Coming back from the ISCEE conference in Eilat a few years ago and having a conversation with one of my friends, it occurred to me why some concepts of AI are so familiar. They are basically like numerical methods for solving unknowns variables. In undergrad, one of my stronger courses was Numerical Methods. Stronger meaning relative to other courses. My mark in the course was above my average, which was very low.
There was something elegant about strategies taught in Numerical Methods, such as guessing the answer and testing if your guess was close to the solution. Then you’d guess again. We learned about binomial x, y, and then going to extremes. What’s the limit of x as it goes to infinity? To zero? For the latter two, I now use this to elicit extreme scenarios in life. What if I had an infinite amount of money? What if I had no money?
It might have been a walk when I was tired, or the desert air where the ISCEE conference took place (Eilat) that this finally clicked for me.
Sometimes it takes me awhile for things to click… it was only in my early twenties that I realized cows had to have been pregnant at some point to start producing milk. I had just never thought about it. I digress…
In this case, it was that the AI label has been given some mysticism but what AI really is — just some new math and computing techniques for finding an answer. That answer might be the start of a pattern, the probability of a sound being a phoneme or a word, or an image being a cat.
Those coming up with terminology in the field of AI have done a brilliant job of adding to this mysticism. “Deep neural networks”… it sounds like we’ve created an artificial human mind. Oh, and it’s so deep, you shouldn’t try to underestand it. It could have been called something like “decision tree matrix” and nobody would think it were special. Even Machine Learning sounds like a steampunk-looking object reading a book and magically doing something. If it were called something like “comparative analysis” it wouldn’t attract so much attention. We could even compress “artificial intelligence” to “pattern recognition” and not lose much meaning.
Saying all this does not take away from the impact that AI will have but the real impact can be credited to now being able to access much more computing power at significantly lower cost so that using new numerical methods is viable. The result is that we can do speech recognition reliably in only hundreds of milliseconds, natural language understanding just as fast, and image processing on a larger and faster scale.
For example, people would die of boredom if they had to tag millions of photos as to whether or not they contained Coca Cola logos. However, with AI-based processes, Coke could look at all Instagram posts and identify influencers who are already fans of their products to target for ads or other marketing effort. And the machines win.
So let’s take the magic out of AI. Not because it’s not leading to a revolution but so that we can make its tools accessible to all of us. Things that we don’t understand can be scary and it’d be a waste of the potential benefits.