QUESTION: Marty,

It occurs to me that some of the most persuasive individuals of history (good or bad) have been so effective because they possessed a keen understanding of Human Nature allowing them to predict the response of their audience, and in turn, coerce them toward a desired action.  The question I would like your thoughts on is this: Do artificially intelligent systems have a “nature” in common making their responses predictable, and therefore exploitable? Thanks for all that you do, there are few opinions I value as highly as yours.

P

ANSWER: First, you have a lot of charlatans who claim to be using AI (Artificial Intelligence) that is really just a program following a predetermined script. That is not the real thing by any shape of the imagination. Real AI takes an awful lot of money, time, and resources no matter what fancy name they give their program. That type of “expert system” masquerading as AI is very predictable and can be easily reversed engineered.

If we are talking about the real thing, the answer is no possible way. I wrote the code for our system. It took me decades. To accomplish something that can accurately forecast BREXIT, Trump, Italy, and Hollande stepping down in France, would be absolutely IMPOSSIBLE to code or reverse engineer. Oh, I could feed in the US election data and come up with an isolated cyclical forecast. But how do you do BREXIT with no past track record of such referendums or that of Italy? These types of forecasts are extrapolations based completely upon all trends globally. There is no single line of code that would read IF-THIS-THEN that ELSE this over here. It is weighing so many variables it is really impossible to reverse engineer that.

The result of a real AI system, or better described as machine learning, is really giving the computer the capacity to learn and in doing so we create a real thinking process, minus the emotion. How it will arrive at a decision on say BREXIT is different from Trump winning. Each process is a different branch. Very interesting from a programming viewpoint.