I couldn’t think of a more appropriate way to start of an initial post than to highlight the most common association between the real world and science fiction – The Terminator. It’s funny really – as you learn more and more about machine learning models, you’ll realize how far we are from the Matrix and Terminator. That said, there are things that give the illusion of general intelligence – or AGI (Artificial General Intelligence) — but those are all parlor tricks.
As you get further into this website, you’ll learn about models, dynamic model routers (which we used to call pre compilers when I was playing with this stuff), and you’ll see that AI models do what they are trained to do – generally one thing – really well. Need an AI to do two things really well? Train two separate machine learning models and stick a routing program in front of them to determine which of the two ML models is better suited to solve each problem put in front of it.
We hear about the US Air force using AI to help prosecute targets on the battlefield in simulated exercises, only to discover that the AI destroys the command post where its human oversight sits (to avoid aborting a mission and completing the objective since a dead approver can’t say “no”). Scary, but not SkyNet, Master Control, or The Source.
My hot take: To make a truly Artificial Intelligence, we have to have a much better understanding of what Organic Intelligence looks like and how it works. There was once a website — the straight dope — who had a slogan: “solving stupidity since 1991 (it’s taking longer than we thought)” — if we can’t solve general organic intelligence how can we do it artificially?
~DT