The joy of Machine Learning

There is something immensely satisfying about watching Machines repeatedly fail at simple tasks.

The coming changes to our economy have been heralded as an inevitable displacement of feeble humans with more capable machines. This is already happening in factories everywhere where robots steadily outperform humans. We watch as AI slowly learns to drive cars and trucks adequately. Our politicians desperately search for a magic pill to convert our current work force into an army of John Henry’s. After all, our livelihood’s are at stake.

Though, from the view behind the monitors, things look a little different. What has been edited out of this public discussion is the hundreds of millions of computer hours in training that each development may require. Not only are the initial conditions of an AI fairly pitiful, the learning takes forever, and to top it all off: little to no knowledge gained is transferable.

Let me say that again: nothing learned, nothing gained. Each computer model is usually thickly mired in its own unique environment. This is precisely why the Alexas and Siris are so fragile: anything outside of their normal operating behaviour, anything new, turns them into bricks.

On the other hand, watching the learning portion may be therapeutical to anyone who feels left behind by this computer revolution. It certainly has been for me.

This is how I lost my fear of a Singularity or other AI apocalypse. Now if these driverless cars could just learn how to stay in there lane, then maybe we can learn to live-and-let-live in the future.