Yesterday evening I had the pleasure of listening to Professor Moshe Vardi talk about the effects of automation on the workforce and the economy, and how the continued development of AI and machine learning-based technologies might further tip that balance. This post is based on the notes I took during that talk. Please read it as more of a high-level summary, rather than a transcript. The talk itself contained more links and references that I haven’t had the time to chase down, so any inaccuracies and misquotations are probably my own fault.
Professor Vardi is a professor at Rice University, and currently Editor-in-Chief of the Communications of the ACM, and the winner of numerous awards including the ACM Gödel Prize, and the EATCS Distinguished Achievements Award. He is also an excellent speaker and it was wonderful to see a coherent narrative formed out of many disparate threads.
The talk started with a brief mention of Turing’s thesis, which can be read as a compelling philosophical argument for thinking machines, and the related intellectual questions. The early history of artificial machines was characterized by unbridled optimism (expectations that general purpose AI would arrive within a generation), punctuated by several AI winters (1974-80 and 1987-93) where funding and support for AI research dried up. However, 1997 started a new era in AI research when a chess playing computer, IBM’s Deep Blue, defeated Garry Kasparov. More recently, DeepMind’s AlphaGo defeated European Go champion Fan Hui. Crucially, AlphaGo combines machine learning techniques (including deep learning) with search space reduction, resulting in an approach that could be termed “intuition”.
With the resurgence of AI research, automated driving has been the holy grail for about a decade. Cars were one of the most important developments of the 20th century. The automobile shaped geography and changed history, and led to lots of infrastructure development. By some estimates, there are over 4 million truck drivers in the US, and 15 million jobs involving operating a vehicle. Today there are about 30 companies, working on self-driving vehicles, attacking an estimated market of $2 to $5 trillion a year. Experts predict that the main technical issues will be resolved in 5-15 years. While this will be a great technological achievement, it will produce profound business disruption. For starters, there is likely to be major industrial contraction (cares are idle 90% of the time), and a major loss of business for insurance, legal, medical fields, as automobile accidents are drastically reduced.
Unfortunately, this industrial disruption follows a trend that has already been in progress for a while now. The last 40 years has resulted in a harsh negative impact on middle & working class. For much of the 20th century there was a “Great Coupling” between productivity, private employment, median income and GDP growth: they all followed a linked upward trend. However, since the 70s, this trend has “decoupled”, a fact observable from many dataset. In particular, there has been increasing inequality: a massive decline in the bottom 50% of earners, and a massive increase in the top 1% of earners. There is a declining chance that a person in their early 30s is going to be better off than their parents.
This in turned has resulted in an “Age of Precariousness”: half of Americans would have trouble affording $400 for an emergency, and two-thirds would have trouble dealing with a $1000 emergency. Labor force participation for men 25-54 has dropped from 97% to 88% and those with high school degrees or less were the hardest hit — almost 20% are not working.
Technology is eating jobs from the “inside out”. High-paying and low-paying jobs are both growing, but middle class jobs are declining. According to a Bloomberg 2016 report: as we move towards more automation, we need fewer people in manufacturing and more people go into the service sector, historically a low-wage sector.
All this paints a pretty bleak future, and from Prof. Vardi’s talk it’s unclear what the way forward is. Universal Basic Income seems like one idea to help offset this dangerous trend, but UBI is still a hotly contested topic. The following discussion raised some interesting questions, including asking what the role of “work” and employment is in a mostly-automated society, and questioning the role and responsibility of educational institutes in the near future.
Personally, I feel lucky to be in a field where jobs are currently booming. Most of my work is creative and non-routine, and thus not amenable to automation yet. At the same time, I am very concerned about a future where the majority of people hold poorly paid service sector jobs where they can barely eke out a living. I am also afraid that jobs that seem more secure today (administrators, doctors, lawyers, app developers) will also be gradually pushed into obsolescence as our machine learning techniques improve. Again, no good solution, but lots to think about, and hopefully work on in the near future. As the Chinese proverb goes, we live in interesting times.