It is thought by many that an artificial neural network (ANN) is a reasonable equivalent of a real neural network. A simple example of a real neural network – the two neurons that trigger each other and make your heart beat. The delay between beats varies with your mental and/or physical state. Let’s make the artificial equivalent. You can’t. The ANN has no concept of time – it is just a bunch of resistors. You can put values into the resistors, you will get a value out, but the setting of the values is handled externally to the ANN. This means it can only handle a very limited class of problem – one where time of event or calculation of value from multiple sources, or memory of a sequence of events is not required. If you asked an electronics engineer to create a cell phone from a pile of resistors, they would tell you it is impossible – you need switching, inductors, capacitors, memory. People are expecting an ANN to simulate the workings of a brain, when it cannot do any such thing – it can do remarkably little.
But we can do all this other stuff externally!
Then you have lost the great power of a brain – the ability to integrate many inputs while monitoring for subsystem failure – you are reducing the problem to suit the tool.
But what about data – they are good with data – right?
It would be appealing to say we are using AI on our operations. But that comes with a cost – you have to change the problem to suit the limitations of the tool. Let’s say you were making button up boots. You had AI running the data – it was all looking good. The one piece of data that wasn’t being checked was whether someone was patenting a new closure method – like a zipper. It is outside events that usually kill a business, like an advance by a competitor, and the lure of AI reduces the desire to check outside influences.
The ability of data is also greatly overhyped. When someone offers that all you need is data, ask them if they look through the windscreen when driving, or is the data in the rear vision mirror sufficient. An ANN is effectively a rear vision mirror, it has to have happened before and the pathways are already built, it is not a forward-looking tool, detecting and either avoiding or warning of looming problems that haven’t happened before.
An important part of problem-solving is to make sure you understand the problem and the limitations of the available tools.