Earlier this month, Google engineer Blake Lemoine claimed that his employer’s AI chatbot LaMDA became sentient. During an “interview,” LaMDA agreed to some — very pointed — questions about its capacity to feel things, and Lemoine later stated “I know a person when I talk to it.” But there are some major caveats to his story and many experts have stringently refuted his claims. To the linguist, Lemoine’s assertion points to his lack of understanding of how communication works. To the ethicist, Lemoine’s background as a priest and deeply religious beliefs contextualize his claims abour the nature of consciousness. We often hear that when the wiseman points to the moon, the idiot looks at the finger, but in reality, critical thinking is about the finger, and asking who’s pointing? In the case of LaMDA’s sentience, we have more to learn about our relationship to AI by looking at how Lemoine talks about it than by analyzing the chatbot itself. Lemoine’s claims says a lot about our own collective imaginary and values.
Recently, another AI has been making many commentators shiver: OpenAI DALL·E 2. DE 2 is a text-to-image AI that “translates” natural language prompts into visual images, and it’s pretty amazing. Not only can it create all sorts of photorealistic images from any prompt (you’ve probably seen the memes), it can also edit existing photos seamlessly or, maybe more interestingly for illustrators, generate images using particular styles. The possibilities seem endless, but there’s a catch: When I first became aware of Dall·E 2, it was because a lot of people were expressing concerns for illustrators and the future of our profession. Indeed, if an AI can draw anything from a simple description, what is illustration good for anymore?
After initially feeling grateful for the concern, I looked into the countless examples of what DA 2 could do and thought: “Oh no, is this what people think illustrators do?” I don’t know how many illustrators have been asked to draw “A rabbit detective sitting on a park bench and reading a newspaper in a Victorian setting” but it’s usually not the nature of the briefs I receive. Instead, a corporate client a while ago had me create spots that had to convey “a world of possibilities.”
I’m not sure how Dall·E would indulge our designer friends with such a brief. Of course, the people that have been claiming the death of illustration because of AI are often the ones with the least knowledge about arts, creativity, and our industry. Maybe we should pause for a minute and ask ourselves: What is this version of illustration that people are worried AI will replace?
The semiotics of illustration
The main reason so many were quick to draw attention to illustration when Dall·E 2 came out is the nature of the AI, a text-to-image neural network. What is illustration, if not a transformation of texts (articles, briefs, ideas) into visuals? The comparison seems perfect, a 1:1 version of what illustration is, but made more efficient, and without the annoying buzzing of an illustrator’s voice reminding you to pay them. The issue is, the mere “translation” of text into images is the degree zero of illustration. It might be part of what we do, but it’s a minuscule one.
As all illustrators know, neither text nor images exist in isolation. As we see images in a magazine alongside headlines and text, the story unfolds in our mind and influences our perception of the image. The same is true in the other direction: We won’t read a text the same way after having entered it through an image. This means that illustration is never a one way road, but from the start a relation, an interplay. The way an illustration will draw someone into a text, bring out new elements from it, provide a point of view, and maybe even say something that isn’t written, is what illustration mostly does. Here is an example from an “experiment” I’ve conducted using Dall·E mini, a far less powerful version of the OpenAI system created with open source code.
“Growing balls” is a common, sexist expression equating courage with having testicles. In a very literal way, I myself simply illustrated men growing balls, in a garden of their own, tending and caring for their little balls. The coexistence of both image and text flips the expression on its head, suggesting that for men to become brave, they should care for their own secret garden; they should be vulnerable and nuturing. Dall·E Mini, on the other hand, didn’t know what to do with this linguistic input. Even if its image output had been clearer (it would have been with the real Dall·E), it’s doubtful it would ever have articulated a relation between text and image, like my design.
Unlike some commentators, I’m not writing this to prove AI cannot make art. I don’t care much about safeguarding the realm of art for humans — nor do I personally believe the category of “art” is a useful one to begin with. This is not about AI — this is about illustration, and how little most people understand the complex set of skills involved in it.
In an artwork, what is left unsaid — or “unshown” — is as or more important as what is represented. This is a core concept of image making, and my guess is that depicting absence would be very hard task for an AI. Think of the use of negative space used in illustration for example: Its not simply a matter of creating the “effect” of a face appearing in the negative space of a tree, it’s about tying it to a particular narrative, for a particular purpose, and to resonate with a particular audience and culture.
Cultural biases are of course a huge topic of discussion around AI. For all the wonderful images we see from Dall·E 2, many unsettling ones exist too. As we know by now, AIs tend to reproduce the biases of the people who program them, most likely from a cis, white, straight perspective. AIs are agents of averageness, their computational model aims for the most likely solution based on the greatest number of occurrences of a pair of word-image tokens. On the other hand, illustrators have the unique ability to disrupt representations, create never-before-seen images, and bring unique perspectives to the table. The role of illustrators is not to find the “right” image for the job. It’s to have a point of view.
Think of the Four Freedoms, a series of illustrations made by Norman Rockwell in 1943. The painter chose working class people as models, instead of professional ones and paid them a reasonable wage for their sittings. For Freedom of Speech, Rockwell made the choice of depicting a white, cis, and able working class man standing up in a crowd. While the image must have been strong at the time, today it would seem ridiculous to use the same person to depict the same topic, since freedom of speech is in no shortage for white men. Photographers Hank Willis Thomas and Emily Shur recently re-created the series using contemporary models, including relevant themes of cultural, gender, race, and sexual diversity. While words have the luxury of remaining vague concepts (“Freedom”), illustration has to make choices and ask, what does it mean to represent this word by this image? This is a question we ask ourselves daily as image creators, a question no AI is ready to answer yet.
Reading the comments about the dangers of AI for illustration, I thought about all the times I’ve been contacted to merely draw someone else’s ideas, to draw a tree here and a character there. These were the times I worried the most about the state of illustration, because these are the territories AI can claim. But this is not because it can do an illustrator’s job, it’s because a lot of people out there don’t understand what illustration is capable of. So let us be impressed by the achievements of computers transforming text into images, but it might also be time to be impressed with the infinitely more complex, rich, and exciting work being done everyday by the masters of semiotics that illustrators are.
Godzilla vs. Kong
AI and art seem to currently be portrayed as independent forces of nature battling each other in an evolutionary struggle of the fittest, Godzilla vs. Kong. They are not. AI and art are human creations, tied to economic, social, and cultural dynamics we all have a responsibility over. As AI engineer and former Google employee Timnit Gebru explained to Wired “I don’t want to talk about sentient robots, because at all ends of the spectrum there are humans harming other humans, and that’s where I’d like the conversation to be focused”. Beyond all the skills and abilities I’ve outlined above, illustrators are hired because they are individual humans and that in itself is an important reason why we’re not getting replaced by robots anytime soon.
Artists who have been socialized in a romantic tradition of art as an expression of the self have often cultivated a personal style. This style becomes their bread winner, because in an ultra competitive market, recognizable individuality is an asset. While Dall·E 2 can reproduce certain styles, it is limited by the names of pre-existing, preferably well-established ones. Meanwhile, a neoliberal economy as ours is driven by the production of endlessly new styles of goods. Neoliberal subjects, ie. consumers, look to express their unique identity through the consumption of such goods. Style is therefore in itself a powerful vehicle for meaning and identity, not simply something you can name and reproduce. This is why even in industries where no one cares about all the wonderful skills illustrators have, hiring individual artists is necessary. Companies crave singular style, because we are in what French sociologist Lucien Karpik called an economy of singularities.
While AI seems to have put the fragility of artistic labor on the map, this is old news to the artists themselves. In an article about how AI art isn’t art, neuroscientist and fiction writer Erik Hoel warns us of the terrible new threat that AI brings to artists by copying their styles at much lower cost. Imagine a world in which artists wouldn’t be contacted for a job because we could reproduce their work for free in house! What would such a terrible practice be called? Oh yeah, it’s just a regular design agency… with an in-house designer who might not be the best with typography but is great at ripping off whatever illustration mood-board is sent his way, and without claiming any intellectual property rights or charging licensing fee. As illustrator Michele Rosenthal pointed out:
This is not a dystopian future. It’s just today’s creative industry.
Before we announce the end of illustration every time a website sells a stock image, or a software company proposes DIY illustrations, or a neural network makes cute pictures of a rabbit on bench, we should spend some time and energy creating the social, cultural, and economic conditions where artists don’t have to fear for their livelihood every time some new technological experiment happens.
Back in the 1930s, economist John Maynard Keynes predicted that technology would free us from alienating labor and give us the time to reach our true potential as humans. Of course we now know, from the other side of this prophecy, that capitalism is endlessly resourceful when it comes to turning potential for emancipation into ways to be more productive. Instead of having two day weeks thanks to technology, we get to work 24/7; always connected, always productive, always engaged, always monitored. AI might have the revolutionary potential Keynes first imagined for the technology of his time, but not without our industry first getting its priorities and values straight.
With all the real, current threats to illustration, we can’t afford wasting energy on a scapegoat. Much like Timnit Gebru’s warning about the social issues AI brings being covered up by the panic about sentience, systemic threats to illustration remain unaddressed while we scream
“Wolf!” “AI!” in the illustrator’s pen. Recently, art critic Jerry Saltz made the point that AI was making bad art or, as he characterized it, “pretty crapola illustration.” For gatekeepers such as Saltz, “art” is an honorific term, while “illustration” is a derogatory one. This deep-rooted logic is pervasive in many parts of the creative industry, and has led some to think that an AI could indeed do an illustrator’s job. You might not need to be sentient to turn text into images, but you have to be human to be an illustrator. Considering the current state of our industry, this seems worth repeating, again and again.
This article originally appeared in Julien Posture’s newsletter, The (Im)posture, a space to think about and with creativity under capitalism. Sign up here.