Ten Months on Robots, Automation, and Artificial Intelligence

Big topics, little communication.  My cupboard is not quite bare, but has some odds and ends, which are worthwhile, even if they add up to one post instead of the three these matters seem well worth.

Artificial intelligence, or AI, seems to be springing leaks, if not in how it is progressing but how people deal with it.  A stern view on one, by George Maliha et al. in Harvard Business Review on July 13^th, was “To Spur Growth in AI, We Need a New Approach to Legal Liability.”  We hit the issue of which humans are legally responsible for post-algorithmic technology with driverless cars, which haven’t spread enough for anything resembling legal precedents, and here we have the straightforward assertion that “the existing liability system in the United States and other countries can’t handle the risks” it entails.  The authors recommend “revising standards of care” especially for medical AI applications, granting radiologists, for example, immunity from malpractice if they provide secondary image reading after AI provides the first; “changing who pays:  insurance and indemnity” including insurers giving better rates for professionals using favored AI systems;  “revamping the rules:  changing liability defaults” such as, if autonomous cars are involved, not automatically blaming a human driver in the striking vehicle for a rear-end collision;  “creating new adjudicators:  special courts and liability systems” using more sophisticated knowledge than most judges have;  and “ending liability completely:  total regulatory schemes” or institutionalizing knowledge that in some cases nobody is at fault.  A good start, all of this.

Julian Jacobs addressed another future problem area in the Brookings TechTank on November 22^nd, with “Automation and the radicalization of America.”  Jacobs found that combining one study assigning mechanization potential to occupations with another giving demographic data on people working them told us that those more likely to be replaced by machines “tend to have a dark and cynical view of politics, the economy, the media, and humanity” and skew left on financial issues but slightly right on “socio-cultural ones.”  He stopped short of predicting revolutionary activity among these workers, but such, if they do indeed lose their jobs, could happen this decade or next.

“Will Robots Really Destroy the Future of Work?”  Peter Coy revisited this old overstatement in the January 24^th New York Times, featuring an interview with labor economist David Autor, who “loves” both robots and unions and wants the two to be coordinated better, ideally on better-paying jobs.  Per Coy, such means realizing that “workers need training so they can use automation, not be replaced by it.”  I see no mention of the number of positions that we can expect to be lost, and it seems naïve to think it will not be substantial, even if some are created – the major point of mechanization is to reduce labor costs, which would not happen if almost as many jobs are created to work with it.  Now, as opposed to two years ago, we can better justify trading lower-paying positions for fewer higher-compensated ones, but there is hardly a guarantee 3%+-unemployment will last indefinitely.  Destroy, no – damage and change, yes.

The Writing on the Wall” was a long April 17^th Steven Johnson piece in the magazine section of The New York Times.  The subtitle of sorts was “A.I. has begun to master language, with profound implications for technology and society.  But can we trust what it says?”  We’re now at the point where such a system can write good-looking essays proposing plausible solutions to complicated problems in a second or so, through abilities to determine missing words and access massive numbers of sites, not all truthful or prudent.  The core of this issue is that the machines themselves cannot judge written material and cannot always identify lies, meaning human input is still needed.  We also are not avoiding the issue of what AI language modules may produce without confidential influences, which could well offend or even upend modern sensibility.  Overall, Johnson’s view that “the very premise that we are having a serious debate over how to instill moral and civic values in our software should make it clear that we have crossed an important threshold” seems appropriate – and solutions may depend on specific assumptions such as “people are basically good” and “guns in houses are safe enough,” which could be revealed to all.  A long way to go we have, and this piece does help.

Shrinking to a less general concern, we have Tanya Moore’s April 19^th New York Times “Can A.I. All but End Car Crashes?  The Potential Is There.”  We don’t have many autonomous vehicles, but there are plenty of others with related software – even my ordinary, year-old Toyota Camry beeps when I cross a center line.  Moore mentioned various other mechanistic improvements, and others in progress – this area is burgeoning.  That means that even if we don’t lose drivers, we will still gain a lot of safety and save many lives.

I end with a robot application with smaller import, but the kind which we can solidly expect.  It’s “Jack in the Box to pilot Miso Robotics’ Flippy 2, Sippy” (Lucas Manfredi, Fox Business, April 26^th).  It will start in only one of the fast-food chain’s locations, and not until late this year, but the first of these “takes over the work for an entire fry station” at a 30% production increase, and the second cuts drink spills as it “efficiently moves cups,” “accommodates a range of cup sizes and groups cups by order for easy delivery to customers.”  At today’s rates, Flippy 2’s $3,000 per month is less than only one full-time fast-food worker, and will work many more hours.  Like it or not, if the trial works, it will propagate, help the business, and potentially save customers’ time and money.  Look for many more – and don’t forget these growing and evolving issues, as, headlines or not, they won’t leave us alone forever.