Bank of America (BAC) held the first half of its A Transforming World: Robotics and AI Conference in London late last month, and the discussion was as scary as it sounds. In stark terms, respected analysts and economists talked about the devastating implications for jobs stemming from these new technologies.
When most people think about the robot apocalypse, they do so through the lens of Hollywood films. We are all familiar with the gray dystopian future where killer humanoid robots march through killing fields of human skulls. But the true apocalypse will be more about erasing humans from payrolls than extermination. The reason: Machines are getting smarter. Advanced algorithms are teaching machines how to see and interact with our world.
When AlphaGo, the deep-learning AI software from Alphabet (GOOG), soundly defeated the best Go players in the world last month, it did so by understanding all of the possible moves and likely countermoves in advance. It knew this because it had already seen those moves in millions of simulations. When an Autopilot equipped Tesla (TSLA) Model S accelerates and merges into traffic autonomously, it does so by seeing lane markers and the speed of nearby vehicles, and then acts accordingly.
These are important developments because previously the machines competing with humans for jobs were dumb. Not only are the new machines intelligent, they’re getting exponentially smarter. Thatbrings an entire new set of job categories into play like anything that requires sitting at a computer screen and assessing information. If that sounds like the basic job description of the average middle-class worker you can see the cause for concern.
The bad news is that these changes are not far off in the future. They’re here right now. For much of the past 100 years, American wage growth had risen in lockstep with productivity but in 1990 the two began to diverge. There is a good reason for this. Work can be divided into four types: routine, non-routine, manual and cognitive. Routine is different than non-routine because it does not vary. Manual is different than cognitive because it involves physicality.
In 1990, the St. Louis Federal Reserve Bank found that the growth of routine manual work, like that performed by factory workers, began to slow because it was relatively easy for software engineers to write rules robotscould follow. More recently it has become easier to write rules for routine cognitive work, like that performed by millions of Americans working in offices.
We can see these changes in real time. When Amazon.com (AMZN) purchased the robotics firm Kiva in 2012, few observers expected the dramatic impact automation would have on its warehouse floors. The online retailer now has more than 30,000 robots roaming about, guided only by computer algorithms . Human workers have been cut by two-thirds, performing only non-routine tasks. And then there is WhatsApp, the software messaging platform Facebook (FB) purchased for $19 billion in 2014. That firm serves more than a billion users and sends 34 billion messages per day with just 55 employees and 32 software engineers.
The World Economic Forum estimated that as many as 5 million jobs could be lost to automation by 2020. That estimate assumes little progress in deep learning and other forms of artificial intelligence. Andrew Ng, a co-founder at Google’s Brain and now chief scientist at Baidu (BIDU), begs to differ. He warns artificial intelligence advances will “create massive labor displacement”. And he is not alone in this assessment. The question is what political leaders will do about it. For now, as investors, the best course is to add to Alphabet and Amazon.com on dips.