Robotic Evolution, Accelerating Automation, and Job Loss
Researchers have recently figured out how to impart hive learning capabilities to robots. It’s particularly applicable to the kinds of complex tasks that require observation, trial, error, learning and further trials. This will enable them to collectively learn, in weeks, tasks that would have individually taken the robots years.
Essentially, a group of identical robots are presented with the same learning challenge. They make various individual attempts to understand and master the task at hand. They compare notes, metaphorically speaking, learning from each others’ mistakes so there is no need for repetition. View here.
It's limited to motor skills at the moment, but I see no reason in principle why similar collaboration couldn’t be extended to mental tasks as well. (Robots are simply boxes with sensors and actuators, that afford AIs means to take actions in the physical world.)
I’ve been criticized in some quarters for an overly aggressive view of how fast accelerating automation can rise to displace human workers. Even today, in the face of AI creativity across many fronts–from invention to musical composition to investment management to scientific discovery–pundits persist in maintaining that there are major domains of creativity and work that will remain uniquely human.
They may be right, but the trends in AI and robotics certainly don’t support their confidence. Further, with such powerful evidence happening within a technological tidal wave of exponential progress, this confidence is not only unjustified but dangerous.
Even though Ray Kurzweil’s work has recently transformed belief in the exponential acceleration of technology from heretical to orthodox, most of us nevertheless persist in projecting in a non-exponential way. Stop for a moment, and consider: the vegetation in a pond is doubling every day. The day before it fills the pond, the pond is half full of vegetation. Now comes the shocking part: all of the previous growth is matched by the last doubling, even if it had been growing for years. And that’s true of every previous doubling as well!
This isn’t hypothetical. Computer power is driving most of the change on the planet. It’s been doubling roughly every two years, and the process has actually been speeding up. Now it’s closer to 18 months. It’s what’s driving automation, and the suddenly much-enhanced capabilities of AIs and robots.
If massive numbers of jobs and indeed whole professions start to rapidly disappear in the 2020s, leaving multitudes of workers high and dry, I am certain that pundits who today deny the threat will not be the ones dealing with those displaced people, who may form angry mobs.
The disconnect between the thinking about the technological aspects of accelerating automation, which tends to be deep and insightful, and the thinking about the social aspects of accelerating automation, which tends to be nonexistent or superficial, continues to greatly concern me.
I agree with the Techno-utopians about the potential of accelerating automation and other technological advances to make society far better. Where we part ways is their nearly universal presumption that such advances are inevitable, and that rationality will govern society in the face of such extreme change.
I can well understand uninformed people making such fundamental errors. I have a much harder time understanding such errors by the Techno-utopians, who tend to be among the best-educated and best-informed people in their countries.
In many such countries, including some of the most technologically and economically advanced, large portions of the electorate have recently supported positions and persons inimical to reasoned progress. Consider the rising popularity of nationalist and demagogic parties in democracies across the globe, all offering simplistic messages of hope, uncoupled from evidence, wrapped in jingoism and demonizing “the other.”
Even those techno-utopians who believe that accelerating automation will generate more jobs than it destroys should appreciate the fact that such jobs tend to be highly skilled, often requiring both training and numeracy beyond the abilities of most workers.
Do they expect the displaced to go back to school, and learn newly necessary skills? Very well; that could happen, for some—though by no means all. And what if those newly skilled jobs start to become automated as well? How many times do they expect the displaced to retrain?
Or, do they expect the displaced to gracefully become homeless, camping out on street corners and begging?
More likely, they expect the displaced to become recipients of a universal basic income (UBI), as is being tested by Y Combinator in a new experiment. While I applaud that experiment, and indeed consider it one of the few conventional experiments capable of fostering wider adoption of such programs, the challenges remain far more daunting than most yet realize. (I explore those challenges here. OTOH, a paper introducing a new approach that appears to address all of those challenges is coming in early 2018.)
I truly don’t understand how Techno-utopians, in particular, can dismiss these warning signs. All of this has happened in recent years, and we have only experienced the first waters of technological unemployment lapping at the shores. The tsunami is yet to come.