A History of AI: The Furthest Extension of Automation (1900-2022)
Introduction
While electricity spawned many important society-altering inventions such as the lightbulb and the electric motor, the invention of computing and its furthest extension—artificial intelligence (AI)—has had the largest runaway catalytic effect on our world. At its core, artificial intelligence is the result of the human desire, and trend of outsourcing human labor. We can see this desire throughout history taking many forms; in the period of colonialism, humans enslaved
others to perform their labor, and throughout history humans harnessed the forces of nature through inventions such as waterwheels and dams. These latter inventions generated mechanical power through harnessing the mechanical power in nature. Even on a more basic level, in many tribal societies, social hierarchies distributed labor in a way such that many individuals performed the labor of the leaders. While this trend of outsourcing has been present throughout history, outsourcing labor still was limited to the mechanical labor that animals and humans could perform. With electrification this began to shift—electricity could transform into mechanical power and magnify the work one individual could accomplish on a far greater scale than any individual could have previously done. Before looking at the reasons for AI being the most runaway catalytic result of electrification, it is first important to gain some knowledge about its history to contextualize where we are today. Since the term was termed by McCarthy, many have opposed it fearing an Ai uprising that would spell the end of human civilization. Conversely others purported it could be the key to a utopian future. One way to understand Artificial intelligence is through investigating both the cultural perceptions of previous mechanization events in history and looking at the public perceptions that accompanied them. To understand the scale of artificial intelligence we must first look to the history of its development, quantify its impact, and then compare it to previous keystone electrical technologies of the past.
Historical Background and Context of AI
The term Artificial intelligence was first coined by John McCarthy in 1956. John McCarthy was a mathematician, and together with Marvin Minsky, convinced Claude Shannon and Nathaniel Rochester (Designers of IBM’s first computers) to create a conference at Dartmouth to advance computers in simulating human intelligence. While there wasn’t much visual progress in AI in the preceding decades, visible steps started to become apparent in 2011. In 2016 and 2017, DeepMind’s AlphaGo defeated Lee Sedol and Ke Jie, who are the world’s top Go players. Because of the complexity of the game, many people thought that this would never be possible for an Ai to beat a human. Since this event, AI has been seen as a serious and relevant topic, and Ai as a tool is now well established in our world with many businesses adopting the technology. The technological progression to this point though was much steadier than the public perception of AI’s progression. Arthur Samuel created a checkers Ai program in 1952 that eventually was able to consistently beat him and the progression of Ai according to Moore’s law predicted Ai to progress to the level where it could beat a human at chess to the exact year that it happened. These small steps were apparent in the small social circles invested in Ai development, yet not to the public. One question that comes up around artificial intelligence is how you define intelligence. Often when trying to answer the question of intelligence, we defer to the human definition which is the ability to reach a goal. This goal can be of varying levels of complexity, but it always comes back to this definition. At the Dartmouth conference McCarthy argued that any human task-oriented goal could be replicated by a computer. This was part of the impetus for creating the conference and this small specific scope is where Ai lived during its development. Even today Ai is still very confined to specific parameters.
Automation in the Factory
Before industrialization the amount of work that could be done was on a human scale—it was limited to the physical abilities of a human and the amount of work they could accomplish. Industrialization and electrification of the factory began to change the scale of what could be accomplished. The invention of the electric motor increased overall factory production and the invention of management technologies did so as well. Some of the first jobs to be taken over by mechanization were assembly line jobs. Whereas before humans were needed to transport materials and products in a factory, with the onset of factory electrification, assembly lines and conveyor belts took their place. The result was that many lost their jobs. While there was a lot of push back from the work force and an increasing fear that mechanization would take over many jobs, at the same time mechanization generally made working conditions better. Like factory mechanization and automation through electrical technologies, we can see AI begin to have a displacing effect.
Computers only ever replace humans when they can perform a given task better and more efficiently than a human could. They also take over jobs when they perform jobs that humans can’t perform. One of the fist examples of a computer AI outpacing a human in a specific task is when a program defeated grandmaster Gary Kasparov in chess. On May 11, 1997, deep blue defeated Gary Kasparov who was arguably one of the best chess players of all time. This set a precedent to the public that computers could perform tasks better than humans. While general AI was far away, and still is today, we started to see AI take over tasks that cut out many workers in the following decades. Deep blue didn’t eliminate the need for chess players, since it is a largely social and recreational game, but when it comes to economics, and business, hiring an AI is the obvious choice for businesses wanting to cut costs. In a study by Eric Dahlin, regression models suggest that many high skill and some middle skill occupations would have more demand while there would be no increase for other jobs. This trend is like mechanization during the first two decades of the 20th century when there was a large demand for engineers and high skill jobs while the blue-collar labor was increasingly automated. One effect of this trend is a growing wealth disparity which is not desirable. This can be noted as one negative effect of mechanization. Additionally, it leads to one potential argument for why automation using AI is one of the most impactful events of the 21st century and in the history of humanity. If large decisions are being delegated to a decreasing number of highly skilled workers with access to AI technologies, and AI technologies are much better at executing specific tasks than humans’ nearly limitless scalability, then the impact of these few decision makers on the world will be extreme. It is dangerous to concentrate that scale of influence in the hands of few individuals for the same reason the emperor model in Ancient Rome was so volatile: individuals are unpredictable.
While a growing demand for high skilled labor is one effect of automation, conversely, mass layoffs of low skilled labor is another side of the same coin. During the first two decades of the 20th century, the worker turnover rate was nearly 300% which was largely due to bad working conditions, layoffs, and poor compensation. In the 21st century we see a similar trend of low skilled jobs being displaced by mechanization that uses AI.
Quantifying AI’s Impact
When it comes to quantifying the impact of AI, it is difficult to know what to include and what to exclude. What kinds of impact are the ones that should be counted and how do we determine these metrics? Looking back to early stages of electrification in the factory, there are several metrics that stand out. First is the metric of safety and health. The invention of the lightbulb led to a transition away from gas lamps which was previously the main form of lighting. This led to less fires and better temperatures. The invention of the electric motor also took away some labor-intensive jobs which one could argue is increasing safe working conditions by eliminating the dangerous jobs. There were still dangerous working conditions resulting from electrification, but many of the accidents caused in the factory during the early 20th century were largely due to bad safety precautions and policies. It was more of a human and institutional error than it was an error of the new technology. While there is a lot of fear surrounding Artificial intelligence, we see a similar condition where it is the human side of the equation that is leading to the undesirable consequences. One such example of this is the Ai algorithms used for social media. Since the proliferation and widespread adoption of various social media platforms including Facebook, twitter and Instagram, rates of teen depression and anxiety have increased. This increase has been linked to social media use—namely the design of the applications and the goal of the algorithms. We have not reached the point where we can have general Ai that performs a variety of tasks and functions. Rather, we have specific Ai’s that perform specific functions. We have speech recognition AI, Image recognition AI, an AI that plays chess, and so on. Each AI is dependent on all of the inputs it receives and is dependent on the goal of the developers. The key is that the individuals making the Ai are the ones who choose the goals of the algorithms. In the case of social media, the goal is to increase clickthrough and time on the applications. This leads the Ai to effectively hijack the users’ reward system to keep them on the applications. This is known to have negative mental health effects, yet these companies still optimize the algorithms for clickthrough. In this example we can see how the technology isn’t necessarily at fault for generating widespread negative externalities, rather it is the human component. Understanding this through, there is still the question of if and how Ai is the most impactful form of mechanization.
Scales of Impact
When we look at impact it is important to see the number of people impacted by the technology and the scale of cascading influence. We can look at the number of people who are affected by Ai each day as a number, and as a percentage of the world population. We can also look at the categories of effects that these technologies have on people. Before looking at Ai though it is useful to get some perspective through looking at the impact and scale of other electrical technologies.
Starting with impact, while it is difficult to quantify the number of people impacted by Ai directly, we can look to the physical technologies that people use daily that use Ai systems. The system that houses AI is the computer which was first developed in the 1930’s largely by mathematician Alan Turing. In his essay “On Computable Numbers, with an application to the Entscheidungsproblem,” he laid out the foundation of modern computer science which eventually allowed him to create a system that could interpret and remember information. At the time his computer was only useful for decoding encrypted messages. In the next 80 years, the computer went from a secretive military tool to a mass-produced product that in 2019 was in 47% of households worldwide. In 2019 the world population was 7.684 billion, 47% of which is 3.611 billion. So, by 2019, 3.611 billion people were on the internet and were interacting with a device that had AI capabilities. The next question is how and when was Ai implemented on a large scale and what did this look like.
As of 2021, 37% of businesses and organizations employ AI. There were 33.2 million jobs in 2022 meaning that 11.9 million businesses were using AI in some capacity by this year. A third is a significant percentage of the market and with individuals interacting with businesses daily, the reach of AI is large on these terms. Considering that the scale of influence of one AI algorithm is global, seeing the wide adoption of AI across various industries, and the trend of decisions being concentrated increasingly in the hands of few individuals. I would argue that the impact that artificial intelligence has in terms of how many people are impacted and to what degree is larger than any other electrical technology in history.
Sources
Russell Stuart J. 2019. Human Compatible: Artificial Intelligence and the Problem of Control. New York New York: Viking.
Nye David E. 2001. Electrifying America: Social Meanings of a New Technology 1880-1940. Cambridge Massachusetts: MIT Press.
Energy and Civilization
Samuel’s Checkers Player original paper
Frank MR, Autor D, Bessen JE, Brynjolfsson E, Cebrian M, Deming DJ, Feldman M, Groh M, Lobo J, Moro E, Wang D, Youn H, Rahwan I. Toward understanding the impact of artificial intelligence on labor. Proc Natl Acad Sci U S A. 2019 Apr 2;116(14):6531-6539. doi: 10.1073/pnas.1900949116. Epub 2019 Mar 25. PMID: 30910965; PMCID: PMC6452673.
https://www.statista.com/statistics/748551/worldwide-households-with-computer/ https://www.census.gov/programs-surveys/susb/data.html
samh.blog 