杏吧原创

WHEN machine learning algorithms that replace newspaper reporters became fodder for a recent episode of Comedy Central鈥檚 The Daily Show, it was clear that the technology had gone mainstream.

But as Pedro Domingos points out in The Master Algorithm, machines that learn have been deeply involved with our lives for a while. If you use Google, Netflix, Amazon, Pandora, Yelp, Xbox or just about any online dating service, your life is being run by algorithms that are learning more and more about you by chomping on the data you, sometimes unwittingly, provide.

鈥淪ociety is changing, one learning algorithm at a time. Machine learning is remaking science, technology, business, politics and war,鈥� writes Domingos, a computer scientist at the University of Washington, Seattle.

For people in his field, the problem is that there are myriad such algorithms, each trying to discern patterns in the masses of data we produce. 鈥淢achine learning is about prediction,鈥� he writes, 鈥減redicting what we want, the results of our actions, how to achieve our goals, how the world will change.鈥�

The book is about the quest for that one master algorithm which would change machine learning, and hence our lives, irrevocably. If it exists, says Domingos, the master algorithm can derive all knowledge in the world 鈥減ast, present, and future 鈥� from data鈥�. In theory, such an algorithm could derive Newton鈥檚 laws from the astronomical observations of Tycho Brahe, with no a priori knowledge of such laws.

But why should such an algorithm even exist? Domingos provides compelling arguments from neuroscience, evolution, physics, statistics and computer science. For instance, the cerebral cortex might be an instance of such an algorithm: some neuroscientists think that it implements the same algorithm all over, just tweaked to learn to see or hear, or to make sense of touch.

Depending on your world view, the development of a master algorithm is either really thrilling or downright scary. It鈥檚 not surprising that Domingos, an expert in machine learning, has a very optimistic view. He clearly sees the master algorithm as desirable and maybe even inevitable. This cheery outlook shines through large parts of the book, when he writes that such an algorithm will 鈥渟peed poverty鈥檚 decline鈥�, that routine jobs 鈥渨ill be automated and replaced by more interesting ones鈥�, that the health of our planet will 鈥渢ake a turn for the better鈥�, and that our own lives will be 鈥渓onger, happier and more productive鈥�.

鈥淒epending on your world view, a master algorithm is either really thrilling or downright scary鈥�

Domingos has few doubts, and those he has mainly concern whether the technology will really happen as promised. 鈥淢aybe,鈥� he muses, 鈥渢he master algorithm will take its place among the great chimeras, alongside the philosopher鈥檚 stone and the perpetual motion machine.鈥�

But what about the future that lies in store for us, should machine learning take over our lives (if it hasn鈥檛 already)? Again, Domingos sees it all as a positive. 鈥淪omeday there鈥檒l be a robot in every house, doing the dishes, making the beds, even looking after the children while the parents work. How soon depends on how hard finding the Master Algorithm turns out to be.鈥�

The implications of machine learning for war and politics may be the most far-reaching of the transformations in store for our world. One reason that Barack Obama defeated Mitt Romney in the 2012 US presidential election is that his campaign embraced machine learning. In the future, 鈥渆lected officials will be able to ask voters what they want a thousand times a day and act accordingly鈥�, writes Domingos. This might work if elected officials always had their constituents鈥� good in mind. But could they not subvert the technology to manipulate the electorate?

As for war, the scenario is scarier. One can envisage fighting robots that use machine learning to get better and better at killing. Eventually, people will be off the battlefield and robots will fight it out among themselves 鈥� and this will prevent human casualties and hence suffering, says Domingos. But surely, all that would do is shift the suffering to other realms of life, not eliminate it?

It鈥檚 hard to avoid the feeling that machine learning is only going to increase the rift between the haves and the have-nots, as we enter a new phase of survival of the fittest. As Domingos writes, 鈥淗e who learns fastest wins鈥�, and machine learning 鈥渋s the latest chapter in the arms race of life on Earth鈥�.

But he鈥檚 still not worried. As machine learning does away with most jobs, the world Domingos envisions consists of a large class of unemployed people living on a permanent basic income doled out by the government, while those in the few remaining human occupations will be stupendously wealthy. 鈥淔or those of us not working, life will not be meaningless, any more than life on a tropical island where nature鈥檚 bounty meets all needs is meaningless.鈥�

Despite my reservations about the value of lives ruled by algorithms, I found the book oddly compelling. Domingos writes with verve and passion, and the book has a strong narrative. This can stall because of lengthy, fairly technical descriptions of the various learning algorithms, such as neural or Bayesian. Reading about their innards can be hard work, even for a former software engineer like me.

But these interludes aside, the book manages to build up a sense of anticipation as we join Domingos and his ilk in their pursuit of the ultimate learning algorithm. He makes you want to know whether they will succeed.

Domingos also provides an insider鈥檚 view, and doesn鈥檛 hold back from dishing out delicious titbits on big names in the field. 鈥淚f the history of machine learning were a Hollywood movie,鈥� he says, 鈥渢he villain would be Marvin Minsky.鈥� That鈥檚 because Minsky, a really stellar name, was deeply sceptical about machine learning, Domingos explains.

The Master Algorithm is a very thorough account of its subject, but I kept thinking that there is another book hidden inside: one that eschews much of the technical stuff and tackles the extraordinary consequences in more depth. That book would build a broader picture for lay readers, to prepare them for what lies ahead. If Domingos鈥檚 work provokes someone to write such a book, then it will have done us all a great service.

(Image: Martin Parr/Magnum Photos)

This article appeared in print under the headline 鈥淎 world remade鈥�