Play It Again, Claude
About 130 years ago, the job of pianist was automated when Edwin Votey created the first player piano. The machine worked by reading music that was encoded by holes punched into rolls of paper, which in turn directed airflows to levers that depressed piano keys. The human’s task was relegated to pumping a foot pedal to create the pneumatic pressure that drove the automaton.
Things got worse for the human pianist from there.
By the early 1900s, player pianos had evolved to more fully reproduce a human performance, including subtle dynamics like tempo changes and the introduction of a damper pedal. The human role went from deskilled to fully deprecated as electric motors replaced foot-powered bellows. With the Seeburg Lilliputian Model L, the only job left for humans who wanted to play the piano in the 1920s was to put in a coin.
Nearly every major pianist of the early 20th century made music for these machines. Echoing AI commentary today, some musicians viewed the player piano as not just replicating human playing, but exceeding it. The Russian composer Igor Stravinsky explained that he wrote pieces specifically for the machines because “there are tone combinations beyond my ten fingers,” and argued that “there is a new polyphonic truth in the player-piano … There are new possibilities. It is something more.”
How could humans possibly compete? Yet today you are more likely to encounter a piano player than a player piano, despite the job being successfully automated a very long time ago. The automatons have been relegated to museums and the rare curiosity. Pianists can be found any night of the week in hotel lobbies, Italian restaurants, and concert halls.
When it comes to the arts, this is hardly an isolated example of automation failing to cause mass unemployment. Musicians broadly have faced mechanical and digital competition for well over a century, first from the phonograph, then the radio, and later the instantaneous on-demand technologies that emerged with the internet and Spotify.
Each of these technological turns produced complaints and concerns that bear a similarity to things being said about AI today. One of these warnings came from John Philip Sousa, who in 1906 used the term canned music to deride the output of “music-reproducing machines.”
The phrase rose to new popularity in the late 1920s, when recorded music began replacing the live orchestras that accompanied silent movies. Musicians were undoubtedly concerned about this new technology. The union-funded Music Defense League spent hundreds of thousands of dollars on cartoons and ads to try to turn the public against this trend. In what amounted to misplaced optimism, the labor union’s president argued that “the public will demand personal appearances instead of mechanized music,” and union officials even declared in 1929 that, as a New York Times report put it, “the decline in the number of movie orchestras” has stopped, and “many houses are re-hiring their musicians.”
Movie orchestras and the jobs they supported were doomed. You would be hard-pressed to find one today. In a narrow sense, this shows the limits of the ability of human artists to resist automation. But the job market for musicians broadly has grown since the invention of recorded music. Data from the Census Bureau show that the number of individuals employed as musicians today is at an all-time high. Musicians were displaced in some tasks, but as society collectively grew richer, the number of paying opportunities for musicians went up as well.
Nor are these jobs available only to those with the talent and determination to achieve the apex of musical success at Carnegie Hall or the Grand Ole Opry. In bars across the country, local bands of limited ability entertain crowds despite the competition from nearly free recordings of the greatest musical acts in history.
What has provided musicians with this protection from replacement during a century of competition with increasingly sophisticated automation? As Sousa argued more than 100 years ago, “The nightingale’s song is delightful because the nightingale herself gives it forth.” In the bloodless language of economics, consumer demand places value on who actually provides certain goods and services. Art is not just a physical good, but a who, what, where, and when. We see this reflected in the marketplace of music every day, in the continued existence of the bar band as well as in the 10 million fans who paid to see Taylor Swift’s Eras Tour.
[Read: What made Taylor Swift’s concert unbelievable]
This may do little to assuage the concerns of those who do not aspire to become a musician and whose current job seems likely to be replaced by AI. But the demand for the human touch is not limited to music. It is valuable across the economy, in a wide variety of goods and services and the jobs that create them.
The demand for the human touch is one reason there are still millions of waiters despite the potential to automate them with QR codes and ordering tablets. There are more than 10 million people employed in sales roles today despite the ability to buy and sell just about anything online, the rise of self-checkout, retail kiosks, and many other automating technologies.
This demand for the human touch appears to grow with income. The level of restaurant service and the number of jobs needed to provide it tends to go up with the size of the bill. Fine dining likely involves not only a waiter but someone who advises on wines, someone to keep the table clean as you dine, someone to bring out the cheese cart. In economics terms, the human touch is a normal good, something that a richer society demands more of.
Our willingness to pay for the human touch does not mean that AI will not be disruptive to the labor market. There are still many jobs that AI will be able to do and where consumers won’t mind and may even value the absence of humans. Like the members of the movie-theater orchestra, some people will need to find their way from tasks where they have been displaced by technology to those where they haven’t.
What the demand for human touch does tell us is that there will always be jobs that consumers do not want AI to do. The existence of some demand for human labor is consequential. When there is some demand for a good or a service, we can spur more. There are many ways we could redistribute an increase in national income: for instance, making taxes more progressive, or sending out annual stimulus checks. We could introduce more novel policy solutions like a wage subsidy, through which the government would boost the take-home pay of lower-wage jobs.
This would be a very different future from one where humans and our labor are fully displaced by thinking machines.