May 04, 2017

Memory Retrieval Model

It is difficult to decide where to begin when attempting to create an artificial mind. But like any problem the size of an elephant, the only practical way to approach it is to begin by feeling our way around bit by bit. Since the mind needs to store information, lets begin by attacking the problem of memory.

Remember me?

Without our memory, we’d be utterly useless. We’d wake up every day as if it were the first time we ever awoke. We’d be born anew each morning, which sounds like it would get tiring very quickly, but because we wouldn’t be able to retain any information, we’d never get used to it and so would be having a new experience every time. We’d live blissfully for the rest of our days continually experiencing the wonders of life. It would, however, ultimately spell certain doom for our species. For if we were left being amazed every day by the things we find inconsequential today, we’d not make any progress. When our grumbling tummies would remind us to eat, we’d struggle to remember where we left the food. So we’d go looking for it. When we came across something, anything, we’d wonder if it was edible and whether this was the food we were searching for. Ultimately, we’d have a pretty tough time of it. It’s safe to say that without it we’d cease to function.

So lets go ahead and recreate this life sustaining property of our brain.

Referencing saves space

The human brain can store vast amounts of information, but to efficiently store this information, it cannot store it redundantly. Every memory is constructed through the firing of multiple clusters of neurons. The brain is able to store information more effectively by linking together the common features of memories rather than storing them individually. Therefore, when we think of a concept such as chocolate fudge brownie ice cream, we don’t have a single neuron for that. Instead, the neurons for more simple concepts such as ice, cream, fudge, brownie, chocolate, etc all fire together to create the combined output of my favourite ice cream flavour. But these can also be reduced further to even more fundamental concepts such as ice becoming water, cold, freeze, and so on. The more we break it down, the less information we need to store to accommodate for the more complex concepts further up the scale.

Let’s explore how a model based on references is far more efficient than storing individual memories.

I’m going to recall a certain memory of mine. It’s a fairly standard memory and whilst brain-to-brain interfaces don’t yet exist, hopefully it is descriptive enough for you to imagine something similar along with the feelings and emotions that may come with it, although in this case I have chosen to be very objective and have left the fluffy stuff out.

We can break down the memory into a word net that is associative yet still retains all of the important information from the memory.

Nobody cares about how you feel

It’s important to talk about why I left out the fluffy stuff for this. My view is that whenever we recall a memory, we don’t necessarily recall the emotions we had at the time, but rather relive the moment and conjure up similar emotions, since we are, hopefully, still the same person. Therefore, storing emotive states is inefficient if we can just recreate them from the same stimuli.

On a side note, this is probably why when we’ve moved on from certain points in our lives and remember back, we don’t always conjure up the exact emotions we felt at the time, but rather are confronted with the emotions we have based on what we know in the present. For example, when you were a kid, it’s highly likely you did a bunch of questionable things. At the time those things may have seemed completely normal and even fun in fact. Now that you’re supposed to be a grown up and it is no longer acceptable to go trick or treating … without a child as cover. It was probably one of the funnest things you did as a child. Going around knocking on doors demanding tax in the form of candy, and then eating all of the candy when you got home - bliss. When you remember it though, and you feel like you can remember how happy you were, I would argue that it’s not that you’re remembering the emotions you had at the time, but rather you’re reacting to the way certain things make the present you feel. It’s not quite recalling, yet we still feel the emotions as if we were reliving the moment. There’s a term for it - nostalgia.

I believe the brain uses tricks like this all the time to allow it to function more efficiently. Whenever it can recreate, it does, so only the most vital information is assigned to memory.

The simplified model

The following is a very simple model to allow us to represent memory in a holistic way. The collection of nodes would represent the complete working memory of the organism to which it would be assigned. Of course the default one here is very basic and I would suspect the organism itself to be pretty useless, but it is good enough to make a start.

It is possible to recall memories by entering an existing term in the top left box. When an existing memory is recalled the connecting nodes are activated. This results in an output which is represented by the list of words that appear in the bottom right. This collection of words ultimately represents the organism’s understanding of the memory that recalled them.

Idea Links

An interesting part of this model is that it also allows for the adding of new concepts. To add a new node or idea, fill in the box on the bottom left. If this idea is linked to anything else, you can input the link too. For multiple links, simply separate with a comma. Adding this new concept adds it to the entire collection of memory. When memories are recalled, this newly learned concept can now easily also be part of the output. So, technically speaking, this model fills all the basic requirements of an artificial memory module. It is associative, like human memory. It can also accommodate new ideas and concepts similar to the human mind. Crucially, it is also not very complex, which means for a simpler overall model. Simple is good.


Short Term Memory

There are very practical reasons for having a section of our brain responsible for long term memory as well as an independent section for short term memory. We'll explore these benefits...