Oliver Scott Curry is wrong about associationism
This article by Dr Piotr Wozniak is part of SuperMemo Guru series on memory, learning, creativity, and problem solving.
This idea must die: Associationism
In the 1970s, when I first heard of Pavlov dog experiments, I instantly realized that our entire behavior is driven by reflexes (conditioned reflexes or similar). I never had a doubt. Three decades of my professional life have been based on associations. Everyday, I do associate questions with answers in spaced repetition (SuperMemo).
For Christmas 2018, I got a nice gift: a book listing all ideas that are supposed to die ("This Idea Must Die: Scientific Theories That Are Blocking Progress"). I do not read paper texts any more. However, I make some exceptions when I am away from the computer (e.g. on vacation).
While contemplating the book lazily, it was a nice jolt to find out a short text that associationism should die by Oliver Scott Curry. Curry is a genius mind that belongs to the young generation of researchers who are not afraid to pose challenging questions and break the old mold. Curry got a PhD in economics, and is passionate about investigating the impact of game theory on social behaviors and morality. In his youth he showed both communist inclinations, and attended a Catholic school as well. In many senses, in Curry I see a younger self. This makes him into a highly sympathetic persona. I like to listen.
When in comes to associationism, my thinking about the brain may indeed be a bit simplistic. Associating questions with answers paints a very simple model indeed. Seeing this as two inputs to a neuron or a connection between concepts makes all reasoning about the brain easy. Too easy? Perhaps it is time for a closer examination?
Associating poison with nausea
I know Curry is wrong, but I always claim that wrong models pose a great deal of inspiration. The need to keep wrong ideas alive was addressed in the same book by Ian McEwan who insists we should "retire nothing!". By studying Curry's position, I could organize my own thoughts using my models of learning. If you are not familiar with my model of thought presented in How to solve any problem?, you may need to preview that model first to understand "concept map activation". This is similar to William James's "stream of thought", and many similar interpretations since.
Curry asks:
At any given time, an organism is confronted by an infinite number of potential stimuli, and subsequently, an infinite number of potential outcomes. A day in the life of a rat, for example, might include waking up, blinking, walking east, twitching its nose, being trampled on, eating a berry, hearing a rumbling noise, sniffing a mate, experiencing a temperature of 5°C, being chased, watching the sun go down, defecating, feeling nauseous, finding its way home, having a fight, going to sleep, and so on. How does the rat discern that, of all the possible combinations of stimuli and outcomes, it was the berry alone that made him feel sick?
Those who study neural networks will probably know the answer: the rat is facing the curse of dimensionality! However, we know it will quickly learn the relevance of stimuli because of feature extraction. It will then employ feature selection to produce concepts or concept maps ripe for the association. Feature extraction is determined by the history of reward generated by the learn drive system. I use the term knowledge valuation network to explain why the taste of a berry is particularly important for the rat while eating, and why the memory of eating is particularly important in the period when symptoms of "feeling sick from eating" may show up. Incidentally, dimensionality reduction stands behind the power of Algorithm SM-17 in spaced repetition. In a chaos of data, without resorting to neural processing, SuperMemo reduces the model of memory to just three key dimensions.
Wrong models that inspire
The value of wrong models comes from their ability to build foundations of an entire semantic edifice. For reasons I explain elsewhere at SuperMemo Guru, such edifice is best built in a single human mind that keeps evolving the model of reality using the same set of rules based on memory stability, forgetting and generalization. If the model is wrong, the edifice will crumble at some point. We all live with hundreds of wrong models, and they do crumble one by one as we keep exploring the world. If a core model is wrong, say my Neurostatistical Model of Memory, a well-crystalized brain may never recover. At that point, we move to a higher level of intelligence: social intelligence. Death of a brain is like an erasure of a single synapse in the act of forgetting. The social neural network may forget the old wrong idea, or more likely, pick it up with brains that will mark it with low retrievability, which will make it easier to self-correct by stronger concepts. I have no doubt my memory models will never be displaced in my brain. They have been seared in protein, and stabilized for good. I will need to die for my ideas to die, in case they are wrong. I know they are not. I have churned through too much data to leave any room for doubt.
Curry also got his model, which seems to lead to the death wish for associationism:
In communication theory, information is the reduction of prior uncertainty. Organisms are 'uncertain' because they are composed of conditional adaptations that adopt different states under different conditions. These mechanisms can be described in terms of the decision rules that they embody—'if A, then B', or 'If you detect light, then move towards it'. Uncertainty about which state to adopt (to B or not to B), is resolved by attending to the specified conditions (A). The reduction of uncertainty by one half constitutes one 'bit' of information; and so a single decision rule is a one-bit processor. By favouring adaptations with more branching decision rules, natural selection can design more sophisticated organisms that engage in more sophisticated information processing, asking more questions of the world before coming to a decision
This is a lovely model, however, the evolution came up with a better one. Literal interpretation of the above description would allow the evolution to work on deterministic automata capable of handling humongous decision trees. However, the best evolutionary approach is the one that provides individual real-time adaptations in the form of network conceptualization and learning, and there is a fantastic learning rule based on association and induction: if B always seems to happen in association with A, then upon A, predict B. Each animal can build its own outwardly non-deterministic automaton for handling behaviors.
Associations separated in time
Curry continues:
For the rat, a rule is, "If you ate something and subsequently felt sick, then avoid that food in future"; it has no such rule fingering sunsets, nose twitching, or fighting, which is why it never makes those connections
I presume Curry's assertion here is that we need to be born with a specialist preset rule on food avoidance to be able to efficiently avoid food. That's correct. However, feeling sick and the offending food are linked by the mechanism Curry wants exterminated: association.
We are indeed born with a specialized circuit for determining acceptable foods. We can see it instantly when observing crawling toddlers who seem to have an algorithmic imprint for tasting all things around and building associations between color, shape, texture, and taste. The omnipresence of associative food avoidance in animals tells us that there is a deeper imprint that makes it possible to make associations separated in time: associate a berry object, its taste, and the feeling of nausea 24 hours later.
This is how I see it work using my concept-based model of the thought process: the active concept map "I have eaten" (working memory) is capable of activating the concept: "I have eaten a berry" (short term memory), and then "I feel sick" (working memory) instantly associates "berry" with "sick" sending a penalty to "berry seeking behaviors" (long-term memory). "Berry" in conjunction with "blue" associated with "tasty" will activate "food seeking" behavior, "brown berry" associated with "sick" with activate "avoid food" concept map. The separation in time is not a problem. All the rat needs is to condition his brain to keep "I have eaten" concept in the active state for long enough. The signal extinction rate would be based on experience (e.g. further prior associations). The degree of specialization in the brain for serving this particular purpose is unclear, but there is definitely a rule imprint associated with food avoidance behaviors indeed.
Innateness of ideas
I still see some of Curry's thinking problematic when we move to the level of humans:
The most likely answer is that humans have a range of innate ideas about the world (to do with colour, shape, forces, objects, motion, agents and minds), which they are able to recombine (almost at random) in an endless variety of ways (as when we dream), and then test these novel conjectures against reality (by means of the senses)
The above statement is largely correct, but I hear the echoes of Geary and Berch, or David Didau, or Hume's "Copy Principle". Innate ideas make it possible to build higher level models that have little to do with innateness, e.g. model of integration in calculus. One student may conveniently see integration as a fruit collecting process along a travelled road. Another will move on to abstract building blocks of calculus that lost all of its innateness. This fallback on innatism tells me Curry must be a strong nativist with a substantial underappreciation of plasticity (based on association). The same book ("This idea must die") includes a nice piece by Allison Gopnik that provides a solid counterpoint.
Long live Associationism
At the core of the creative progress in generating new ideas is that one connection: association. We call it an "association of ideas". I have already associated the label "Oliver Scott Curry" with "interesting" and with "not always right". Next time I see his texts, my attention will triple, and my creative circuits will jingle: ready for more associations. Like a rat which can easily recognize a tasty berry, I will easily recognize Oliver. This new skill will be born from the power of association!