Will We Have AI Whisperers?
To Divine AI Decisions That Confound Us
Some people are amazingly effective at dealing with dogs. Unruly canines that won’t obey their owners happily listen to these people. It’s as if they have a mysterious power that lets them divine whatever is going on inside that dog’s head. We call such people dog whisperers.
I was reading something about hallucinating art forms made by AI, and that got me wondering: what about an AI whisperer? Someone who is effective in understanding Artificial Intelligence?
“Understanding an AI?” Your brows knot. “Now, wait a minute,” you say. “Why would someone need to divine an AI?” And that’s a fair question.
After all, an AI is not an animal with a brain whose inner workings we don’t understand. We, the humans, created AI. Surely, we know how AI works. Just look at all the intelligent systems we have working for us: systems that decide who makes parole, who’s approved for a loan, and who gets hired for a job. If you want to know how that system arrived at its decisions, all you have to do is look at its source code, the mathematical model underlying its decision making process. It’s all very logical and straightforward. It’s easy to understand the reasoning behind a computer’s decision.
Do you still think like that? (the paragraph above) If you do, I have some news for you. The world of AI is undergoing a revolution, thanks to twin breakthroughs of neural network and deep learning, or together, DNN (deep neural networks).
I went in more detail about these technologies in an earlier post, but basically we are building computers that simulate brain with millions of interconnected electronic neurons. These electric brains are then taught skills—translating a language, recognizing pictures, driving a car—like humans, by trial and error, by analyzing huge volumes of training data.
The problem is, we don’t know how these AI systems are learning. During training, an AI continually readjusts thousands of internal parameters until it can reliably perform a task. But we have no clue what exactly is happening inside that electronic brain. It’s a radical, fundamentally different way of programming computers.
Take for example, apps that use DNN to serve ads or recommend songs. If you ask their creators how the apps arrive at their decision, they won’t be able to tell you. Those apps have programmed themselves by interacting with you. They have figured out your preference on their own.
And that takes us back to my original question: Will we have AI whisperers?
The rate at which AI is advancing and worming its way into human civilization, it won’t be long before they are ubiquitous, everywhere. That will increase cases where AIs will makes puzzling decisions (hopefully not dangerous), which in turn will necessitate the need for individuals who can quiz these faulty machines, and figure out how to tame the unruly AI. Not very different from a dog whisperer, is it?
More @ RigelBlu.com