Automated Art

A new economics book, The Rise and Fall of American Growth, by Robert Gordon, analyzes productivity growth in the U.S. Economists associate productivity growth closely with the long-term standard of living in a country, and Gordon finds that U.S. productivity growth peaked in the 1950's and is unlikely ever to reach that level again or even come close. Gordon believes that there will be almost no growth in household incomes through 2040. This flies in the face of rhetoric in the media and politics proclaiming that with the right political leaders and policies in place the U.S. could return to 1950's-like conditions in which nearly everyone lives well. I don't think I'll read the book, because its thesis seems rather obvious to me, but I thought I'd mention it because it relates to my thoughts about technology's effects on how people earn their livings. There doesn't seem to be much research underway on what people are likely to be doing for work in fifty years, and it seems that the work environment, if it still exists then, will be radically different due to advances in technology.

The technology-related changes during my career in printing were enormous. As of the late 1970's, most print preparation involved pasting set type onto a board in the correct position and then photographing it. Black and white photographs were manually converted into halftones by placing a screen over the negative and photographing it again. Color photographs were once manually separated into four colors with filters on a large camera, and later a color transparency was electronically separated using a scanner, which was quite expensive in the 1980's. Once all of the image components were in negative form, a stripper would place them in the correct positions and manually align all four color negatives. These "flats" were then placed in step-and-repeat machines to make lithographic printing plates. Stripping and platemaking departments used to employ a relatively large number of people, but now, with the electronic processing of images, stripping is obsolete, and a sizeable plant can operate with just one platemaker per shift. Similarly, improved electronic controls on printing presses have made it possible to operate them with smaller crews, and newer presses run much faster than earlier presses, reducing the amount of labor per sheet. On top of this, Internet advertising has been cutting into print sales. The net result of new technology is the disappearance of thousands of jobs in a once thriving industry.

I bring this up because the same phenomenon has occurred in all manufacturing industries and has been spreading to white-collar jobs for some time. With advances in AI it is easy to imagine a state in which one doctor, lawyer or engineer will do the work that twenty once did. Some manual labor and customer service jobs may survive for a few more years, because low wages obviate the need for expensive technology, but if the cost of technology falls low enough, those jobs could go too. I like thought experiments and speculate that some of the last vocations to fall may be in the arts. This will come about because technology will be able to mimic human behavior and skills even if we never arrive at a singularity. Though I think something resembling a singularity will probably occur, that would not be necessary for reconfiguring art as we know it; it could happen before then if machines began to pass the Turing test with the art they produced. All this means is that people would be unable to distinguish a painting, sculpture, novel, poem, film, musical composition, etc., created by a machine from one created by a human.

You can see the early beginnings of this transition in the algorithms currently used to determine what films a Netflix subscriber would enjoy. Despite the fact that Netflix's recommendations for me are almost always wrong, they seem to have identified some of the characteristics that I like in a film. For instance, they know that I like suspenseful psychological movies, foreign movies, thrillers, understated movies and cerebral movies. That isn't much, but it's a start. Why couldn't a sophisticated computer analyze a large body of art, identify the characteristics that appeal to people and then formulate new works based on human preferences?  Several kinds of fiction-writing software already exist, and there is no reason to assume that a completely automated novel that satisfies most readers couldn't be produced in the future. With the right software, a computer could sift through a large database to identify themes and subplots that appeal to people and generate writing styles that would be agreeable to specific groups. The result would not necessarily be robotic, because the program could be calibrated to insert random events that simulate events in novels written by humans. As a matter of fact, it is soothing, repetitive qualities that draw many readers to fiction, and repetitive qualities are the ones that would be easiest to detect through analysis.

One art that would be particularly amenable to automated creation would be painting. Ever since Warhol, the critical standards for what constitutes a good painting have fallen frighteningly low. There is little reason to suppose that an extraordinary number of mediocre paintings couldn't be deemed brilliant and fetch large sums if promoted in the right way by the right people. Capitalism has infiltrated the art world to such a degree that aesthetic values are often predefined by the financial interests of members of the relevant art community and never subjected to the judgment of independent experts who have no conflicting interests. For example, it might be possible to create what looks like a Warhol painting and sell it for a million dollars if it were marketed properly by insiders of the art world. We usually think of artistic value as a private experience, often in terms of the beauty of an object, but art's commodification in recent decades seems to have redefined it so as to place an emphasis on market value as the dominant factor.

Perhaps the fundamental question here is whether creativity, vision and skill can be automated. Certainly skill can be, as machines are already capable of performing with greater physical precision than humans are capable in some areas. With the current state of technology, creativity and vision can be simulated, perhaps convincingly enough to fool some people as to their authenticity, but without escaping the notice of more sophisticated observers. However, if you believe, as I do, that humans are not the ne plus ultra of the universe, an allowance must be made for the possibility that our best art could one day be matched or surpassed by superintelligence.