Nice little cartoon illustrating the GIGO principal (Garbage-in, garbage-out) What most people don't understand about the current state of artificial intelligence software is that it's really, really narrow in scope. For example it's common to see AI tackling problems such as a board game like chess and playing such games to a very high standard. This is because the rules of chess are very well defined and the problem space is very narrow, i.e. once you know the rules, you have everything you need to play. Take a more "life-experience" oriented problem, like consulting a medical patient or digesting a legal case, these kinds of problems require all kinds of tangential expertise, for example having an intuition about the state of mind of the person involved and things like dealing with exaggeration or lying etc. There are no "data-sets", i.e. collections of "results" of the form "X happened and Y was the outcome" in order to train a statistical or "AI" model with sufficient variation and breadth to be of much use in more broad and general use-cases like this. The current technology doesn't even get close.
No comments:
Post a Comment