My non-prediction for generative AI in 2024
ππ ππ¨π ππ πππ§ ππππ
I decided to stop using generative AI to produce the images for ππ ππ¨π at the end of last year when I expanded to Substack. I would like to say this decision was reached after a considered analysis of the ethical problems raised by the technology. The energy used for each image generation may be equivalent to fully charging your smartphone and asking an LLM a question may consume as much power as leaving a low-brightness LED lightbulb on for one hour. The diminishing job opportunities for freelance illustrators and writers seem real. Those considerations played a role! However, the honest answer is that the biggest reason I stopped was that I got bored. Trying to coax DALL-E or Ideogram into giving me the image I wanted was frustrating. I get more satisfaction from searching Unsplash or Smithsonian Open Access for human-created images, and it takes less energy, from me and from the electrical grid.
I started using image generators for the same reason I started the blog, to better understand the tools that are changing our work as teachers and writers. Maybe
convinced me with one of his posts last spring, or maybe it wasnβt based on reason at all. I saw Mollick, Laura Dumin, , and Jason Guyla doing something cool on LinkedIn1 and like a teenager on TikTok, I copied them and posted the results. We are social animals, after all.There has been intense coverage of the news about the New York Times suing OpenAI over copyright infringement. A bit undercovered is that the lawsuit represents a breakdown in the negotiation between the two over how much OpenAI should pay to use the content NYT produces. So this is really an argument about price, not principle. For an interesting take on AI and copyright recorded before the big news, see the latest episode of Decoder with Nilay Patel featuring Adobeβs general counsel Dana Rao. Adobeβs approach to generative AI is similar to Getty Images in that they only use images they own or license instead of scraping the web for everything.
Is my boredom with DALL-E a sign of how things will go with generative AI this year? It seems entirely possible. Will the hype cycle end with a whimper, not a bang? Will we grow tired of our new toys? Will their social costs will come into sharper focus? Will the moral panic over homework fade? Will the companies that stole much of the data used to train their models face judgment? Will GPT-5 and Gemni show that we have reached the upper limit of the transformer technologyβs capabilities? All possible, even likely.
predicts that generative AI will turn out to be βjust a big upgrade to existing machine learning systems.β He says, βThink of it like replacing a Playstation 3 with a Playstation 5. Thatβs a huge improvement. But itβs still a gaming console.β On the other hand, we have a compelling post on One Useful Thing that the coming year will see impacts on education, knowledge verification, and work that are, perhaps, more intense than last yearβs. Plus, the usually sober reporters at the MIT Technology Review think we will see video generation and large models out into the world via robots. So maybe itβs a Playstation 5 with a great VR headset?Impacts depend not just on a technologyβs capabilities but also on the speed with which these new tools are adopted. And that is nearly impossible to know. The companies selling the products will only show us the numbers they want us to see. What happens in classrooms and offices is mostly a black box and I think we really want to keep it that way (Please letβs not use this new technology to surveil teachers and workers). In the absence of reliable data, we are left to our own devices, literally, peering into our screens at regurgitated press releases, the few academic studies trickling out, and our own interactions with large models. Then typing words into our devices about what we see there. And hopefully remembering to put down our devices, at least occasionally.
I have always admired Wendell Berryβs reminders that so many of the bad things happening around us are the result of collective decisions for which each of us has some small degree of responsibility. After all, we can say no. Berryβs βWhy I Am Not Going to Buy A Computerβ appeared in Harperβs Magazine in 1988 and generated several letters in response, which were later published along with Berryβs replies. The essay describes the tools he uses to write and why he prefers them to a computer. The opening lines are:
Like almost everybody else, I am hooked to the energy corporations, which I do not admire. In my work, I try to be as little hooked to them as possible. As a farmer, I do almost all of my work with horses. As a writer, I work with a pencil or a pen and a piece of paper.
One letter accused him of hypocrisy, saying βI find it ironic that a writer who sees the underlying connectedness of things would allow his diatribe against computers to be published in a magazine that carries ads for the National Rural Electric Cooperative Association, Marlboro, Phillips Petroleum..β
Berryβs response to this letter has always struck me as a good description of how to live in a technological world we did not ourselves create but want to change for the better.
That I am a person of this century and am implicated in many practices that I regret is fully acknowledged at the beginning of my essay. I did not say that I proposed to end forthwith all my involvement in harmful technology, for I do not know how to do that. I said merely that I want to limit such involvement, and to a certain extent I do know how to do that. If some technology does damage to the worldβ¦then why is it not reasonable, and indeed moral, to try to limit oneβs use of that technology?
I use a computer and I will continue to use generative AI, but not that which is trained on data that is stolen and that includes the absolute worst output of humanity. My point is not to urge you to draw the same line, merely that drawing lines is something we should all do.
The work of AI Log is toβ¦well, log my ongoing engagement with generative AI. Here is an essay about my experiments with AI image generators back when I used them. I just republished it on Substack but it was originally posted on LinkedIn.
I am aware that βdoing something coolβ and βLinkedInβ are mutually exclusive. My kids keep me honest about such matters.
π¨π° π³ππ Β© 2024 by Rob Nelson is licensed under CC BY-SA 4.0.
I like your approach!