Saturday, March 19, 2016

Robot Sunrise


A friend of mine is a capable physicist working for a tech company. He dabbles in Artificial Intelligence for fun with a particular interest in “deep learning.”  Deep learning machines are able to rewrite their own programs and devise their own rules of thumb in response to real world experience; they can be as basic as the “recommended for you” AI of Amazon, as specialized as diagnostic medical devices, or as generalized as IBM’s Watson, which famously won at Jeopardy in 2011 against reigning human champs despite having its response time deliberately slowed down in order to give the humans some hope of reaching the buttons. (“Devices” can be a misleading word, since the computing sometimes is done in the cloud rather than on discrete devices.) Watson’s achievement is more surprising than it first appears when one considers that it had to interpret human idioms and puns; the most recent version of Watson is twice as fast as the Jeopardy model and operates on the cloud.

I sometimes joke with my friend about how he is coming along with Skynet, but to some people this is no joke – and some surprising names (including Stephen Hawking, Moshe Vardi, and Elon Musk) are among them. Married to robots, AI according to Hawking could mean “the end of the human race”; Musk similarly calls robots an “existential threat.” The primary risk is not the prospect of Terminator-style lethal autonomous weapons systems, though (non-humanoid) robots are in fact a growing part of the battlefield. Most of the concerned analysts, including Vardi and Martin Ford (author The Rise of the Robots), see the indirect threat of economic disruption as more worrisome than direct threats to life. Robots, they say, will continue to eliminate ever more semiskilled jobs as they have done for decades, whether warehouse workers at Amazon or fast food chefs in burger joints. Self-driving vehicles eventually will eliminate jobs of taxi and truck drivers. What is more, AI is able to replace white collar workers – and not far down the road but right now. How about journalism? Ford gives this example:

“Guerrero has been good at the plate all season, especially in day games. During day games Guerrero has a .794 OPS [on-base plus slugging]. He has hit five home runs and driven in 13 runners in 26 games in day games.”

There is nothing very remarkable about that sports item other than that it was written by a computer with no other instructions than to write about the day’s baseball games. Numerous news organizations (Forbes, for one) use similar technology to produce business and news articles, some of which need a little touch-up and some of which don’t. According to the consultancy company McKinsey, 45% of the work people currently are paid to do could be automated economically including 80% of a file clerk’s job and 20% of a CEO’s. As everything from retail to education continues to move online AI can take over more of the tasks. Tech companies – the biggest business success stories in the past two decades – already employ very few workers for their valuation. Youtube was started in 2005 by three people and employed only 65 when Google bought it out for $1.65 billion, or $25 million per employee. Facebook acquired Instagram (13 employees) for $1 billion in 2012, which is $77 million per worker. This contrasts with the old metal bending industrial giants that employed thousands of workers. In consequence, some of the doomsayers project unemployment rates of 50% by midcentury.

But can a machine intelligence be truly creative? Can it compose music, for example? Yes. The London Symphony Orchestra in 2012 performed the neoclassical Transits – Into an Abyss, composed by an AI algorithm called Iamus [who communicated with birds in Greek mythology] running on several computers. Critics liked it, one calling it “artistic and delightful.”

I have no doubt that AI robotics will be disruptive both at work and in private life. I’m less convinced that the result need be mass unemployment, much less extinction of the species. Automation always has led to net economic gains in the past even though it was hard on the individual workers affected. The argument “this time is different” is rarely accurate; relying on it is what prompts investors to buy into asset bubbles and then miss the post-crash rebound. We will adjust, even if most of us eventually end up servicing robots for a living. Nonetheless, the road ahead might be a bit bumpy; fortunately, robot drivers are pretty good at navigating bumpy roads.


Dinner at Eight (1933) – Marie Dressler might be wrong

4 comments:

  1. I'd wondered what Hawking was meaning about the fear of AI, but mass unemployment makes sense. Where I worked they had just started using remote control engines, which displaced people and jobs. I don't think of that, however, as AI or robots, but more automation which is just as bad. Though it seems like data could be fed into some sort of artificial thinking too that could benefit mankind, how to perhaps solve the unemployment problems, ways to solve some of our economic dilemmas, etc. I'd think there's always things that could be made better given that mankind could somehow agree on the matter.

    ReplyDelete
    Replies
    1. There also seems to be a general unease that our AI creations are bound to replace us, just as our biological children do, even if we can't foresee all the paths by which this can happen. Perhaps they will, but I wouldn't bet the farm on it.

      Delete
  2. But if AI takes over all our jobs, think of all the free time we'll have. More time to watch You Tube videos and play Candy Crush. That can't be all bad right?

    Well, now that I think of it, Candy Crush is just a hop, skip and a jump to the Matrix, right. And with those VR headsets coming out at the end of the year maybe we are closer to "The Matrix" than we thought.

    Has anyone mentioned using Asimov's rules of robotics in the programming of AI. Seems like that might solve many of the problems we'd have with them right there... unless we go all "Caves of Steel" on them.

    ReplyDelete
    Replies
    1. One of the big mistakes John Maynard Keynes made when predicting the 21st century economy was the assumption that people would continue to trade income for leisure time as compensation rose. This was true only through the 1950s. We work more hours now than then. Perhaps robot labor can indeed make Keynes right after all, albeit a few decades late.

      Asimov gets short shrift it seems. Military robots violate the three laws by design.

      Delete