What does it mean when people say that not even the creators of large neural networks or AIs like ChatGPT or Midjourney fully understand how they work?

Question

I’ve heard this sentiment a fair bit in recent years and I’ve never been able to find a good answer online.

In: Technology

Answer 1

AIs like ChatGPT and MidJourney are types of Neural Networks.

Neural networks are made of one part code and one million parts training data, which is then compressed into a “model” which is somewhere in the middle in terms of size.

Many people understand the code, but it is much harder to understand the models, mostly due to their sheer size, and also because they change so often, either due to tweaks in the code or introduction of new datasets.

Unlike traditional programming, which was more based on control over the machine, machine learning approaches are not concerned with controlling or understanding the machine, rather they are concerned with achieving results by supplying massive amounts of computational energy.