Using Artificial Swarm Intelligence to solve complex math problems

Long before ChatGPT was a thing I persuaded OpenAI to let me help train their GPT model. My specialist subject was predicting if the UK interest rate would rise or fall over both the short and medium terms. I also tried to use GPT to predict if property prices would rise or fall subject to economic conditions. I soon discovered that generative AI models are not particularly good at complex math. They predict the answer, rather than work it out. I wanted an AI that could process complex math problems by applying novel and optimised methodologies. I wanted the computer to develop a solution itself, rather than me program one.

I spent months trying to find a solution, fortunately I am hooked on cosmology and philosophy so I know my way around nature a little bit. By studying the behaviour of other types of intelligence I concluded that swarm intelligence might hold the answer.

In fact there are a number of artificial swarm intelligence experiments that you can watch on YouTube. However, they all seemed to have a number of common problems. So you can see the processing taking place all the experiments run very slowly. I believe that the best solution must run as fast as possible, thousands of times natural speed. Secondly, the “things” swarming in the experiments are all exactly the same, I thought why? In my concept the dants (digital ants) are organised into a large number of groups, with each group of dants having different abilities – rather like real ants, they have different jobs.

Also the environments that the experiments run in are all very plain. I believe that the environment (or hive) could be rich, meaning that it contains complex rules that cause the dants to swarm in a certain way, and to stick together to form “process chains”.

The question is, can artificial swarm intelligence solve complex math problems where generative cannot? I believe it can. My colleagues and I are building a new AI architecture called NEXUS, which is based on artificial swarm intelligence. This is a completely different approach to the neural networks found in generative AI. NEXUS will build optimised processing chains rather than predict an output through a neural network.

NEXUS will be rubbish at text and image processing, but when it comes to solving complex math problems, I believe it will excel. This is important when you are landing a space ship or designing a new drug.