Thursday, December 20, 2018

A Year-End Wrap-up on Today’s Hottest Jobs-Related Topic: Artificial Intelligence


Yes, in amount of recent press, AI has gone past guaranteed income, driverless cars, minimum wages, and even robotics.  It’s sort of an oxymoron, as it still seems to use only algorithms, which forces its developers to assume that what knowledge they want must be linear. 

We’ll start our set of dispatches from the last seven months of 2018 with one from a series called “Dispatches.”  This entry by Henry Kissinger, who, if he penned this himself, is one of the world’s best 95-year-old writers, is titled “How the Enlightenment Ends” and was published in the June 2018 Atlantic.  Kissinger was concerned that our “new, even sweeping technological revolution” has “consequences we have failed to fully reckon with” – a sentiment around since at least George Orwell in the 1940s, even if some of the problems he mentioned, such as “the ability to target micro-groups” causing politicians to be “overwhelmed by niche pressures,” are newer.  He warned of the dangers of “an AI program that is acting outside our framework of expectation,” especially one that learns much quicker than humans and can “surpass the explanatory powers of human language and reason,” and considered the nature of consciousness, which in one of many views comes from computations and is thus present in pocket calculators.  As opposed to the historical Enlightenment, in which the situation was the opposite, Kissinger wrote that we now have “a potentially dominating technology in search of a guiding philosophy.”  Well selected and well refined insights.

In the June 20th New York Times, Steve Lohr asked “Is There a Smarter Path to Artificial Intelligence?  Some Experts Hope So.”  The author first hinted at a rising problem, that “a growing number of A.I. experts are warning that the infatuation with deep learning may well breed myopia and overinvestment now – and disillusionment later,” a view which may be driven as much by concerns about the politically incorrect conclusions the deep learning, or massive data analysis, systems Lohr discussed sometimes reach.  Although they seem to “learn” by identifying commonality, such schemes are ultimately only based on computations, running faster but not making intuitive human-style connections, and indeed, per Lohr, “deep learning comes from the statistical side of A.I.”  That theme was also the point of Melanie Mitchell’s “Artificial Intelligence Hits the Barrier of Meaning,” on November 5th in the same publication, which said that “today’s A.I. systems sorely lack the essence of human intelligence:  understanding the situations we experience, being able to grasp their meaning,” and pointed out how vulnerable such schemes, which may need to learn from the beginning when only small things are changed, are to hackers.      

One section of artificial intelligence has hit a roadblock, as Julie Creswell reported on June 20th, also in the New York Times, that “Orlando Pulls the Plug on Its Amazon Facial Recognition Program.” That police-department effort, powered by Amazon’s two-year-old Rekognition product, was ended in the wake of protesters rightfully concerned that it could be used to track them “or others whom authorities see as suspicious, rather than being limited to individuals who are committing crimes.”  This won’t, though, be the end of large organizations vacuuming up millions of faces and attaching them to credit reports, medical files, purchasing histories, cell phone records, and so on, and it is unrealistic to think your face will not be with your name in plenty of huge databases within ten years.  There will be good and bad aspects to what is already the means to attach a personal history to a high share of facial photographs, making everyone an even faster and sharper data analyst than Penelope Garcia on CBS’s television show Criminal Minds.  Even if such technology is legally restricted, it will still be used.

Moving on to human resources issues, we have, also in the Times, “A.I. as Talent Scout:  Unorthodox Hires, and Maybe Lower Pay” (Noam Scheiber, December 6th).  That piece showed how companies can use a baseball-style Bill James or Moneyball system to identify winning job candidates lacking conventionally expected credentials, for whom, per a Columbia economist, the organization “doesn’t seem to have to compete… as much.”  A fundamental improvement over the long-time use of automated scanning for key words on resumes, the service Eightfold can look for related but not identical experiences.  One drawback Scheiber named was that people might feel “a sense of unfairness… if a computer were to make hiring, firing and promotion decisions” – that made me laugh, as those verdicts, when made by humans, have been notoriously erratic since long before the first computer was made.  Just as modern technology can put chubby or awkward-throwing but superbly hitting players into major-league careers, it can do the same for those with credentials slightly off balance in cubicle jobs.

We end on a pessimistic note from Timothy Egan, in the December 7th New York Times.  As before, we’ve known about threats from “The Deadly Soul of a New Machine” for a while, but it isn’t the whole story.  Yes, October’s Lion Air flight disaster may have been caused by an excessively dominant, in effect faulty, “advanced electronic brain,” but such things have saved many more lives than that, and saying that one driverless-car pedestrian death means that “there shouldn’t be any rush… to hand over the steering wheel to a driver without a heartbeat,” and that a good summary for artificially intelligent devices is “Our invention.  Our folly,” is out of touch with the reality of what such systems are doing now, let alone how they will perform when bugs that caused both disasters are removed.  We don’t need naivete, but we can also do without looking only at the worst.  Artificial intelligence is still only algorithmic, and will continue to present problems, but it is still overwhelmingly positive.  As for its future, stay tuned. 

No comments:

Post a Comment