Manufacturing Desire

Although most famous for her work on another banality, Hannah Arendt wrote forcefully against the use of cliché verbiage. Said Arendt, “clichés, stock phrases, adherence to conventional, standardized codes of expression and conduct have the socially recognized function of protecting us against reality.”

This seems especially true for those of us who work in media, on both the right and the left. Phrases such as “coastal elites” and “the mainstream media,” not to mention the banality that is “making space” for transgender voices, all too often stand in for actual thought. Such clichés do our thinking for us, and when we use them, as we are especially prone to in speech, it is we who are fashioned to fit language, rather than fashioning language to fit our meaning.

This dynamic is freshly relevant in recent conversations about the impact of artificial intelligence. As technologies such as DALL-E and ChatGPT develop the ability to illustrate and write for us, the question is not, as many have posited, whether the work belongs to us or to the machine, but who is shaping whom.

My husband is not an artist, but this week he created a beautiful painting. The central figure is a blond-locked warrior under a stone lintel, surrounded by three children in saintly white. The background is a sunlit forest. It is a delightful picture, styled not unlike something from the Renaissance. He created it using the platform Midjourney, one of several new AI art generating tools that takes a string of alternative text (“triumphant strong warrior surveying a pleasant garden containing a lovely wife and small children peaceful beauty”) and generates an image from it. The same text at a different time, or on a different computer, will create a different image.

The tool has been heralded by the tech optimists as a breakthrough in the world of art and graphics, saving artists time and money by doing enormous legwork for them. Some artists have already begun to use these images as source texts for paintings, while others might touch them up gently for use in graphic design. More numerous, however, are those using AI platforms to generate the ascendant art of modern democracy: memes. (Various historical figures taking selfies has been a popular one.)

Midjourney, along with two other tools of the same type, Stable Diffusion and DreamUp, is currently being sued for copyright infringement. The artificial intelligence behind these systems was trained to translate text into images in part by mining billions of images from the web—images that may or may not be protected under copyright law.

This new tool is under considerable fire from more traditionally minded artists, too, who see it as a threat to the glory of their craft. Their argument is the same one we’ve heard against Jackson Pollock and Andres Serrano: If anyone could have made it, can we call it art? Defenders of AI art, and of Pollock, will say the difference is that you didn’t make it, and Pollock did, but this is unsatisfactory and anyway it is beside the point.

The point is not one of skill. Rather, it is a matter of taste. We know the difference between Michaelangelo’s Pieta and Salvador Dali’s Lobster Telephone. It may soon take great restraint, however, to prefer the Mona Lisa to the AI Mona Lisa, and this will not simply be because AI will eventually get good (it is not there yet), but because our taste for the real will be supplanted by our access to visuals that are increasingly piquant. In the same way that a chef who has once cooked with MSG, a chemically modified sodium compound that enhances savory flavor well beyond ordinary salt, will rarely return to the usual variety, AI imagery has the power to satiate our desire for ever more exciting or perfect visuals, and to elevate that desire by providing a constraint stream of new delights. We’ve already seen the power of images to rewire what we consider appealing through the epidemic of pornography.

Not only does this tool encourage an insatiable preference for the piquant in those who consume art; it also inhibits the artist from producing art that is genuinely creative. This is helped by the speed at which an AI image is generated. Once, an artist conjured an image entirely in his own head, discovering as he chipped away at the plaster or dabbed in bits of color what the precise shapes of the woods, flowers, and warriors would be. The image in his head was only as sharp as his mind; it came into focus through the act of creating it. In AI art, that image is given immediately. It may be beautiful, perhaps even more beautiful than anything the artist himself could have imagined or created. But is it the same image as the one in his mind’s eye? This is impossible; the artist himself does not know it exactly yet. Instead, he is told what his image could look like—and therefore, what it does become—by the machine.

OpenAI, the creator of the most prestigious of these AI image generators, DALL-E, is also the creator of ChatGPT, the AI text bot. As you might guess, AI in language looks a little different than AI in art, but operates on a similar principle: Users input their request—perhaps for a kind but firm breakup message, or, as many students are discovering, an essay on Shakespeare—and the bot ejects prose. Users can then run variations on the prose or otherwise edit it according to their preferences.

ChatGPT has awakened an even bigger storm than AI art, likely because language, unlike art, is a tool of communication that everyone uses. It is also a medium in which it is much harder to distinguish real from fake. The Washington Post has already released a guide on when it’s acceptable to use a robot to speak for you, an ethic that mostly focuses on transparency, the greatest of all goods in the modern era.

Here again, however, the question of telling real from fake is a serious one, but not the most serious. More concerning is how robotic text generation continues to standardize language, whether for something mostly inconsequential, such as an auto-generated email response, to things far more serious, such as personal messages, essay writing and argumentation, or even poetry. It doesn’t matter if we never use AI to write a letter; once we are accustomed to certain (often sanitized, bureaucratic) turns of phrase in one area, we begin to use them in others, even in speech, allowing them to colonize our meaning and autocorrect our thought. ChatGPT cannot replace Shakespeare, we are told (as though we did not know), but it can replace our ability to think in a way that even begins to approximate the great language wielders, because it replaces our ability to think for ourselves. As in AI art, it is we, rather than the machine, who are trained.

This aspect of machine learning is already true in many aspects of modern life. Most people under the age of 30 cannot recall bits of history or information on their own, since the tool of Google provides these details at will. Even our ability to do mental math is plummeting as we depend on digital calculators to do the simplest addition and multiplication. And, of course, if it weren’t for GPS, we might not know how to get from our house to our parents’ without a robotic voice telling us which exit to take.

Perhaps this is a good thing. The geniuses of any era will stay away from such tools that inhibit real thought and creativity; the rest of us were not geniuses to begin with. When we find ourselves with desires that outstretch nature, and tastes trained to uniformity, perhaps there will be a few men remaining who have retained the true happiness of an ordinary soul. 

The post <strong>Manufacturing Desire</strong> appeared first on The American Conservative.