Welcome to this edition of The Process, a free weekly newsletter and companion podcast discussing the future of technology and philanthropic work.
If you enjoy reading and listening, please consider subscribing and sharing with your community!
Thank you!
The famous pipe. How people reproached me for it! And yet, could you stuff my pipe? No, it's just a representation, is it not? So if I had written on my picture "This is a pipe", I'd have been lying!
— René Magritte
Belgian surrealist, René Magritte’s famous 1929 painting, The Treachery of Images, depicts a wooden pipe above a caption that reads, “Ceci n’est pas une pipe.” This is not a pipe. The image is clearly of a pipe, and yet Magritte’s point is to remind viewers they are actually looking at oil paints arranged on canvas in a way so as to cue our brains to the concept of an actual wooden tobacco pipe. I wonder what Magritte would make of generative AI.
This recipe for leek and potato soup, for example, spun up by ChatGPT looks right and may even function well (let me know if you try it!), but is it a recipe or is it something else?
Leek and Potato Soup Recipe: Ingredients: 2 tablespoons of butter; 2 medium-sized leeks (white and light green parts only), washed and chopped; 3 medium-sized potatoes, peeled and diced; 4 cups of chicken or vegetable broth; 1 cup of whole milk; Salt and pepper to taste; Chopped fresh chives for garnish. '
Instructions: In a large pot or Dutch oven, melt the butter over medium heat. Add the chopped leeks and sauté for 5-7 minutes until they soften and become translucent. Add the diced potatoes and sauté for another 2-3 minutes. Pour in the chicken or vegetable broth and bring the mixture to a boil. Reduce the heat to low and let the soup simmer for 15-20 minutes or until the potatoes are cooked through and soft. Using an immersion blender or transferring the soup to a blender, puree the soup until it is smooth. Add the whole milk to the soup and stir until it is well combined. Season the soup with salt and pepper to taste. Serve the soup hot, garnished with chopped fresh chives. Enjoy your delicious and warming bowl of leek and potato soup!
Leek and Potato Soup — ChatGPT
These large language models (LLMs) create their outputs by choosing each successive word based on probability. The recipe above, while not entirely random like, say, last week’s Powerball numbers, was generated in a mechanically similar way by popping up what the model determined is the most-likely-next-word, one-by-one.
ChatGPT, like Magritte nearly 100 years ago, has arranged words on a screen in a way so as to imitate a recipe for leek and potato soup. Ceci n'est pas une recette. This is not a recipe.
In the tweet above, OpenAI’s CEO, Sam Altman, reminds people of the power of simple directives, iterated at scale over time. Applying unfathomable amounts of computing power to these prediction programs is revealing something almost like a mathematical DNA sequence for content including text, images, video, and, perhaps something much more.
Images of people’s faces, for example, may all look unique to us, but LLMs see common patterns and probabilities in the construction of human faces that allow them to generate photorealistic images of people who have never existed, one pixel at a time. Here’s a website where you can test yourself to see if you can distinguish real people from AI generated photos.
This raises at least two uncomfortable question. First, with enough computing power and data, can everything be decoded? And second, if so, what is human creativity?
There doesn’t appear to be consensus among AI experts about the theoretical limits of what LLMs can achieve as their size and complexity continue to increase. Some believe these models will plateau at some point and the marginal gains from, say, GPT7 to GPT8 won’t be as dramatically felt as from GPT2 to GPT3.
Others believe there is no ceiling, or that it is a long way off, and we will see continual 10X increases in the abilities of these models on a regular basis. In the near term, it’s reasonable to assume anything humans have digitized, from graphics to genomics, humans will also put these data into AI models to see what patterns and abilities emerge. It’s not only the power of these models that we will feel, but the rapidly increasing diversity of their abilities will be astonishing as well.
On the question of creativity, here’s an excerpt from a recent podcast conversation between Ezra Klein and his guest, science fiction writer, Adrian Tchaikovsky, in which they begin parsing the nature of creativity.
EZRA KLEIN: The size of the island that is human creativity, that is what counts as a creative act, in that telling, it’s worth noting how small it is now. When you say, can A.I. create work that is not derivative? Largely, I agree with you, it can’t.
But most human beings create work that is derivative. And a tremendous amount of certainly the creative economy — and I don’t say this with any sense of judgment. I don’t think there’s anything wrong with this — is work that is derivative.
My son adores the — my four-year-old adores the TV show, “Blaze and the Monster Machines.” I actually think “Blaze and the Monster Machines” is great. Big fan. But episode to episode it’s pretty derivative. But a lot of people are working on that. They’re putting their time into it. But the meaning of it isn’t really the point. And the non-derivativeness isn’t the point.
It’s a common complaint now that we’re endlessly recycling the same IP. There’s nothing we will not turn into a movie from old board games. And we’re remaking all kinds of things. And everything is a sequel on top of a sequel on top of a sequel on top of a sequel. And the secret of it — the thing I think people don’t really want to face up to is people like things with familiarity to it. They like a certain amount of derivativeness in their work.
And so, of course, there’s great art being done that is truly new, truly non-derivative, truly made with meaning as the central intention. But so much of human creation is not about that. And so many of the people who are creators, that is not what they are doing. And it is not what they have been asked to do. And to some degree, if you’ve got to beat the A.I.s that are going to come in five years, it may not be what most of us can do. I mean, there may always be a space for the very best.
But I find it really much more frightening than I think people want to give it credit for if you say that the A.I. will be able to do anything so long as it is somewhat derivative of everything that has come up until this point, or so long as meaning — a really deep structure of meaning is not central to the project. Because just how many things that we make really do satisfy the meaning is central and this is non-derivative conditions? I think it’s pretty small.
ADRIAN TCHAIKOVSKY: Yeah, and I have no answer. I think everything you said is pretty much bang on the money. There was a rather depressing cartoon that might have been in the British magazine “Private Eye” of how we thought the future would be, which is a human painting a picture, and in the background, the robot is doing the vacuuming. And then how the future is going to be, which is the robot painting the picture and the human doing the vacuuming
Klein points out much of what we commonly refer to as creative content is actually derivative, a riff on something else that already exists. He clarifies this is not a bad thing or a judgement, but a recognition of our reality, which is people seek familiar enjoyable experiences. What worries him and Tchaikovsky is derivative creation, remixing new versions of what came before, is what AI excels at far beyond what humans are capable of.
When it comes to picking and choosing between the benefits and drawbacks of generative AI, we probably can’t have our cake and eat it too. Many creative jobs will be replaced by AI systems, and there will also be many more people who gain the opportunity to be creative. Both will be true.
The X factor will be how the value of creativity evolves as it becomes decoupled from profit-making. As I wrote about last week, it is almost impossible to predict or imagine what will happen to creative endeavors as people gain back time to be creative and create for reasons other than making money. Both of these directions seem to be far more human-centered than our current system, which why I am eager to encourage progress in this direction.
Thanks for reading this edition of The Process. Please share and comment or you can email me at philip.deng@grantable.co
Philip Deng is the CEO of Grantable, a company building AI-powered grant writing software to empower mission-driven organizations to access grant funding they deserve.
Free Webinar: Unlocking the Potential of AI for Nonprofits: Overcoming Fear with Curiosity
As Artificial Intelligence grows in usage in society, what does it mean for nonprofit and philanthropy? How should we feel about it? How would we use it? Can it take on some tasks like grant applications? What are the ethical and equity considerations? Will it take over our jobs?
Join us for this conversation with Beth Kanter and Allison Fine, authors of The Smart Nonprofit, Philip Deng the creator of Grantable (an AI-supported grantwriting platform), and Vu Le of NonprofitAF.com
Automatic captions will be enabled