A conversation with Artificial Intelligence expert Daphne Koller

San Francisco Ballet discusses artificial intelligence with Daphne Koller

Daphne Koller is a Machine Learning pioneer and the founder and CEO of insitro, a biotech company that leads AI-driven drug discovery. She is also the co-founder of the online learning platform Coursera, and was a professor in the department of computer science at Stanford University and a MacArthur Foundation fellowship recipient.

How has the development of AI and/or Machine Learning changed over the last five years? And how do you think it will change in the next five, ten, or fifteen years?

We’re on what is clearly an exponential curve in terms of the capabilities of this technology, but the challenge of an exponential curve is that the human mind doesn’t really understand it. It’s not possible for us to understand something that gets twice as good every year. But you can look back and observe the capabilities of the technology in, say, recognizing what’s in an image. If you go back ten years, it was barely at the level of, “Oh, I think I see a dog.” And half the time maybe it was a dog, and half the time it was something different. The technology we have today can not only write an entire story about what’s in an image but can even generate an image based on a verbal description. We can describe an image and have the machine generate it and then say, “Could you make the dog jump higher, be furrier, be friendlier?” with all sorts of qualitative descriptors, and the machine will generate something that looks real. It’s incredible. And I think because this is an exponential curve, that progression will only continue and it’s very difficult to foresee even two to three years ahead, let alone ten years ahead. As I said, exponential curves are something that a human mind just doesn’t get.

What is the potential you see in using AI and/or technology as a tool for accessibility to something like arts, education, or healthcare?

Those are quite different use cases. If I had to generalize, I would say that those are going to be transformative to accessibility but in different ways. In the arts, if you think about someone who is, for example, visually impaired and having the computer be able to describe what is being perceived in ways that anyone can understand, or conversely, if you’re someone who’s hard of hearing, you can make a piece of music and create visual imagery that’s evocative of what’s being played, which I think is incredibly exciting. In education, I think that this is going to be a truly transformative capability. There’s a study by Benjamin Bloom from 40 years ago called “the two sigma problem”, where he showed that a student who has access to a private tutor performs two standard deviations better than a student who’s in a standard classroom. The problem is that we’ve never had the ability economically to provide a private tutor to every student, but now we can. So there’s the opportunity to take kids who are basically performing at the median or even far below the median and ‘up level’ them in a way that would not have been imaginable without this technology.

In what ways do you see your colleagues using AI in your day-to-day life?

One of the challenges I see specifically in the biological sciences is that we don’t have the ability to understand it, because it’s so complicated. It’s this emergent thing that doesn’t have natural laws. And that makes it very difficult for us to come up with meaningful ways of intervening in it. The benefit of what we can do with AI, and what we’re doing at insitro, is to take biology, measure it in a whole slew of different ways enabled by cell biologists, bioengineers, and multiple technologies, and then let machines do what they now do far better than people, which is find patterns in these data. The data provides us with the ability to predict what certain interventions might do. To me, this is the beginning of a new discipline, which I call digital biology, in which we’re able to read biology at unprecedented fidelity and scale; interpret what we read by the tools of data science, machine learning, and AI; and then write biology to behave differently than it would normally otherwise do. That could have an impact in healthcare, in environment and sustainable agriculture, in biomaterials and biofuels. There’s a tremendous opportunity to transform the world around us by being able to understand and edit it.

How can the Bay Area community cultivate a curiosity about AI?

I think you could reasonably say that the Bay Area is in some ways the birthplace of AI, the home of AI, even all the way back to John McCarthy, who’s the father of AI in the old days, who was a professor at Stanford. Certainly, when you look at some of the newer technological developments, they came out of Google, they came out of Stanford. As such, I would hope that a lot of people, whether they are technical or otherwise, are able to develop an excitement and a curiosity of being at this inception, if you will, of this new world that we’re going to be part of. And the nice thing about this new generation of AI, unlike a lot of the earlier incarnations, is that it’s accessible to everybody. Everyone can go and get an account on chatGPT, Llama, Claude, or any of the others. You don’t need to know how to program, you just need to have a good idea and good structured thinking to get the computer to do something that maybe no one has ever gotten it to do before. I think it’s an incredibly accessible technology that will become more so over time and I encourage people to go and try it out.

Is there anything you’d like to add before we conclude?

I think, based on my own personal experience, that some of the most exciting impacts for this technology are going to lie at the boundaries with other disciplines. Whether that discipline is biology and the work that I do, education that we’ve alluded to earlier, the arts, or many others. I would encourage people to think about where they could see AI pushing past those boundaries into these other disciplines. The best work is going to come from the synergistic discussions between people with very different backgrounds and perspectives coming together, creating something that has not existed before. I think these partnerships are going to give us the most exciting new developments.

Headshot photo © insitro

Header background image: Imagery trained and prompted by Hamill Industries // Courtesy of San Francisco Ballet, source photos by Lindsey Rallo