In May, 1997 the world's chess champion, Gary Kasparov, lost a 6-game challenge to Deep Blue, an IBM supercomputer. An advantage to the computer's programmers was the full history of Kasparov's previous public matches -- his style -- as well as that of dozens of other grand masters and their moves in various opening-, middle-, and end-game scenarios.
To give Deep Blue credit, its moves were the result of algorithm-driven analyses (e.g. how important is a safe king position compared to a space advantage in the center, etc.), according to the results of 700,000 or more grandmaster games. The machine could explore up to 200 million possible chess positions per second in this way. This wasn't just a memory exercise.
The rest is history.
Since then, developers have combined various forms of data (financial, scientific research, social, etc.) with application algorithms and massively parallel processing to aid human observation, pattern recognition, detection, forecasting, and decision making. Meanwhile, improved software and heuristic decision making have permitted much smaller dual I-Core microcomputers to perform as well as or better at chess than Deep Blue.
But, until recently, a computer's ability to truly "learn," that is, to write its own algorithms and rules according to data presented, has not been achieved -- and here comes the part you may not like....
You see, processing and working through mountains of data is no problem for computers these days. The problems arise when the rules (algorithms) become so complex for certain tasks that they just can't be written -- by a human. And, as it happens, identification of people and other objects by comparisons of characteristics or traits (for example, five pictures, same person, different pose) is an incredibly difficult set of algorithms to write.
Now, who might have access to those five mug shots of you, as well as your name "tagged" in each one? And who might have the same data on 50 million or more additional individuals? In a word -- Facebook. We and our Facebook friends have created a huge data garden from which to enable algorithm development via AI.
Just wondering. How does that make you feel?
We at E.T.I. are looking forward to the day when AI in a supply chain gives us five more days notice on a rush order for a difficult part, or, when our AI can talk to your AI. Until then, from our point of view, caution is required on where AI is heading, who's taking it there, and whose rules they're following.
No comments:
Post a Comment