Suppose it’s 1794 when Eli Whitney invented the cotton gin.
Or perhaps it was 354 years earlier when Johannes Gutenberg invented the printing press. Maybe you prefer December 17, 1903, the date of the Wright brothers’ first flight? Or 1957, when Sputnik was launched into space, man’s first venture outside Earth’s atmosphere?
Those extraordinary inventions – alongside dozens of others – revolutionized how we farm, transmit information, travel long distances or explore realms beyond our planet. Yet, they may prove inconsequential compared to the arrival of artificial intelligence (AI). That, at least, is how many fans and foes alike of AI view the transformative potential of machine intelligence.
AI is defined in the Oxford English Dictionary as the “development of computer systems able to perform tasks normally requiring human intelligence.” However, that definition is sufficiently vague to encompass tasks that ordinary calculators have managed for several generations with the ability to add, subtract, multiply or divide.
Therefore, what distinguishes AI from mere computing is the ability of computer systems to learn, adapt and hence acquire intelligence, allowing them to respond dynamically to changing situations in ways that mimic and even exceed human ability. Or, as the father of the term AI, Professor John McCarthy of Stanford University put it in 1955, AI is “the science and engineering of intelligent machines.”
AI spans autonomous systems (e.g., robots that can self-navigate), machine learning (algorithmic pattern recognition based on large volumes of data), learning via labeling (using images to accelerate pattern recognition) and deep learning (e.g., using neural networks to draw inference, including from smaller samples).
Using AI, machines can now recognize voices and images, respond to spoken and written queries, best grandmasters in games such as chess and “Go,” and navigate complex and dynamically changing routes.
While there is considerable debate about when AI was first “invented,” it is now old and advanced enough to be commonplace in many everyday usages, among them:
Economists have long understood that sustainable increases in living standards require continuous innovations that enable ongoing increases in output per hour worked (i.e., productivity). For most of human history, innovations either belonged to improvements in human capital (human know-how) or in physical capital (better tools with which to carry out tasks more efficiently).
AI, however, does not fit neatly into those distinctions. It may or may not improve human knowledge, skills or intelligence, and it may or may not provide workers with better tools to become more productive.
Instead, it may replace them.
Put differently, traditional innovations made existing workers more productive because they became “smarter,” gave them tools to produce more, or replaced humans with more productive machines.
But not all earlier innovations replaced workers.
For instance, printing innovations and, much later, the internet allowed workers to increase their skills and knowledge to become more productive. In pre-history, the taming of fire vastly improved human caloric input (in particular, of animal-based proteins), while the invention of the wheel enabled people and goods to be transported to where they were most needed.
From the end of the 19th century to the final decades of the 20th century, space and time shrank sharply owing to inventions that allowed instantaneous communications worldwide (e.g., the telegraph, telephone and internet) and to innovations that compressed geography (e.g., the steamship, train, automobile, jet aircraft). Compressing distance in communications and transportation underpinned significant productivity gains associated with rapidly growing international trade and finance, which unleashed millions into the global labor force, many with much-improved jobs and incomes. Those gains, moreover, were distinctly positive sum.
Over the course of human history, therefore, productive innovations have both complemented human labor and replaced it. Fire, the wheel, and information were chiefly human-productivity-enhancing innovations. Other innovations, such as tractors, threshers, harvesters, and trucks, replaced humans in agriculture. So did word processors and answering machines, which displaced secretaries and other administrative staff in the workplace.
One of the biggest emerging questions in economics, therefore, is whether AI will tend to be more of a complement or substitute to human inputs in production. In truth, no one yet knows, but perhaps for that reason, AI today is as much feared as it is admired.
The substitution of machines for human inputs has been a recurring theme of modern capitalism, beginning with the depopulation of agriculture in the United Kingdom and the United States in the century between roughly 1850-1950. That period saw over half of all Americans move from farming into industry and services, becoming uprooted from rural areas to urban and suburban ones. More recently, machines (including robots) have replaced factory workers, resulting in large-scale declines in employment across many industries (most famously, along the automobile assembly line).
Captured in literature, movies and the arts, those vast dislocations, and the human pain and suffering they induced, are etched into the consciousness of peoples around the world. Modernity has always had a rough edge and has probably invited more trepidation than utopian dreaming over the several centuries of its existence.
It is, therefore, unsurprising that modernity’s latest incarnation – human-like machines powered by AI – strikes fear among much of the population.
But if history is our guide, it demonstrates that most innovations proceed apace, not respecting fear or tradition and that, at best, guardrails can be established to limit some of the innovations’ most nefarious impacts. Given the number of actors engaged worldwide in the development of AI and the fact that by its nature, it can develop autonomously, it is difficult to see how it could be stopped, even if that were desirable (which it may not be).
Accordingly, the test will be whether institutions of human invention that establish the rules under which we co-exist – democracy and the rule of law chief among them – will have the durability and flexibility to establish the conditions under which AI improves the human condition rather than harms it.
Innovation often evokes images of vast fortunes that are quickly made: Rockefellers and oil, Ford and automobiles, Gates and personal computers, Zuckerberg and social media. But while those narratives are demonstrably true, they also embed the notion of memory bias. History is written by and about the winners, with little space given to the losers.
Some of our readers may vaguely recall other once-famous names: Wang Computers, Pets.com, Netscape, or Friendster. Those are but a few in a long list of names in the age of the computer, the internet, and social media, that were once darlings of Wall Street but subsequently fell flat on their faces.
There is a lesson in the juxtaposition of groundbreaking innovation that makes fortunes for some and footnotes for others, and it is that separating the two in real-time is difficult, if not impossible. It could be advisable to spread one’s investment capital across the sector to capture the outsized potential returns of the theme without unwarranted reliance on separating winners and losers before that is even possible.
At the same time, a deeper understanding of how broad themes can be applied to the security selection process and provide a larger potential for excess returns is needed because these themes are often viewed as too difficult to measure. Thematic investing leaves space for more macroscopic decisions – those who can deeply understand a theme can potentially earn a larger alpha than has been earned recently through traditional active management.
In the analogy of baseball and statistics, scoring runs and winning games is about avoiding outs and advancing runners on offense, pitching and fielding well on defense. Teams with superstars don’t always or perhaps most often, win pennants.
At least, that is what ChatGPT told me.