The Similarity Engine

The first computers were called Difference Engines, after their mechanism to perform complex calculations all based on adding and subtracting numbers: their difference.

At heart, computers have not changed. They still add up or subtract to do all their work. Increasingly, though, the interface to the human race has changed from a numerical one to one of programming language, which the computer subsequently inverts back to numerics, its only working medium.

And then someone had the radical idea to stop using programming language and let computers loose, without definitive control, on human language. They were pointed at our species’ entire record of the written word, and caused to understand it all in some way. How, exactly, no-one knows. We gave the computer systems enough learning capacity to make their own way, to develop their own algorithms and just get on with it.

Those systems, which we named Artificial Intelligence, developed algorithms which reduced language back to numbers. Back to differences. They would look at a string of words and work out what the probability of the next few words would be, choosing the most likely. That is what AI does, and all that it does.

In some ways, it is a huge breakthrough in statistical analysis, and as such, is driving many new advances in technology. It can take human-type problems and apply a huge firepower of statistical analysis that would be unthinkable to humans. Literally, because humans will always attempt to reduce big problems by making intelligent assumptions about the answer and narrowing their workload. Computers don’t need to, so they are not limited by our own assumptions. So, artificial intelligence does its best work by being stupid.

In other ways, it is a huge step backward for civilisation, because, away from the technological arena, AI dumbs everything down to the average. The next few words it seeks, those which have the highest probability, are those which are average. Not the outliers, not the strokes of brilliance. If you look at wannabe writing groups on social media, you will see people who want to write, but don’t want to study the craft, announcing that they have just told ChatGPT to write a scene, or heaven help us, a book, and the results are great. Those who know their writing craft will try AI out of curiosity, look at the results and know that the strings are words without depth or feeling. AI is unable to process either.

I think much of the problem we have as a global race dealing with this emergent technology is that two very different applications of the same technology yield very different results for humanity. When we look at technology, AI has much to offer. When we look at art, culture, human wisdom and sentience, AI is a threat, a lowest common denominator, an entity which will devalue our arts while diluting, polluting and corrupting our knowledge base. 

Perhaps we should think of this technology not as Artificial Intelligence, which imbues it with properties it does not have, but as the Similarity Engine. It seeks the most likely answers based on historical information. Great for deep probability research, but pretty dumb at the things humans do best. Being human. Being alive, artful and compassionate.

By writing, by being human, by doing all we can to protect our written human record, we can help humanity adapt to the technology. It cannot be displaced, it cannot be avoided, but if we can accept it for what it really is, and do what we can to protect our human heritage from its hungry, uncaring maw, maybe we can limit its harm and enhance its good.

_{Image: Gaussian word curve by Rachel Whitby}