Artificial Intelligence and Society

I have been a fan and a subscriber to Daedalus, the Journal of the American Academy of Arts & Sciences, for many years. In 2021, the journal became an open-source journal, making its new issues available to all, and the academy is in the process of digitizing back issues to make them available. The most recent issue, Volume 151, Number 2, is titled “AI & Society” with an all-star group of contributors.

Fortunately, the 374-page issue is organized along the lines of topics. The topics are:

  • On beginnings & progress
  • On building blocks, systems & applications
  • On machine vision, robots & agents
  • On language & reasoning
  • On philosophical laboratories & mirrors
  • On inequality, justice & ethics
  • On the economy & future of work
  • On great power competition & national security
  • On the law & public trust
  • On seeing rooms & governance

Each of these topics has several articles. Most of the authors have an academic background as a researcher or pedigree with a PhD and a great deal of experience outside of academics. I found most of them to be thought-provoking but must admit that some captured my attention and imagination more than others.

The topic, On the economy and future of work, interested me the most. There were three articles and the one written by an economist fascinated me the most.

Automation, Augmentation, Value Creation & the Distribution of Income & Wealth was written by Michael Spence. Michael Spence is Senior Fellow at the Hoover Institution, Dean Emeritus & Professor Emeritus in the Graduate School of Business at Stanford University. He was awarded the Nobel Prize in Economics in 2001.

When an author begins an article by defining specific terms, I try to pay attention. In this case, Mr. Spence defines “use cases” as applications of digital technology such as mobile payments, social media, online commerce, and location-specific services such as maps. He further writes that automation and digital machine augmentation are a class of use cases. Automation replaces people with machines and augmentation adds machines that make people more productive at work. The two are flip sides of the same coin.

Mr. Spence writes that augmentation is how humans have used tools for years. While augmentation can increase unemployment in the short term, over the long term, incomes rise due to augmentation and, as incomes rise, demand for goods and services increases. Also, hours worked steadily decline over a long period of time.

In the pre-digital era, mechanization did not displace jobs because machines did not work by themselves. Machines were used to make things more quickly, with higher quality outputs, and even to make things that were impossible to make by hand. Machines augmented and replaced humans in physical tasks but controlling and governing the machines required humans. That’s no longer true in the digital era.

Mr. Spence writes that for the first time in the 60 plus years era of computers, we have powerful machines in the information, coordination, and decision (ICD) layer. It was the ability of the ICD layer to function remotely that allowed many economies to keep operating during COVID with limited mobility and physical contact. The ICD layer is the governing and control mechanism in the economy. It has powerful machines and software that automate, replace, and sometimes outperform people in some tasks while augmenting people in other tasks.

The full economic impact of AI has yet to be realized writes Mr. Spence. The pandemic accelerated digital adoption across a range of sectors that had been lagging. The first round of automation was not AI-driven but required codification of tasks. The second round of automation and augmentation enabled by machine learning is in its early stages with its full impact not yet known.

Around the year 2000, white-collar and blue-collar jobs in which routine tasks were a large component began shrinking as automated tools were able to replicate their tasks digitally. Many of these jobs were middle income and thus, many middle-income jobs were eliminated. Income distribution was flattened, the ends of the normal distribution curve were increased as the peak in the middle was reduced. If the peak in the middle continues to flatten, there could be a barbell effect where there are few middle-income jobs but many lower income and many upper income jobs. The outcome of this would be a greater disparity in economic well-being.

According to Mr. Spence, the skills and human capital side of the job market are not fixed. They move at a slower pace than the market itself. People look for and invest in skills and human capital that are in demand in job categories with higher incomes. The pace depends on the number of entities that are investing. Sometimes partnerships are formed with the government and business or education and training institutions to increase the numbers of people qualified for newly created positions.

Another gatekeeping factor with respect to transitioning skills is the distribution of income and wealth. Individuals who want to invest personally will need to invest time and financial resources regardless of the quality of their outside support (e.g., company tuition reimbursement or financial aid). If the income and wealth inequity is extreme, then those in the lower end of the distribution will find it difficult to make the investments in their own human capital, particularly if there is not the availability of outside funding for the education or training and low-cost public services for childcare, etc. This reminds me of the findings in Nickel and Dimed, the book written by Barbara Ehrenreich. Ms. Ehrenreich found in her undercover roles as a low wage earner that many people had to work at least a second job to afford housing. If you must work a second job to keep a roof over your head, you’re unlikely to be able to afford the time to retrain as well as the monetary capital.

There is a circularity in this cycle as Mr. Spence points out. When middle-income jobs are eliminated through automation, wages are suppressed in the middle-income distribution. For people who are pushed toward the lower deciles on the income distribution curve, it becomes more challenging for those individuals to invest their way out. Income distribution is both an outcome of and an input to the digital transitions in work. Mr. Spence writes that “the appropriate conclusion seems to be that policies that directly address high income inequality will turn out to contribute to successful work transitions, even if that is not the primary purpose of the policies.”

There is a beneficial impact from the skills adjustment process on income distribution. According to Mr. Spence, the skills adjustment process increases the supply of people that are in high demand increasing their incomes and partially offsetting the adverse initial effect of automation. Mr. Spence states that it is impossible to know if the skills adjustment process can eliminate the adverse effects of automation because we don’t know how many people can be upskilled from the lower income population.

There are four takeaways that Mr. Spence wants the readers of his article to understand:

  1. We are dealing with complex structural changes and transitions in work, skill requirements, and human capital and it is not balanced.
  2. The purpose and endpoint of this transition is to turn automation into digital augmentation.
  3. Technology is not stationary and creates a constantly moving target, especially as it concerns applications of AI.
  4. Extreme income inequality combined with institutional and policy shortfalls risk turning a complex transition into a trap for the lower-income part of the population.

Mr. Spence adds that the initial part of his article focused on the impact of AI and machine learning in developed countries, but the impact does not stop there. He writes that there are two important classes of developing economies where the impact of automation and augmentation has large current and future impacts. These are middle income countries (also known as emerging economies) and lower-income countries in which growth and development is in the early stages.

Emerging economies have resources and reasonably well-developed digital infrastructure according to Mr. Spence. However, they still have lower income portions of the population with little to no access to information and related services. In these economies, digital transformations are a net positive. These economies have left behind the labor-intensive manufacturing employment engine and are utilizing e-commerce, mobile payments, and fintech to close the service availability gap. The service sectors are expanding as a share of the economy and employment. Digital transformation is creating more jobs than are being eliminated. In these economies, it’s more about training than retraining. These economies are leapfrogging intermediate steps of growth.

The rapid spread of the digital economy is leading to the expansion of entrepreneurship in emerging economies. As this trend gains momentum, it creates employment opportunities for the younger part of the population. Entry barriers to digital automation are low and the capital requirements are low making them ideal for fostering innovation and entrepreneurship.

For lower-income countries, the situation is similar in some ways and challenging in others. The mobile internet has closed the gap in terms of digital infrastructure although more is needed to catch up with emerging economies. Lower-income countries need a growth model that leverages global economic demand and technology. The core of the traditional exports for these countries has been labor-intensive goods and process-oriented manufacturing. Its low labor cost model has been a powerful competitor globally.

AI, machine learning, and robotics are cutting into the economic model that relies on low-cost labor. Because of this, it is less important to locate manufacturing plants in areas of low labor cost. If the digital infrastructure is in place, Mr. Spence believes that the benefits of the digital advances enjoyed by emerging economies can be applied to lower-income economies.

While automation can eliminate jobs, Mr. Spence writes that the most common result is augmentation where automation replaces part of the job and changes the nature of the work. Being prepared to step up to the higher level of work is imperative for someone in a middle-income job to avoid slipping to a lower income role.

I wish I could provide a synopsis of all the articles in this issue of Daedalus. As I mentioned before, this issue isn’t intended to be an AI and machine learning primer, but rather a wide-ranging number of thought-provoking articles on the impact of AI. There is no monetary cost to read them, just the human capital (your time). I think you will find it worthwhile.

Subjects of Interest


Higher Education

Independent Schools


Student Persistence