Archive

Archive for August, 2011

Regressive Learning

Mon 15 Aug 2011 22:23:22 2 comments

Following my last post, I visited some friends of mine who have young children. As my visit carried on into the even I was present for the evening reading ritual where the children each had a turn at stumbling their way through to words, or in the case of the youngest, through the pictures. I began considering other times I’d see or heard about early learning. At that stage learning seems to be centred around the process of trial and error, and as the motto “no pain no gain” tells us, the emphasis is on the error. When attempting to speak a word, a child’s first attempt will often be far from the mark. It’s only through a process of repeated failures that they end up being able to pronounce it correctly. It’s as if they aren’t learning the right way to say the word, rather they’re learning a collection of wrong ways and then attempting to pronounce it from the possible ways that remain uncorrected. Once they have learnt to reliably pronounce the word they begin exploring its meaning by using it everywhere. Again, through a process of repeated failures, the meaning of the word slowly becomes entrenched in their minds, as something of a collection of not-meanings.

So we can model the mind as a thing with a certain chaotic potential, pursuing experiences in a seemingly random pattern. Each experience chips away at this chaotic potential to slowly reduce it in scope while shaping it into something that appears to contain positive knowledge1. Looking closer, each individual idea2 would initially be a bulky rough edged block, with the truth being some mysterious shape buried within. Each time failure for that idea is experienced a piece of the block would be chipped off until the block beings to roughly represent the shape of the truth within. Given learning is not going to be perfectly experienced, each negative experience will only chip off a fraction of the portion of the block that is offends the experience. Positive experiences become a process of hardening the block as it is, reducing the ability of further negative experiences to change the block.

In a sense what I’m describing is the way a neural net functions; beginning as network of nodes containing a collection of random values with each iteration of the feedback loop slowly tuning those values until the output becomes more reliably correct. The resulting collection of values is a distilled version of the input and output information from the “experience” of learning. There is a useful trait with this regression model of the learning process. It demonstrates the damage that can be caused by mis-learning something, both in terms of chipping off too much through a false negative experience (chipping off too much, requiring relearning from scratch) and premature hardening through too early and frequent confirming experiences (a misunderstood idea becoming resistant to later corrections).

In this model, complex ideas are the connection of the individual idea blocks, including both the blocks and the connecting material. The complex idea remains vulnerable to any fault in the underlying idea but also requires correction of the connecting material corresponding to any modification of the underlying idea. It would also be possible for the complex idea to be formed first, and the underlying ideas revealed with further refinement3. The end result being highly complex ideas are formed through a long process of refinement and remain vulnerable to a compounding element of fault in the individual ideas. Thus the more complex the idea, the more dependant it is on negative correction to become accurate, and the more it is at risk of premature confirmation.

It’s important to consider also what this complexity means. In this model, a relatively simple idea is one that has a direct feedback mechanism from which the brain can facilitate learning. An example of this would be the skills of hand-eye coordination, or the lesson learned from the pain caused by touching something appearing to be hot or sharp. A slightly more complex idea might be the act of walking or the vocalisation of a word. More complex again would be movement to a particular place or performing a social greeting. Engaging in conversation becomes more complex again. The idea(s) constituting the concept of language becomes built on this understanding of conversation. The idea(s) for general concepts of information and communication built further. The important pattern is that the more abstract, or the more generalised the idea, the more complex it becomes for the mind to learn. The more likely the mind will misconceive the complex ideas through a compounding of the faults in the ideas from which it is built.

This stands in contrast to how we build ideas when thinking and communicating rationally4. When thinking rationally we like to start with the general rules and use deductive reasoning to add detail and end up with a conclusion. When viewed in this fashion, it is the general rule that seems simple and the specific conclusion (observable experience) that seems like the complex result. So how do we model conscious and rational thought in a model of thought based on regression?

Through a continual process of regression, the mind will build up layer on layer of ideas until it learns abstract ideas that can constitute a form of logic. These mental structures form tools the conscious mind has access to, in the same way has access to muscle movements and sensory observations. This form of intuition logic won’t necessarily match the structure or accuracy of formal logic; however it would enable a form of conscious reasoning. By learning to use and trust these mind structures, in the same way we learn to use and trust our limbs and senses, we gain to ability to think at a level abstracted from immediate action.  The sheer complexity of the abstract ideas of logic and reason, in terms of how the mind learns them, means that the mind is going to be inefficient at using them and would continue to rely heavily on its natural method of regression. Rather than attempt to structure an idea within a structure of regressed logic elements the mind will examine small portions of the overall idea and construct a new regressed concept as the idea. Thus what may seem like rational or logical thought to the conscious thinker, is actually just a thin layer of rationality on top of a complex web of irrational intuition and regressed conceptualisations. We could take the model future and suggest rational thought is nothing more than the intuitive selection of the mental structures that have been constructed in association with the language elements we associate with rationality, where the intuition is guided by the experiences associated with the use of those structures.

In the model learning language (anything from English to mathematics or formal logic) becomes about learning the symbols and their associated rules in a fashion somewhat separate from the intuition based reasoning. Learning to use language becomes the regressive process of associating the symbols and rules with the intuition elements; given each person will have developed their own structure of intuition based reasoning, understanding and application of language will also vary. It will take time, for the mind to adapt to language to the extent that it’s use becomes intuitive. Rational thought is the application of these formal ideas to process and refine other ideas the mind contains. This in itself will cause the mind to train itself to better adhere to the formal ideas in the language, causing the language and the ideas within it to become more intuitive. Given the ideas of language are socially sourced, motivation to use language based rational thought must, at least in part, rely on social desires. The elements of language would be mentally formed in a similar process to the intuition logic ideas, slowly being refined with repeated failures, generally through social experiences.

This chaotic and regressive model of learning and thought is something that I’ve been developing for a while, particularly over the last year. Undoubtedly the model is influenced by my experiences with neural networks, and my experiences education more generally. I sense it has been influencing how I approach thinking about a range of issues, from education strategies to political philosophies, so I thought I’d try to get it out in a post. I’ve also enrolled in a introduction cognitive psychology class this semester, so it’ll be interesting to see how my ideas fit in with the academic theories and evidence.

 

1. It might be described as an identity or person, but that’d be a bit of a tangent from my current train of thought.
2. I use the term “idea” here somewhat abstractly as a concept of mental datum. An initial idea could be the meaning of a word when the word is first heard.
3. Noted because while I had the general idea I hadn’t figured out exactly where this post was headed or how it would get there until I wrote it.
4. Could be synonymous with consciously.

Categories: Uncategorized