Judgment Day? (reprinted from the Metro Spirit 3/28/13)
Every once in a while, you come upon a headline that makes you scratch your head. No, I’m not talking about the Wired.com article this week, “Want to Make an Alligator Angrier Than Normal” (uh…why?) or the Cnet article “Your next phone’s screen will be incredibly strong.” (Really? Like, duh.) The headline that really made me wonder if we are traveling aboard some unstoppable force toward an immovable fate was another Wired story, “Darpa Sets Out to Make Computers That Can Teach Themselves.”
First of all, did these guys NOT see Terminator?!? Remember, the whole SkyNet artificial intelligence thing becoming self-aware and trying to wipe out mankind? You know, John Connor? Arnold? “I’ll be back?”
Apparently none of those things ring a bell. Instead, DARPA kicked-off a 46-month development effort called Probabilistic Programming for Advanced Machine Learning, or PPAML. According to Wired, Program Director Kathleen stated that the goal of the program, “is that future machine learning projects won’t require people to know everything about both the domain of interest and machine learning to build useful machine learning applications.” DARPA wants to make it easier for non-experts to build machine-learning applications.
Now, I get the jist of the principle. While in college, I developed some very crude genetic alogrithms, which I suspect are related to instantiations of PPAML components, to solve some relatively simple orbit trajectory problems. While extremely computationally intensive, the genetic algorithms were effective in optimizing solutions without the need for multi-dimension calculus, least squares linearization or even knowledge of orbital mechanics theory. These algorithms can get you to the “what” without knowledge of the “how”, or even more importantly, the “why.”
From one point of view, these algorithms let us explore the art of the possible. Instead of just dreaming the future, machine-learning can quantitatively show us what can be real. The process of discovery can be better managed as computers help navigate between true discovers and dead-ends. These are powerful tools that can move us forward a great deal.
On the other hand, a path to knowledge without full understanding creates long-term problems. Subject matter experts are important in every field in order to guide organizations along the healthiest path. No matter the program, the age-old “garbage in, garbage out” principal applies. Somebody needs to ensure garbage doesn’t go into the machine.
Unless, of course, the machine itself gets to a point where it starts making garbage-in, garbage-out decisions for itself. In that case, we’ll have to change the name of PPAML, to “Please Pray, All May Be Lost.”
An Augusta First – BTW, I heard my first college joke about GRU this week:
A UGA student, a Georgia Tech student, and a Georgia Regents student all go into the men’s room (yes, they’re all guys…roll with it).
The Bulldog does his business, then washes his hands, then completely dries his hands with a truly profligate amount of paper towels.
“Georgia Bulldogs are trained to be thorough,” he explains.
The Rambling Wreck does his business, then washes his hands. But he uses a minimal amount of paper towel, while making sure his hands are as completely dry as the Bulldog’s.
“Yellow Jackets are trained to be thorough and efficient!” he explains.
The GRU student does his business, and walks out without washing his hands!
Flabbergasted, the UGA and GT students demand an explanation.
“Jaguars don’t pee on their hands.”
Until next time, I’m off the grid. @gregory_a_baker