The Gorilla-in-the-Room: Income Stratification

Original article can be found here (source): Artificial Intelligence on Medium

The Gorilla-in-the-Room: Income Stratification

By Brady Engelke | 3/15/20

If the 2020s are anything like the past decade, the world will be a vastly different place by 2030. There are powerful forces at play that make it an exciting time to be a software engineer but quite an unnerving time for chess grandmasters and truck drivers. Alphabet’s Waymo currently has autonomous vehicles (AVs) navigating US streets and Alphabet’s DeepMind developed an algorithm in 2017 that trained itself to play chess within four hours only to dominate Stockfish8 — the worlds computer chess champion of 2016. As an aspiring data scientist, these feats invigorate me to sharpen my skillset and immerse myself in the expansive literature of the field. The potential these technologies have to change society is unprecedented; to be a part of this global transformation is something I am not willing to miss out on.

The possibility of AVs ruling the roads is not only fascinating but promises to mitigate the rate at which car accidents occur. In his latest book, 21 Lessons for the 21st Century, Yuval Harari claims that “replacing all human drivers by computers is expected to reduce deaths and injuries on the road by about 90 percent [1].” Harari attributes this profound impact to the enhanced efficacy a connected and updatable network of AVs would have over individuals disparately driving by their intuition. Every AV would be synchronized with the other AVs around it, substantially reducing traffic density as well as accident rates. Every time an AV encounters a new scenario on the road, resulting in a crash or swerve, the underlying algorithm could immediately learn from that event and update all AVs across the network on how to proactively resolve similar future situations. Harari also denotes the potential benefits of a connected network of artificial intelligence (AI) agents in the realm of Healthcare. As of now, it is near impossible to ensure that doctors across the world are acting upon the most up-to-date information. This is something that would be far easier if a connected network of AI doctors could be streamed updates on the latest research and techniques. This scenario may be a bit far-fetched as of 2020, nonetheless, AI is still modestly making its mark. The American Cancer Society recently published a case study with Slalom that applied machine learning to identify abnormalities within 1,700 tissue slides, a task that would have taken a seasoned pathologist an unduly amount of time [2].

Harari is not the only one to acknowledge the potential benefits of AI. Andrew Yang, a former candidate of the 2020 presidential election, understands the potential economic value AI boasts as well. He leveraged a projection by Morgan Stanley that autonomous freight alone will save businesses $168 billion a year to fortify a campaign initiative around smoothing the transition for the 3 million truck drivers across the US who will be displaced by autonomous freight [3]. Harari and Yang imagine it being a far more challenging effort for displaced workers to transition to other professions in 2020 compared to the 1980s. In 21 Lessons for the 21st Century, Harari points out that in 2015, despite the Air Force facing a shortage of skilled candidates to man their drones, the Air Force was still unwilling to accept applications from supermarket cashiers displaced by self-checkouts. Yang is aware of this problem as well, calling out the unimpressive results retraining efforts for displaced manufacturing workers have had in the past [4]. These revelations led both men to conjecture that Universal Basic Income (UBI) could be a potential remedy, Harari diving into the details of what “Universal” and “Basic” could potentially mean, and Yang trying to enlighten America on why giving $1,000 a month to families across America is prudent.

After following Yang’s presidential campaign, it is evident that UBI could benefit the American people although I remain cynical as to the feasibility after reading about the lengths at which the uber-wealthy are willing to go to evade taxes in The Laundromat [5] and how disingenuous big banks are willing to be to enhance their profits in Rocket Boys [6]. Harari brings up an even more daunting point: even if wealthy Americans do choose to help the middle-class, it is irrational to believe that Americans will accept the burden of supporting countries that depend upon the US for a majority of their income once it is economically viable to automate their services [1]. Let’s say Harari is wrong, and Americans decide to assist these countries, the definition of “Universal” and “Basic” will be critical in the coming decades and I can only hope policymakers emphasize the well-being of humanity over GDP.

If a New York Times bestselling author and a tech entrepreneur audaciously running for president are not enough to convince people that automation should be a primary concern of politicians, I am not sure if there is really anything of value I can add to the party. But I decided to take a shot anyway and see if I could better understand the extent to which income stratification exists in the US today. Before we dive into my study, I would like to mention that Bain & Company already wrote a rigorous 68-page report in 2017 on this topic [7]. Below are a few key excerpts from the report:

  • Analytical, administrative or clerical service sector jobs with highly repetitive or rule-based tasks are surprisingly easy to automate. Law firms already are deploying algorithms to scan legal documents in lieu of highly educated and highly paid junior lawyers. Hedge funds have been developing machine learning systems that perform the work of thousands of analysts at inhumanly fast speeds.
  • Our base-case scenario, in which automation displaces 20% to 25% of US workers by 2030, will hit the lowest end the hardest. Workers currently making between $30,000 and $60,000 per year are likely to experience the greatest disruption from automation: up to 30% could be displaced, and many will suffer depressed wage growth. We expect automation to have a lesser impact on those with incomes between $60,000 and $120,000 a year and the least negative impact on those earning more than $120,000.
  • By 2030, employers will need 20% to 25% fewer workers, equivalent to 30 million to 40 million jobs in the US. To put these numbers in context, during the Great Recession, US employment fell rapidly from its peak in January 2008 to its trough in February 2010 by nearly 9 million jobs, or 6.3% of total employment.
  • By 2030, we estimate that operating costs at an industry level could decline by as much as 10% to 15%. Under those conditions, increased profitability would largely flow to owners of capital and further reduce the share of national income allocated to labor.
  • Rapid deployment of automation technologies is likely to exceed the pace at which economies can reabsorb and redeploy the millions of workers who may lose their jobs to automation.
  • Across geographies, the decline in manufacturing costs due to the next level of industrial automation is likely to negatively impact large exporters that compete based on lower labor costs.
  • Our analysis does not take into account additional third-order effects such as the introduction of new job categories. However, given the magnitude of disruption in our base-case scenario, we do not believe new job categories will temper the degree of labor force disruption in the 2020s.
  • We acknowledge a material risk that social and governmental resistance delay extensive adoption of automation technologies. It is impossible to handicap that risk as the response across different countries is unlikely to be uniform. However, we believe that our base-case scenario of rapid deployment is more likely.