemplifies that the map you follow when embracing representationalism is misguiding. It is 14 pages long, and you can find it here.

Historically, Titchener (1895) proposed stimuli to cause linear series of mental acts, at the end of which is a behavioural response. Which mental acts occur, and in which order, are for the experimenter a matter of speculation. It was critiqued by Dewey (1896) already the year after as being subject to the “empiricist fallacy”. This term is used in a broader sense in the thesis concerning unobservable events.

Chemero (2007) argues that Hegelian Arguments (arguments marshalled in an attempt to constrain empirical research and close down developing research programs a priori) shut down alternative interpretations, even ones that hold promise to give satisfying explanations. In my own view, cognition is often treated as the pinnacle of evolution and it enforces an arbitrary argument but powerful consequence. It heightens the credit towards the subject matter of psychology, a discipline often under fire from competing disciplines for being non-scientific. It is a left-over from the establishment of psychology as its own discipline, Unfortunately, many still live within the perspective that brain and mind are separate or “just different perspectives of the same thing”. Nonsense. Science necessarily relies on materialistic monism, no room for dualism.

The issue presents itself when we are unaware that this is excluding other alternatives and theorists act through a theoretical filter, biasing assumptions and interepretations in experimentation. If we assume that everything is represented in the brain then we will only look in the brain and interpret results on this basis also. An example of this is the curious case of mirror neurons. Barrett (2011) proposes that mirror neurons are difficult entities to account for without representations. Their function has been severely de-dramatised as of late, but observations made state that they fire both on others’ specific movement as well as one’s own. This finding is not contrary to Ecological Psychology, for example, but because of ignorant theory-ladenness, explanations given and research on, link them with representations. Instead, ignoring what contemporary cognitive theory forces us to believe, what could explain their function? It may just be simultaneous activation between stimulated sensory modalities and/or movement, and due to strengthening of simultaneous neuronal activation, they are just more so activated, or activated in different ways, than other neurons due to their multiple sources. Explanations like these are however more than discouraged due to Hegelian arguments and theory-laden contemporary cognitive psychology.

Because of the unfounded assumption that language is an abstract symbolic system following laws of grammar, we came to the false conclusion that it would be easy to construct what was supposedly so easily accomplished in our brain. Under a representationalist understanding of human enterprise, it really should have been simple, and justified, to put resources to projects like CYC and DARPA. When failed, it should have given some indication on that perhaps underlying theory is not correct in its assumptions.

A person is led into a room, seated and asked to read a list of words. After some time, the same person is asked to write down as many of the words as possible. The conclusion to this type of experiment is that the invisible process underlying the explanation of recall is called memory, and consists of representing the words in the brain, storing them, to later pull them out and write them down. Popper (1963) had the idea of theories to be non-scientific if any result could be explained in terms proposed by the theory. Posit that a participant in the above experiment does not write down any, or very few, words. Is the theory to blame and a rejection of representationalism in order? No, and in all honesty, it would not be justified to do so because the participant’s result does not directly falsify the claim. The first issue with this is that empirical observation cannot falsify the claim, and secondly, it cannot falsify it because the claim does not strictly deal with what is observed, but rather, what is not. It would be claimed that the participant failed, but not only, it would also be claimed that the participant failed to live up to the already assumed unobservable process posited to exist.How can a methodology be accepted that, without anything else to refute alternative explanations on than Hegelian arguments, posits an unobservable process to exist and then compare any observable behaviour to live up to its presumptuous ideals? Instead, the question needs to be, what is it that actually is observed? There are two behaviours, reading the list and writing down words previously on that list. Everything else is an assumption.

To exemplify that logic is without perspective and not a reasonable norm, the classic four card task (Wason, 1966) residing on truth-table logic will be used. Trivers’s model was introduced to this task, yielding the if-then statement “If a previous employee gets a pension from the firm, then that person must have worked for the firm for at least 10 years.”. The four cards read, “got a pension”, “worked for 10 years”, “no pension” and “worked for 8 years”. Perspective, as mentioned earlier, is crucial. When participants were told they were an employee, they turned up “worked for 10 years” and “no pension”. When told they were an employer, they turned up “got a pension” and “worked for 8 years”. The latter situation renders choices of participants consistent with both Trivers’s model of cheater detection and the laws of the truth table. The former situation however, is not consistent with truth-table logic, but is explained by Trivers’s social contract theory. Gigerenzer (2008) argues that this is essentially a frame-of-reference problem and it is unfair to set up (albeit, perhaps, unintentionally) an experiment in this way in order to confirm a hypothesis. In contrast, it is important to note that logic can be appropriate as a criteria, but its domain is restricted (Gigerenzer, 2008). Truth-table logic experiments have not explained human enterprise, but rather, explored the limits of logic as criteria.

Challenging traditional cognitive psychology is an uphill battle against tradition, norm, life works, unfair criteria, Hegelian arguments, the Empiricist Fallacy, theory-ladenness and non-falsifiability. However, on a theoretical basis it has, thus far, little to stand on.

It is not with neuropsychologists I lay blame, they often know of all the issues inherent in methodology and apparatus, it is with those who draw unfounded conclusions from this field. First of all, it is not a natural environment for humans to lie frozen in an enclosed area fixating on a screen, but more so to a general point; can we ultimately say that performing no task is a valid baseline to compare with performing a task? The assumption is that it is, but again, it only comes about because of the restrictions on methodology because of the practical restraints in testing participants. What other baseline is possible? A second issue is that the difference in activation between the two conditions, depending on particular method, shows a maximum of 5% difference in activation. The remaining 95% of the activation is at the same levels under both conditions (Pfeifer and Bongard, 2007). What are those 95%? Contemporary cognitivists tend to ignore them and usually only point to the difference (for example Ochsner & Gross, 2008, or see Logotheitis, 2008, for a discussion on fMRI-techniques), which is clearly all too simplistic. A third issue is brought to our attention through Naghavi and Nyberg (2005), whom caution against too much enthusiasm by stating that “functional neuroimaging techniques can at best specify the coincidence of regional brain activations with specific cognitive demands. These methods cannot determine which brain regions are essential for a specific cognitive process.” (Pfeifer and Bongard, 2007, p. 321). It is important that we do not let unwarranted assumptions and generalisations taint the neuropsychological field, turning it into a modern version of phrenology where different brain parts do different things in isolation. A fourth important aspect is the assumption that the images show “thoughts” or other vague definitions of cognition. What we in fact see, taking fMRI as an example, is firstly an inference between ‘more thoughts’ and ‘more activation’, secondly an inference from ‘amount of activation’ to ‘amount of blood flow’, and thirdly, an inference from blood flow to an averaged out numeric value between spatial areas, participants and timeframes. There are thus three steps of inferential logic which makes it vulnerable to both a priori and ad hoc assumptions of what it is we are actually looking at when we are presented with these images. There is thus little support gained, at the present moment in time, from the maturing field of neuropsychology. We simply do not yet have enough knowledge, specific or enough controlled techniques to confidently state what the brain is doing. What are we actually looking at?

Although representations are unobservable entities with only assumptions to rely on, yet are essential for contemporary cognitive psychology, and thus needs to physically exist, the only thing left to deal with in representationalism, is that of Entity Realism. This is the proposition that you can still be justified in assuming a realist standpoint for theoretical entities (and representations fits this bill), if one has pragmatic use for them as tools in experimental investigation of other entities (Chemero, 2007). As in the case of asking participants to recall a list of words, the explanation given for their current behaviour is by reference to a previous behaviour, but, what went on in participants have not actually been observed, the word ‘memory’ is just used to fill this gap (Barrett, 2011). In other words, the issue with this proposition is that, because representations are necessary for the internalist account, yet have not been established empirically to actually exist, the assumption does not really explain anything. It is merely stating that, this is one possible process that may occur because it would fit the criteria for linking one behaviour to another. There is doubt that cognitive scientists would resort to this however, since the power of the concept is drastically reduced, and in all right.

All we have done is named unobservable, hypothetical processes, leading us down a garden path, away from the core subject of psychology. We want to understand why humans behave the way they do, we want to understand what the brain does. Representationalism does not provide these answers. “…if we cannot do any better than this, we should stop using the word…” (Gibson, 1986, p. 254).