Modeling the Mind

In psychological science, there are different approaches, or perspectives, different people bring to the field. As a science, psychology is descriptive. A true scientist doesn't actually 'prove' anything, they use methodological rigor to seek a consistent result under a particular set of circumstances.

Truth is consistency.

These results then serve as evidence to theories. These theories are what most people think of when describing findings. 'Study shows that sleep is important to memorization.' Well, yes and no. The study showed that people scored higher on a test if they slept beforehand. We extrapolate from that that sleep is important to memorization. (Alternative extrapolations could include: sleep reduces stress, sleep deprivation impairs general functioning, etc. These possibilities are narrowed down by repeating testing the same hypothesis with different methods that control for alternative explanations.)

But then comes the question of how. And in attempts to explain multiple findings and multiple theories, psychologists develop models. Although few scientists will disagree with the raw data of a well-designed study, they may disagree on which model best accounts for results in the larger scheme. Models are statistically tested for goodness of fit, which helps lend some objective credence to them, but ultimately they are based on incomplete information. Most of these models are focused on a concept, and highly abstracted, such as the various models of memory. They look kinda like this. Here's a more detailed explanation of one you may be familiar with.



Based on my own criticisms of various models, including the aforementioned issues of overfocusing and abstractness, I started developing a general model of mental processes some time ago, based partially on the biological structure of the brain, cybernetics, and personal theories of the significance of pattern recognition to learning and intelligence (to be covered in another post). In explaining it to a professor, she remarked that it resembled connectionist models, and from what I have researched thus far on connectionism, I am inclined to agree.

So, disclaimer: what I'm describing here as resembling connectionism is actually my own theory, which has apparent convergent conclusions with connectionism. As I learn more about how this does and does not conform to the communal understanding of connectionism I may make another post comparing the ideas.

The general idea is to look at the brain as a network. At the basest level, this consists of connections between neurons. These connections are of varying strengths, and change and adapt in response to both internal and external stimulus. Individually, each of these connections is meaningless, but the patterns they form begin to take on function. Mental activities are patterns within that network. To use an imperfect analogy, it's like how an individual bit in a computer means nothing, but put them together in a particular pattern and - voila - you have a working program. And as new connections form, old connections weaken, the patterns slowly shift.

Species are genetically coded to form a similar (though not identical) base network (our nervous system). But through a combination of genetic differences and environmental differences (including causes of mutation), our final networks are not identical. Hence, individuals are different.

The complexity of this system is astounding. Well beyond our current computational ability, let alone cognitive ability, to completely map and understand the interactions of. Thus, we address it at a higher 'level', in the form of psychological concepts. Much in the same way that we will describe running a program by saying, 'run smutcannon.exe from the C: drive' rather than detailing the very particular binary sequence that occurs within the computer to make that happen. (For a much more complete and detailed explanation of the idea of network 'levels', and the effects of looping within them, I recommend reading Godel, Escher, Bach. It's one of my favorite books for a reason, and helped me formalize these ideas.) But while we cannot visualize the exact processes, it is our understanding of the method of interaction between elements that is important in models.

Analogies to computers are common, and among the easiest metaphors to use, but neural networks are even more complex than computer programs. Whereas each individual unit of a program has a single two-stage state (bits), neurons can form multiple connections to other neurons, each with a variable gradient of strength. Neurons have a threshold that when reached cause the neuron to fire signals to connected neurons ("action potential"). The patterns of these firings run through this network in a complex chain reaction that manifests as our thoughts, actions, and other processes. Sound complicated? Here are some connectionist models of memory for comparison (warning, . Those are grossly incomplete and oversimplified. The same can be said of all models, though the degree of magnitude here is orders above.


^ This one, I believe, is one of the most accurate, because it doesn't label the nodes as concepts. Nodes alone have no value. It is the activation of particular patterns between nodes that hold meaning. And remember that not all of those lines are equal to each other.

My perception of the structure is that you can look at these models at multiple 'zooms' as well. Look closely at one of those nodes and see that that 'node' may in fact be its own network of strongly-connected nodes, and so on and so forth. It's that 'thinking in levels' thing again (seriously, go read GEB).

That the brain is a neural network is not really debated. We know the brain is made of neurons and that they communicate in this way. The difference here is how its represented in theoretical models, how we relate scientific findings to what we know about neuroscience. The above models are still abstractions, however the intention is that they are isomorphic abstractions - that is, they can be related on a (closer to) 1-to-1 basis with the physical reality of the brain, rather than having to be translated through subjective concepts and the definitional limitations of constructs, such as "long-term memory".

It could be argued that this perspective is too complex. That the idea of modeling is to made something understandable and eliminate the need to talk in obstructively convoluted terms. I'm not advocating the elimination of other models per se. Rather, I believe models need to use such a perspective as their foundation - to be able to isomorphically map onto the physical brain. In physics, there is a goal to find the "theory of everything", a single framework which unites all known aspects of physics. Ultimately, psychology is not any different, it is just less talked about  because the elements are still so uncertain. Isomorphic models are a logical necessity if we are develop a comprehensive model of the brain, i.e. the mind.

The earlier such a stance is adopted, the better. Not just to save conversion work in long run, but because the fact that it does directly map to physical reality means that it is a better jumping-off point for exploring new concepts and completing our understanding of existing ones. Really, if everything is connected, there is no difference between the two.