The introduction of the book informed us about the components and goals of cognitive science. In this next chapter, Bermúdez takes us on a historical tour of the key developments across psychology, linguistics, and mathematical logic that preceded the establishment of cognitive science. Here, he focuses on four of these developments, namely those in behaviourism, algorithmic computation, linguistics, and information-processing models.
Incidentally, this is where Bermúdez explicitly places cognitive science in the information-processing tradition, asserting that "the guiding idea of cognitive science is that mental operations involve processing information, and hence, we can study how the mind works by studying how information is processed". As an ecological psychologist, I naturally disagree with this stance, but I reserve my judgment for now as I attempt to work through this material with an open mind.
That being said, this post will focus on behaviourism in psychology, and I'll come back to cover the rest of the mentioned developments in due time. Here, we'll look at what behaviourism is and the key studies that led to its eventual rejection in psychology.
Behaviourism was the dominant paradigm and approach to psychology from the early to mid-1990s. The fundamental assumption here was that, instead of worrying about unobservable mental states and processes, psychology should concern itself with observable, measurable behaviours. The most famous finding from behaviourism was that of conditioning, which is a mechanism linking responses and behaviours to certain stimuli. There are two main types of conditioning -- classical and operant.
In classical conditioning (see Fig. 1), an unconditioned stimulus (UCS) is accompanied by an unconditioned response (UCR). For example, a dog might start to salivate (UCR) when exposed to a piece of meat (UCS). These terms are slapped with the label 'unconditioned' because this stimulus-response requires no prior training or learning to emerge -- they are simply there beforehand. For classical conditioning to occur, a neutral stimulus (NS) is introduced alongside the US. Going back to our dog example, an experimenter might ring a bell (NS) every time a piece of meat (UCS) is brought to the dog. Like in the previous case, the dog starts to salivate (UCR) in the presence of both the bell (NS) and meat (UCS). The magic happens when you repeatedly expose the dog to both the neutral and unconditioned stimulus over an extended period of time. Slowly, the dog starts to associate the bell with the piece of meat, and eventually salivates at the ring of the bell when there is no meat present! At this point, the neutral stimulus (the bell) becomes a conditioned stimulus (CS), and salivation in response to the bell is termed the conditioned response (CR). Importantly, behaviour here is seen as an involuntary response to a conditioned stimulus learnt through a process of conditioning by association.
In operant conditioning, behaviours are influenced by their consequences. These consequences refer to either rewards or punishments. The basic idea here is that if you want to reinforce and encourage certain behaviours, follow up the behaviour with a reward. Meanwhile, if you want to discourage the behaviour, simply follow it up with a punishment. In the famous Skinner's box set-up (see Fig. 2), rats increased and decreased their button-pressing behaviour, respectively, in response to rewards (e.g., food pellets) and punishments (e.g., electric shocks, bright flashes of light) after the behaviour. Here, behaviour is seen as a voluntary choice that is influenced through a process of conditioning by reinforcement.
Learning without reinforcement
The two types of conditioning above led behaviourists to conclude the following. Firstly, all learning is a product of conditioning, and secondly, that conditioning can occur either through association (classical) or reinforcement (operant). Tolman & Honzik (1930) challenged these conclusions in a study about learning processes in rats. Here, they had three groups of rats run through a maze with the goal of finding the food box (see Fig. 3). When they did, the first group of rats got a reward, whereas the other two groups didn't. Using the principles from operant conditioning, rewards should strengthen the behaviour of finding the food box. Unsurprisingly, then, the first group quickly learnt to navigate through the maze, while the other two groups simply wandered around aimlessly.
The key manipulation came next. After 10 days of providing no reward, the experimenters started to reward the third group of rats every time they found the food box. Well, they probably just learnt how to navigate the maze like the first group, right? Yes, and astonishingly, they also learnt to do this much faster than the first group! Given that learning through reinforcement only occurs with rewards, the behaviourist account argues that no learning occurred in the first 10 days and that this only starts after rewards were introduced. How then can we account for the rapid learning rate observed for the third group of rats?
To explain this, Tolman & Hozik argue that during the first 10 days, the rats must be undergoing some process of latent learning. While they might not be directly learning through reinforcement, the rats must somehow be picking up information from their environments (the maze) and storing this in their mind. These bits of information can later be activated and used to promote a faster learning rate once rewards and reinforcements are introduced.
Cognitive maps
If information can be stored through latent learning, what type of information is being stored? On the one hand, the rats might be storing information about how the maze actually looks like, thereby building an internal spatial map of the maze and using this to guide their behaviour. On the other hand, the rats might simply be memorising the order of the movements. Think about how you might sometimes remember how to get somewhere not because you have an accurate map of the area in your head, but simply because you tell yourself, "go straight, then right... then left... then left again, etc".
So, which is it? Imagine a crossed path (see Fig. 4) with their respective north, east, south, and west poles. Now compare two scenarios. In one scenario, rats start at the north and south poles on alternating trials, with a reward awaiting them if they end up at the east point. Note how this requires different movement responses depending on their starting position. If they start up north, the rats have to turn left to get east, and if they start down north, a right-turning response is required. On the contrary, successful performance, regardless of starting position, would suggest place-learning, where rats store information guiding them east in some spatial map in their heads. In the second scenario, the rats once again start at either north or south poles from one trial to the next. This time, however, they are given a reward only if they reach east when they start up north, and reach west when they start down south. This means that to successfully get the reward, the rats have to turn left no matter their start position! This means that successful behaviour requires response-learning, since what is stored is not an internal map, but instead a sequence of responses.
And this was exactly what Tolman et al. (1946) did! In their experiment, they found that rats learnt the goal behaviour much faster in the first scenario (place-learning condition, instead of response-learning). The authors took this to be evidence that, at least in the case of rats-navigating-a-maze, information about how the environment was arranged was what was stored as internal representations in the mind. These representations, called cognitive maps, form an internal model of the world and are (somehow) used to guide the goal-directed behaviours of rats (and other animals).
Complex behaviours and behaviourism
There is one last bit that I will cover very briefly. It comes from a 1951 essay by Karl Lashley, who argued that the stimulus-response approach to understanding behaviour could not account for much of the complex behaviours we see. The problem is that this approach views behaviour as a linear, unbroken, and unbranching chain from cause (stimulus) to effect (response). Instead, Lashley argues that behaviours are organised hierarchically, with an end-behaviour influenced by many different subfactors.
He goes on to outline two hypotheses. The hypothesis of subconscious information processing asserts that while we might be aware of a higher-order goal or intention, the actual computations that form the basis of information-processing happen rapidly and unconsciously in our minds. Next, the hypothesis of task analysis argues that scientists can analytically break down a behaviour into its different subcomponents as an approach to studying that behaviour. The classic example of this is the study of memory. To understand memory, scientists routinely break down memory into different functions and sub-functions. From memory, we get the study of long-term and short-term memory. From long-term memory (see Fig. 5), we get episodic memory, semantic memory, procedural memory, etc. The idea is that studying these components gives you a better idea of the overall higher-order function (memory), especially once you combine the components.
This section really links to my understanding of the limitations of traditional, information-processing psychology. Lashley argues that behaviourism takes too much of a serial and linear approach to studying human cognition, and then goes on to propose ANOTHER linear approach as a solution. Yes, his response might be more complex in the sense that it introduces a hierarchical structure, but this still depends on the assumption of linearity and that you can simply combine different components in a purely additive fashion to get the 'higher-order' behaviour. For more on this, I've covered some issues of assuming linearity and additive combination in a separate post.
Some thoughts
There is a lot in this section that I disagree with. The appeal to internal representations, the alleged need for information processing, the linear decomposition of tasks etc. These will be recurring themes as I work through the book, so I'll try not to repeat myself so often. That being said, I will also be the first to admit that I don't have an immediate answer to how we can explain the limitations of behaviourism from an ecological perspective.
What I will say for now, though, is that I do agree that the stimulus-response explanation is far too simplistic to account for the wide variety of complex behaviours we see animals perform. What I disagree with is the explanations that cognitive science has put forth to account for this. There has been a lot of talk about 'information' that is stored as representations. Off the top of my head, I suspect that ecological psychology will also attempt to explain these limitations with the notion of information. The key difference is that 'information' here specifically refers to ecological information, that is, higher-order kinematic invariants in the energy media that are (a) created by dynamic properties of the world and (b) are specific and map 1:1 to those dynamic properties that created them. There is no need to process ambiguous information (in the cognitive tradition) if your information already uniquely specifies the world. Only detection is required.
References
Elliott, M. H. (1928). The effect of change of reward on the maze performance of rats. In University of California Press eBooks. http://ci.nii.ac.jp/ncid/BA18664955
Lashley, K. S. ( 1951 ). The problem of serial order in behavior. In A. L. Jeffress (ed.), Cerebral Mechanisms in Behavior: The Hixon Symposium. New York: Wiley.
Oyigeya, M. (2021). Reflex memory theory of acquired involuntary motor and sensory disorders. The Egyptian Journal of Neurology Psychiatry and Neurosurgery, 57(1). https://doi.org/10.1186/s41983-021-00307-2
Simply Psychology. (2024, February 1). Classical Conditioning: How it works with examples. https://www.simplypsychology.org/classical-conditioning.html
Tolman, E. C., & Honzik, C. H. (1930). “Insight” in rats. In University of California Press eBooks. https://ci.nii.ac.jp/ncid/BA18692881
Tolman, E. C., Ritchie, B. F., & Kalish, D. (1946). Studies in spatial learning. II. Place learning versus response learning. Journal of Experimental Psychology, 36(3), 221–229. https://doi.org/10.1037/h0060262
No comments:
Post a Comment