Friday, 18 March 2016
Smart Machines Trump Trump
Last week I wrote about an algorithm that beat the world`s leading human player of a game called Go. The algorithm learns from experience. It mimics how natural neural networks function, and it is augmented by an innovative twist known as deep learning which drives greater speed and efficiency. Neural network computing methods are the brainchildren of many, but deep learning is the purview of Geoffrey Hinton, formerly of the UK, now at University of Toronto. In 2013, Google bought a company that Hinton jointly owned with grad students which held the rights to a specific form of deep learning. Now Hinton splits his time between the University and Alphabet/Google's main California campus, just one more instance of leading edge science being dragged into the widening maw of what we once considered narrow commerce. Now commerce leads the way. [For more on the Hinton story see my last book, SMARTS.] Clever technological application of deep learning is advancing artificial intelligence research by leaps and bounds, though the idea of copying how neurons remember and learn (through increasing connectivity) dates back a very long way. Turing discussed the idea at a symposium in 1949. Attempts to design computers that mimic how humans learn went in and out of fashion over the next fifty years. A leading American computer scientist, who helped direct the assembly of the human genome project, told me he began working on neural networks in the early 1980s. Hinton himself studied neural networks with a man who was convinced they were a waste of time.
Success with neural networks, in other words, has been achieved only after long years of failure, which is itself a reflection of the way humans learn.
Go requires inference, intuition, and vision, mental skills usually considered to be the sole province of smart humans, not algorithms. Yet the algorithm beat Sedol in four out of five tries. In a brief apres le deluge interview posted inside this Guardian story, Sedol said that before the first game, he was pretty confident that no machine could beat him. He was surprised to find himself losing. He only won game four because the algorithm made a mistake. This gave Sedol hope that he could also defeat the algorithm in game five. Although it made another mistake, the algorithm learned from it, corrected itself, and defeated Sedol in a way, he said, that no human being would have been able to do. After the games were over, the algorithm was awarded a 9th Dan, the highest level of attainment in competitive Go.
Which brings me to something almost as important as this artificial intelligence milestone -- the amazing algorithm known as the democratic process.
Even as Sedol was losing game five, Donald Trump was trouncing his competitors in Tuesday's Republican primaries, with the exception of Kasich who beat him in Ohio. This was an utterly dispiriting result for those in the upper reaches of the Republican Party who had hoped that if Rubio took Florida, and Kasich took Ohio, Trump would not get enough delegates to win on the first ballot at the convention. A second ballot, with the delegates free to change their votes, might permit the leadership to get a nominee more Republican in his opinions, more amenable to the interests of the Party's big backers. But now Trump holds the Party's future in his hands-- unless the Republican Party votes to free all delegates in the first round of voting. Trump has said that if anything like that happens, there will be riots. He is not the only one predicting them.
Much has been said about how Trump is lighting the bonfires of hatred in America as he chest thumps about shutting America's doors to all Muslims, deporting 11 million illegal migrants, getting Mexico to pay for a giant wall across the southern border. His dystopian vision appeals to fringe voters who have never been happy living in a neighborly way with anyone whose skin color is not lily white. But hatred alone is not why Trump is winning. His intuition about what ails America has taken him far beyond the simple calculus that there might be enough mean spirited folk in the US to lift him to office. It's what he has to say about the US economy that has attracted all kinds of other voters, not just that loony fringe. He's channeling the anger and despair of what entertainment executives used to call the fly-over zone, those places in the US that are not Los Angeles, New York, or Washington. Bernie Sanders has been doing the same in the Democratic primaries, although he has voiced the anger of the student-loan burdened young more particularly, and he doesn't drape his argument in xenophobic rhetoric.
These two are not the first "outsiders" to run for high office as voices of the voiceless. Remember Ross Perot? What is so fascinating is how, at certain turning points in the natural histories of democratic societies, communities push forward such champions, or perhaps I should say welcome them with open arms as soon as they say out loud what those "in charge" would prefer to suppress.The interaction between champion and community, the way they feed off each other, learn from each other, and find new directions together is a hugely complex group version of what our own individual neural networks do when confronting the world. This mega neural network calculates who will win key elections without anyone involved really understanding how this works or being able to predict outcomes.
Americans are angry, says Trump, because professional American politicians-- unlike him--have been bought and paid for by special interests who don't care a fig for the interests of voters. He speaks often, and loudly, about how he intends to make American corporations which have relocated their plants to low wage zones like Mexico, Viet Nam, Malaysia, and China, pay serious taxes to import finished products back to the US. In effect, he is saying he will make big business sorry for destroying the livelihoods of a generation of American workers in order to reduce costs and increase profits. Trump also wants to force major American corporations to repatriate their ill gotten profits from the tax havens where they've stashed them, places like the Cayman Islands, Bermuda, etc. He has mentioned Apple in particular in that context which I'm sure went down like a rubber donut in Silicon Valley.
His rant is the anti-Republican Rant, the exact opposite to the normal one shaped by the Koch brothers through their billions in donations to Republican PACs and think tanks. The normal Republican Rant promotes the notion that government is bad and taxes are worse and corporations should reign supreme.
Trump argues that it's the free trade deals that have enabled the devastation of the American middle and working classes and the consequent rise of the 1 percent. He says he wants to rip up trade deals that benefit big corporations and enemy countries. In effect, he is re-litigating a battle that was already old twenty years ago. Remember Perot's argument against free trade? Remember his warnings about "the giant sucking sound" that would echo across the land as American jobs disappeared to Mexico thanks to NAFTA?
Unfortunately, tearing up trade deals will not fix the problem that afflicts America now, and by extension all the rest of us in the western world. The problem is the onrushing technological future, not just the globalization of the economic system. No matter who becomes the next American President, as the Google/DeepMind algorithm has demonstrated, jobs will not return to the fly over zone. The auto industry will not get rid of its robots and rehire all those folk it let go over the last twenty years. Software will continue to replace lawyers and accountants and hedge fund executives, as well as marketers, reporters, editors and home care personnel. The giant sucking sound is already being heard in the tall towers of New York and Los Angeles, and in the back rooms in Washington. Humans will become ever more superfluous to the functional performance of our globalized economy. Even the jobs done by the 1 percent will eventually be done by intelligent machines, machines imbued with the capacity for inference, intuition, and vision thanks to new algorithms like the one displayed in Seoul.
Trump is standing eyeball to eyeball with the past instead of squaring up to the future.
He doesn't understand that the machines have already trumped him.
But the democracy algorithm gets it.