| dc.description.abstract | Machine learning has come of age. And just in case you might think this is a mere
platitude, let me clarify.
The dream that machines would one day be able to learn is as old as computers
themselves, perhaps older still. For a long time, however, it remained just that: a
dream. True, Rosenblatt’s perceptron did trigger a wave of activity, but in retrospect,
the excitement has to be deemed short-lived. As for the attempts that followed, these
fared even worse; barely noticed, often ignored, they never made a breakthrough—
no software companies, no major follow-up research, and not much support from
funding agencies. Machine learning remained an underdog, condemned to live in
the shadow of more successful disciplines. The grand ambition lay dormant.
And then it all changed.
A group of visionaries pointed out a weak spot in the knowledge-based systems
that were all the rage in the 1970s’ artificial intelligence: where was the “knowledge”
to come from? The prevailing wisdom of the day insisted that it should
take the form of if-then rules put together by the joint effort of engineers and field
experts. Practical experience, though, was unconvincing. Experts found it difficult
to communicate what they knew to engineers. Engineers, in turn, were at a loss as
to what questions to ask and what to make of the answers. A few widely publicized
success stories notwithstanding, most attempts to create a knowledge base of, say,
tens of thousands of such rules proved frustrating.
The proposition made by the visionaries was both simple and audacious. If it is
so hard to tell a machine exactly how to go about a certain problem, why not provide
the instruction indirectly, conveying the necessary skills by way of examples from
which the computer will—yes—learn!
Of course, this only makes sense if we can rely on the existence of algorithms to
do the learning. This was the main difficulty. As it turned out, neither Rosenblatt’s
perceptron nor the techniques developed after it were very useful. But the absence
of the requisite machine-learning techniques was not an obstacle; rather, it was a
challenge that inspired quite a few brilliant minds. The idea of endowing computers
with learning skills opened new horizons and created a large amount of excitement.
The world was beginning to take notice. | en_US |
