Building Intelligent systems: biological and software

There are amazing analogies between computer processors and the brain, and it is just a matter of time before the algorithms that define electronic processors mirror the functionality of the chemical processors of our brains. The biological neural systems rely on inputs and pattern recognition to learn. Computers are excellent at storing (remembering) facts, and are becoming proficient at recognizing relationships as defined by statistical patterns and neural networks. However, computers cannot yet create metaphors or learn how to independently process new information.

Metaphors are an important part of how humans think. We understand new information as we draw parallels and connections between it and prior information. Over-simply put, we learn by recognizing relationships between new information and old. Computers could do the same when their sets of data include enough relevant fields to be able to computationally identify systems that are described by similar dynamics. In other words, when a computer has data about how things works, it can find systems that work similarly to each other. The leap from there to drawing metaphors is a data-mining process: it can be solved by computational rote, where statistical relationships are identified, prioritized, and used for prediction.

This process also leads to “learning” about how to process new information. By recognizing the metaphors, new data can be classified and described according to how it is understood. And just like in our brains, there will be errors. Misunderstanding will occur as metaphors are calculated based on incomplete data sets. As more data is input, systems will have to be able to make corrections and re-calculate all of the other metaphors that included the corrected data. New corrections will be made and a cascade of corrections will result in a modified historical data record. A large number of calculations and recalculations will occur with each new input, and the storage of historical data (and calculated results) will require substantial processing and storage.

Recognizing metaphors will allow machines to output statements like: “It appears that ABC is driven in many similar ways to XYZ. The result we are seeking might be accomplished by A because a similar result was achieved in XYZ when X was applied.” Put more simply, computers will be able to express creative suggestions.

Interestingly, storage could be massively reduced by deleting large volumes of data that support the relationships that are strong enough to overcome some threshold level of certainty. For example, if everything falls, then we don’t have to keep all that data, just the relationship that everything falls. This may be analogous to forming intuitions.

Leave a Reply

Fill in your details below or click an icon to log in: Logo

You are commenting using your account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s