Friday, October 23, 2015

One Algorithm to Rule Them All

As a student in a 1970s high school computer lab, I used a teletypewriter to punch holes in a paper tape, dialed a phone number and jammed the handset into two rubber cups. Somehow, an unseen computer on the other end was told to do something. I didn’t see the point.
     In the 1980s, as a newly minted account executive at a consulting firm, I assured a colleague I’d be dead long before any work I might do would require learning to use a computer. (I also was sure that I wouldn’t ever need to type — that’s what the secretaries outside my office did on electric typewriters, which corrected typos at the touch of key!)
     Thirty years on, the opportunities that I’ve missed in computer science — a set of disciplines that is arguably the most influential and powerful in the world — are obvious. To avoid further embarrassment, I took up Pedro Domingos’s new book, The Master Algorithm: How the Quest for Machine Learning Will Remake our World (Basic Books, 2015). 
     Unlike me, Domingos did not miss the import of computer science. While I was ignoring the computers popping up on every desktop, he was earning alicenciatura in electrical engineering and computer science from Instituto Superior Técnico at University of Lisbon. After reading about machine learning in a book on artificial intelligence, he skipped an MBA and, instead, earned a Ph.D. in information and computer science from the University of California at Irvine. Now a professor at the University of Washington, Domingos is a leading expert in the fields of machine learning and data mining. His Tolkienesque quest — and the book’s subject — is the most powerful algorithm of all.  
     This master algorithm, which many experts, whose views Domingos gives a hearing, think is a Computer Age chimera, would allow machines to learn without human assistance. “Every algorithm has an input and an output: the data goes into the computer, the algorithm does what it will with it, and out comes the result,” Domingos explains. “Machine learning turns this around: in goes the data and the desired result and out comes the algorithm that turns one into the other.”
     Learning algorithms are already commonplace. Netflix uses them to pick movies for us; Amazon to recommend books; Google to search out Web pages. But Domingos is pursuing something much more far-reaching. “In fact, the Master Algorithm is the last thing we’ll ever have to invent because, once we let it loose, it will go on to invent everything that can be invented,” he writes. “All we need to do is provide it with enough of the right kind of data, and it will discover the corresponding knowledge.”
     The corresponding knowledge includes a cure — or, more accurately, myriad cures — for cancer. In theory, Domingos argues, the master algorithm could create a program capable of spitting out the exact formula for a therapy designed to kill a specific patient’s cancer — based on a tumor’s genome, the patient’s medical history and profile, and a “vast database of molecular biology.” Unfortunately, little of the data necessary to fuel such a program exists as yet. But then neither does the master algorithm... read the rest here

No comments: