Science, Art, and Stable Diffusion

There’s nothing especially new about observing common ground between art and science. Any good scientist can tell you that some creative imagination is required to ask the right question, to construct the right experiment, to generate new and meaningful data, and to present the entire process with a clear and exciting narrative.

But there is something new happening right now in the world of art creation—and it has the wheels of my scientific imagination turning.

Stable Diffusion is one of a handful of new text-to-image models (see also Dall-E 2 and Midjourney) released in 2022. As someone possessed with the unfortunate combination of a ferocious imagination and little visual-artistic talent, I was an early adopter of Stable Diffusion and have found it to be a revolutionary piece of technology.

The process by which Stable Diffusion works is itself a fascinating achievement. But setting that aside, I want to focus on this exciting piece of technology purely from the user’s perspective: Stable Diffusion allows me to go from a visual idea in my head to a snippet of text on the screen to a realized image of that idea in a few seconds. With some additional tweaks to model parameters and the text prompt, I can essentially edit this image for content, composition, and style. And then I can generate countless iterations and variations of that image until I have exactly what I need.

If you had told me about this technology two years ago, I would have said it was impossible. And yet now this technology is here and it is allowing me to explore countless creative ideas that would have otherwise been impossible to realize.

In the coming year, Prism is experimenting with something similar. How can we use our meta-science technology to go from a question about the state of biomedical evidence to a credible analysis and answer in mere minutes or seconds?

If we can accomplish this, then we can empower strategic, evidence-based thinking in biomedicine that would be too slow and expensive otherwise.

Indeed, we are currently living in the world where, despite knowing that we ought to conduct a thorough systematic review and meta-analysis before undertaking any new clinical trial, we almost never do so. There are many reasons for this, but one of them is because in the months it would take to complete said review, the state of the evidence will have evolved and likely rendered the opportunity for the trial, and the purpose of the review, obsolete.

But what if this were not the case? What if getting an insightful picture of the evidence on any question was no more difficult than writing up a new text prompt for Stable Diffusion? To be sure, there would still be some work required to know how to compose the prompt/question and input parameters such that the “system” (i.e., Prism’s platform) gives you back a meaningful answer. But this is nothing compared to the time and labor required to conduct a systematic review.

Imagine if all you had to do to generate a systematic review was type in your question… You would be foolish not to do this. You should generate a hundred systematic reviews, if they would help you better design your next trial or experiment!

This vision is another way of articulating what we mean at Prism when we talk about our offerings as “consulting at the speed of software”. The industry currently can, and does, hire research consulting firms to do many of their systematic reviews. But you would never hire a firm to do a hundred reviews at once. It would take far too long. It would cost far too much.

But when a hundred reviews are just a few keystrokes away… Think of all the questions you would ask, all the answers you might have, the mastery of the evidence that you could achieve.

And, of course, the inverse of this thought-experiment is to appreciate how many questions, answers, and insights we currently ignore because the industry-standard methods to synthesize evidence and turn it into actionable knowledge are too slow and too crude.

I am delighted to say that Prism’s technology will change this. We do not have the complete text-to-evidence-review product to share just yet, but are working to realize this dream alongside our innovative clients. And just like with Stable Diffusion, I can see the revolution ahead.

If you had told me about this technology two years ago, I would have said it was impossible.

Stay tuned.

‍