Folks,
I just read an article suggesting that each newly-deployed robot will eliminate 6 human's jobs.
The authors may be right, but I think they miss a point. In the early 1800s, when water-powered weaving machines started putting hand weavers out of work, the Luddites protested. It is certain that this technological advance was disruptive and did put hand weavers out of the work of hand weaving. However, the price of cloth became much more affordable and resulted in clothes becoming considerably more available to the middle and even lower classes. Admittedly the hand weavers had to enter new trades, including possibly designing or assembling, or installing, or maintaining, or selling water-powered weaving machines. In addition, as the industrial revolution proliferated, there were certainly more than a few downsides, however, eventually almost all benefited from industrialization.
I see modern robots in a similar light. The article referenced above goes on to state that the types of jobs replaced by the robots are boring and repetitive, such as welding in automobile manufacturing. In addition, these robots usually require more assistance from humans than many people think. I’m reminded of the electronics assembly “lights out factory concepts” of the 1990s that never occurred. Electronics assembly hasn't yet developed to the point of not needing humans, many humans.
Figure 1. Robot Designed for Electronics Assembly
The robots themselves will require maintenance, “assists”, programming, and to be designed and manufactured. I still believe the end result is more productivity and cheaper products for all, not to mention more fulfilling and challenging work for us humans. To remain engaged, the worker of today must update their skills or they may be left behind in a world of fewer non-skilled jobs. However, I don’t see the “robot apocalypse” envisioned by the Washington Post. The Post sees a tremendous loss of jobs due to automation.
Even some white collar jobs, such as stock picking, can now be handled by computer algorithms. Blackrock is betting on this trend. However, I believe that artificial intelligence (AI) will mostly work with humans, not replace them. My friend, Aaron Brown has pointed out that, while computers now routinely beat grand masters at chess, the most unbeatable team is a human chess expert and a computer.
But, just how smart is AI? Is it such that we will all be replaced by robots in a generation or so? Hardly.
Early AI pioneers had a confident view of future AIs. Witness a quote from a 1956 workshop at Dartmouth:
“The field was founded on the claim that human intelligence ‘can be so precisely described that a machine can be made to simulate it’”.
Wow! We now know enough to recognize that “precisely describing” human intelligence is not within our grasp today or even in the foreseeable future.
Even before this Dartmouth meeting, the 1950 Turing Test was considered the ultimate test of an AI. In most envisionments of the test, a human sits at a terminal and communicates with either another human or an AI. If the AI is perceived by the human as human it passes the test.
In recent years, researchers have realized that the original Turing test could be fooled by deception. Gary Markus’s excellent article in the March 2017 issue of Scientific American addresses this concern. Marcus argues that there are now four categories of tests that an AI would be expected to pass. I encourage those interested to read his commentary in full, however I will discuss two of the tests to give a sense of how close AI is to passing this modern suite of Turing tests.
The first test is the Winograd Schema Challenge. This test provides simple, but ambiguous, sentences that require real world commonsense to understand. Winograd himself poses an example:
“The city councilmen refused the demonstrators a permit because they feared violence.”
The test question is: “Who feared violence?”
Humans can reliably figure out that it is the councilmen by employing common sense and cultural norms. Marcus points out that the best AIs got less than 60% of these types of challenges correct in 2016.
The Physically Embodied Turing Test is another. This test could be as simple as building a wall out of blocks after having been giving instructions. Clearly this is a task a two year old child can do. Marcus points out that these types of problems are a considerable challenge for the current level of AIs.
I was reminded of this type of problem when I received a weight lifting rack to hold a barbell for doing squats. It came with no assembly instructions. I would not claim to be especially handy, but I was able to assemble the rack in about 10 minutes without instructions. Marcus’s article suggests that AI is decades away from this ability. Even with instructions an AI would not be able to assemble the rack with today’s technology.
Figure 2. This weightlifting squat rack I assembled with no instructions was a simple task. No robot with AI today could come close to accomplishing this task.
Writing this post was an eye opener for me. Even though I was a skeptic of AIs believed advanced state of development, the studies I performed suggest that many decades of work will transpire before an AI can be hoped to pass a suite of Turing tests. My sense is, when an AI does pass the tests, skeptics will quickly point out weaknesses in the tests and it will likely be more years before skeptics are convinced.
My conclusions on AIs, robots, and self-driving autos always end up with the same result: interesting progress is being made, but the uniqueness of humans is always underestimated. I see all of these technologies not significantly affecting jobs that humans do for quite some time. However, it will be important for all of us to constantly be evaluating our skill set and improving it. Recall that quite a few horse manure shovelers were displaced by the introduction of the automobile, and that numerous wood choppers had to retool with the advent of the automatic water heater. Few of us long for the return of either of those jobs.
Much as the industrial revolution created new companies and, with the help of labor laws, eventually raised the standard of living for most people. I think today’s advances will create new industries and, hence, new jobs. The current situation would appear to support this view, as most of the highest value companies, like Apple, Microsoft, Facebook, Google, etc. did not exist 50 years ago. New technology created these companies and the many jobs they provide. Is there any reason not to expect this trend to continue?
Cheers,
Dr. Ron