A friend going through the motions. Worn husband in a rut out of habit. A traveler’s lagged message of longing fraught with exhaustion. A pent-up, unwelcome or badly timed kiss. These were some of the explanations that resonated in my mind after I watched the Strange digital art trifle Powered by the Emoji Mashup Bot, a popular but no longer existing Twitter account that combines the parts of two emojis into new, interesting, and surprising combinations. The robot took its hands and eyes from the yawning emoji and mashed them together with the mouth from the 😘 heart emoji. That’s it.
Compare this simple method to the supposedly more sophisticated machine learning-based generative tools that have become popular in the last year or so. When I asked Midjourney, an AI-based art creator, to create a new emoji based on those same two, he produced compositions that were definitely emoji but had none of the style or significance of simple blending: a series of yellow, heart-shaped objects with Protruding tongues. One of them seemed to be eating another’s tongue. I was all struck by the kinds of atrocities that might be offered as prizes for carnival games, or as junk-mailed stickers intended for children’s cancer fundraisers.
ChatGPT, the beloved text generation bot, has never looked better. I asked him to create descriptions of new emoji based on parts of existing ones. Her ideas were fine but mundane: a “yawning sun” emoji, with a yellow face and open mouth, to represent a sleepy or lazy day; A “multitasking” emoji, with eyes looking in different directions, to represent the act of juggling multiple tasks at once. I took these descriptions back to Midjourney and got succinct but nice results: a group of glaring suns, a series of eyes on a yellow face dripping from above with a black tar-like spray.
I could probably have drafted better prompts or spent more time improving my results on ChatGPT and Midjourney. But these two programs are the pinnacle of AI-driven generative creativity research, and when it comes to making new, expressive emoji, they’re outdone by a very simple computer program that picks facial parts from a hat and puts them together.
People have dreams of creating artificial intelligence. For starters, they’re computer dreams: that by simply feeding terabytes of text and image data, software can deploy something like a machine’s imagination. author It works instead of just taking it out. But this dream involves visualization: AI generators like ChatGPT, DALL-E, and Midjourney can make any A kind of creativity with the same ease and performance. Their creators and advocates consider them capable of handling every form of human intelligence—like everything is born.
And not without reason: these tools can create a copy of almost anything. Many of these versions are false, misleading, or even potentially dangerous. And a lot of them aren’t very interesting, as the emoji examples show. Using a software tool that can make a particular thing is completely different — and a lot more fun — than using a tool that can make anything at all, as it turns out.
Kate Compton, a computer science professor at Northwestern University who has been making generative arts software for more than a decade, doesn’t think her tools are artificially intelligent — or smart at all. “When I make a tool,” Compton told me, “I’ve made a little creature that can make something.” This thing is usually more expressive than useful: its bots visualize the inner thoughts of Lost Tesla Independent and draw pictures Alien virtual spacecraft. View similar tools Hipster cocktail recipes Or naming fake British cities. Whatever their goal, Compton does not aspire for software generators like this to master their field. Instead, she hopes they will present the “small, slightly dumber version of him.”
This is a far cry from ChatGPT creator OpenAI’s ambition: to build artificial general intelligence, “highly autonomous systems that will outperform humans in most economically valuable work.” Microsoft, which has already invested $1 billion in OpenAI, is reportedly in talks to inject another $10 billion into the company. This kind of money assumes that technology can generate huge profits in the future. Which makes Compton’s claim all the more shocking. What if all that money is chasing after a bad idea?
One of Compton’s most successful tools is a generator called Tracery, which uses content templates and lists to generate text. Unlike ChatGPT and its cousins, which are trained on huge data sets, Tracery requires users to create an explicit structure, called “context-free grammar,” as a model for its output. The tool has been used to create Twitter bots in various forms, including Thinking playgrounds And abstract landscapes.
Context-free grammar works like nested Mad Lib. You write a bunch of templates (say, “Sorry I didn’t make it in [event]. I had had [problem]. ’) and content to fill in these forms (problems can be ‘gut’, ‘freak’, ‘explosive diarrhea’ or ‘ [conflict] with me [relative]”), and the rules bring them together. This requires the author of the generative art to consider the structure of the thing he wants to create, rather than requiring the program to output it, as he might do with ChatGPT or Midjourney. The creator of the Emoji Mashup Bot, a developer named Louan Bengmah, had to split each Code the source emoji into a bunch of parts before writing a program that puts them back together in new configurations.It takes a lot of effort, not to mention some technical proficiency.
For Compton, that effort isn’t something to disavow—it’s the point of the exercise. She told me, “If I just want to make something, I can make something.” “If I want to make something, I can make something.” A contra open AI mission, Compton sees the purpose of generative software differently: The practice of making software tools is like giving birth to a software creature (“a chibi version of the system,” she told me) that can make something—often bad, weird, or, anyway. Anyway, caricature versions of it — and spend some time communicating with this creature, as you might with a toy dog, a small child, or a kind-hearted alien. The goal is not to produce the best or most accurate likeness of a hipster cocktail menu or a mountain scene at dawn, but to capture something truer than reality. ChatGPT’s ideas for new emoji are workable, but the Emoji Mashup Bot offers seem to fit; You can use them instead of just posting the fact that the computer generated them.
“Maybe that’s what we’ve been missing in generators of everything,” Compton said: an understanding of what the machine is trying to create in the first place. Looking at a system, seeing the possibilities within it, identifying its patterns, encoding those patterns into programs or data, and then watching the thing work over and over again. When you type something in ChatGPT or DALL-E 2, it’s like tossing a coin into a wishing well and pulling back the bucket to find a pile of kelp or a puppy in its place. But Compton generators are a lot like putting a coin in a file gashapon machine, knowing in advance what kind of object the thing will parse. This effort refers to a practice whereby the author hopes to help users seek a relationship with their software rather than draw a conclusion from it. (This also explains why Twitter has emerged as such a fruitful host for these bots — the platform originally encouraged caricature, brevity, and repetition.)
Much is gained from showing how a software creator worked, and how its creator understood the patterns that define his subject matter. The Emoji Mashup Bot does this by displaying the two emojis from which it created any given combination. One of the first text generators I remember using was a weird software game called Kant Generator Pro, made for Macs in the 1990s. It used context-free grammars to compose bound text reminiscent of the German Enlightenment philosopher Immanuel Kant, although it also included models for less esoteric compositions, such as notes of thanks. The program came with an editor that allows the user to view or compose grammar rules, providing a way to look under the hood and understand what the program really is.
But such transparency is difficult or impossible in machine learning systems like ChatGPT. No one really knows how or why these AI systems produce their output — and the output can change from moment to moment in inexplicable ways. When I ask ChatGPT about emoji concepts, I have no sense of emoji theory – what patterns or paradigms it interprets as important or relevant. I can examine ChatGPT to explain its action, but the result is never illustrative – it’s just text generated: “To create ideas for emojis, I used my knowledge of common concepts and topics often represented in emojis, as well as my understanding of human emotions, activities, and interests.”
Perhaps, as creative collaboration with software generators becomes more widespread, everything generators will be recast as middleware used by dedicated programs with more specific goals. Compton’s work is charming but doesn’t really aspire to utility, and there are certainly plenty of opportunities for generative AI to help people make useful, even beautiful, things. However, realizing this future will involve a lot more work than simply chatting with a computer program that, at first glance, seems to know something about everything. Once that first color fades, it becomes clear that ChatGPT doesn’t actually know Any thing—instead, he produces compositions that mimic knowledge through a compelling structure. And as the novelty of that surprise wore off, it became clear that ChatGPT was less a magical wish-granting machine than an expository sparring partner, and is the most interesting tool when it’s bad rather than good at its job.
Nobody really wants a tool that can make anything, because that need is a theoretical illusion, a capitalist fantasy, or both. The hope or fear that ChatGPT, Midjourney, or any other AI tool might end expertise, craft, and work betrays an obvious truth: these new tools require whole new systems of expertise, craft, and work. We were playing with tech demos, not finished products. Ultimately, the raw materials for these AI tools will be used for things that people will unfortunately pay for. Some of this new work will be stupid and offensive, as organizations demand to generate value around the AI systems they’ve invested in (Microsoft is reportedly considering adding ChatGPT to Office). Others can be satisfying and even suggestive—if they can convince creators and audiences that the software makes something specific and speaks with intent, giving them a chance to engage in dialogue with it.
At the moment, this dialogue is more simulated than real. Yes, sure, you can “chat” with ChatGPT, and you can loop over images with Midjourney. But an empty feeling arises from many of these encounters, because the program runs through the motions. It seems to be listening and responding, but it’s just processing the input into the output. AI creativity will need to abandon the silly, arrogant dreaminess of general AI in favor of specific details. An infinitely intelligent machine that can make anything useless.