Gigaom AI Minute – May 5

Transcript

We recently surveyed Gigaom's audience about their views on artificial intelligence. Our audience it primarily made up of business people that have an interest in technology and I would like to use the next several AI minutes to explore the answers.

The fifth question has to do with the right to know. It's the idea, according to European legislation, is that if an AI makes a decision about you--like it denies a loan--you have the right to know why it made that decision. So, we asked our audience, "Do you think that should be the law where you live?" Seventy percent of the people said, "Yes, I think it should be."

Then we asked, "Do you think that, if that were the law, would that limit the development of AI?" And only 1-in-10 people said they thought that it would. Now, I find this pretty interesting because putting the constraint that it has to be explainable, almost by definition, limits the kinds of things that AIs can do. Right now we use unsupervised learning to identify clusters and large pools of data, and we associate those clusters together, and there may not be a why behind it other than that AI put you in that cluster. So, there may be a disconnect between what people want and their expectation of essentially what it's going to cost.

Now, when asked, "Is it impossible to pass that law?" in other words, "Can an AI always tell you why a decision was made?" only 1-in-20 people said it's not possible. For certain kinds of systems, though, which reach a certain amount of complexity, it may not be decipherable. People, for instance, don't know why we make the decisions we make, but, it isn't necessarily going to be the case that you can just open up a machine and look inside it and understand why it came out with the solution that it did.