We know that AI has helped us to make big advances in medical science: in mapping the human genome, and in solving for proteins (with Alphafold, etc) and in various kinds of diagnosis, drug discovery, and beyond. In fact, that may be the single biggest contribution of AI to date, and saves quite a few lives on a daily basis. There are the more quotidian types of screening that may trigger an important intervention, and then there are the broader paths of clinical analysis, that take time, and money – and a lot of patience.

I was privileged to interview Daphne Koller at the Imagination in Action event in California, our Next Endeavor summit Sept. 14-15. (Disclaimer: I help to put on these events.) Koller founded Coursera, so we covered education a bit, but right now, she’s involved in research that helps to promote better avenues for the kinds of medical innovations I mentioned above. So we talked a good bit about biology and the adjacent sciences, too.

Physics and Biology in Practice

“The world that we live in right now is the world where AI bits meet real-world atoms, Koller said. “And I think people often underappreciate just how much complexity exists at that interface, and the fact that when you move out of the realm of the world where all of this data is being generated by the minds of eight billion people—images, text, and so on, on the web, code that you can generate and test immediately, and so on—and you have to exist in a world where (your efforts are) actually working, you’re working at the pace of the physical world. You’re no longer working at the speed of light.”

But, she noted, it’s also the world of the human body, and that’s where a lot of our conversation went, in terms of applied biology aided by AI.

“You can make a prediction of what will happen if you perform an intervention in the real world,” Koller explained, “the ‘real world’ being the real-world body of individual patients like all of us.”

“How do you make that prediction about that counterfactual world that has never happened, with data that isn’t actually doing that experiment? And so what we’ve had to do in the last eight years—and that’s why it took eight years—is to collect a unique-of-its-kind data set that combines observational data from millions of people with a causal underpinning in human genetics, and combine that with causal data in experimental systems that we can generate and perturb at the scales of tens of billions, soon to be hundreds of billions, of cells. And then bring that together, to the point that we can make accurate predictions in what we call the virtual human, which is a first-of-its-kind, we believe, AI model for human causal biology, making causal predictions in the real world of human beings.”

I asked Koller about her essay, “drug discovery has no magic wands” – and why she chose the particular moment to write it.

“I think that there is a lot of enthusiasm and excitement about the role that AI can play in the physical world, and specifically the role that AI can play in human health,” she said. “And I think that is not an unreasonable thing, and we all would love for AI to do something that helps make the world better, as opposed to, you know, all the predictions of doom and gloom. “

We talked about predictions that are aspirationally correct, but on timelines that might not be realistic. Koller pointed out that, in the grand scheme of things, a lot of the cures implemented to date have been applied to infectious diseases, and that it’s easier to slow a pathology than to eliminate it.

“We need to understand what is feasible and what it will take to get there,” she said.

Alphafold, we noted, has done a lot.

Continuing Work on Solutions

One challenge that Koller referred to in medicine is that we can solve for something in terms of chemistry, but the way to real progress might be a little different.

“We do not understand the biology,” she said of these kinds of situations. “It is not a chemistry problem. It is a biology problem. And the amount of data that we’ve been able to acquire for understanding human biology is a de minimis fraction of what we need.”

For one thing, she added, there’s the challenge of collecting things only from humans. Other factors, she suggested, include risk aversion and current numbers in the drug world, and the challenges of FDA approval.

One example of a tough condition is ALS, a severe condition that affects tens of thousands of Americans.

“The drugs that are out there extend lifespan by two to three months,” Koller said. “That’s all they do. They’re palliative at best … that is the problem that I think presents an incredible opportunity for uncovering new biology that will give us all, you know, potentially, the opportunity to go after disease after disease after disease, and come up with drugs that actually work, that are actually, in another term of art, disease-modifying.”

Correlation, not Causation

We need AI, she said, to be causal.

“If you talk to any statistician or economist or epidemiologist or clinician, they will tell you that correlation and causation are very, very different things,” Koller explained. “And the place where people often go wrong is because they interpret correlation as causation, and they do something where two things happen to be correlated … but… one has no consequence on the other.”

One response to this that Koller described is the creation of what she called a “causal data factory.”

“We take billions and billions of cells from different cell lineages,” he said. “Some of them are neurons. Some of them are liver cells, cardiac cells, immune cells. We intervene in them by knocking down a particular gene, amplifying a particular gene, and then seeing what happens.”

This leads to great observability.

“It’s amazing what AI combined with advanced microscopy techniques can see in a human cell, especially when it’s a living human cell, because we have live-cell microscopy,” she continued. “We also look at what happens to other transcripts using sequencing-type technologies. And we’re collecting the largest causal database of human biology that has ever existed, when you look at cells. And then we’re combining that with causal data that you get from humans.”

All of that is going to be vital to cutting-edge medicine in the years moving forward.

In going over the interplay of education and continual progress in science, we discussed what things will be likely to look like soon.

“I believe that more and more companies are going to be AI companies that have a strong vertical presence,” Koller said, promoting the value of proprietary harnesses and specialized tools in a particular vertical that “go deep” to solve modern problems.

“We’re absolutely developing cutting-edge causal AI models that do not currently exist, but at the same time, we’re solving a problem that matters to real people,” she said.

In terms of education, where AI tutors are now ascendant, I asked Koller: does the traditional lecture survive in the age of AI?

“I don’t know that lectures are entirely dead,” Koller said, “because there is something so inspiring about being taught by somebody who’s a truly gifted and inspiring teacher.”

“I think that we absolutely need to be mindful of the consequences of our actions in the world today,” Koller said, in aid of exploring our role as engineers. “And frankly, I think that a lot of people would benefit, not just if you’re a machine learning scientist, but in general, from thinking about the consequences of our actions in terms of environmental impact … machine learning is very much on our minds here in this room, but there are other disciplines that have, I think, significant world consequences … and I think that we would all benefit from giving more thought to the consequences of our actions in the world, and trying to pivot more of our activities, whatever discipline we’re in, toward doing it not just for gain, but also for good. So, doing good, while we do well at the same time.”

Koller also had this to say about the role of universities:

“I think that, in general, universities should be trying to make the walls of the ivory tower more porous, so that there is a greater movement of people, ideas, and IP from industry into the academic world and back again,” she said. “I think that would make the work in academic labs more impactful, and better able to impact society.”

In examining this further, Koller talked a bit about why she left Stanford, trying to achieve the transfer of ideas to their real-world applications.

“I think universities should take a deep, hard look at how they operate and how they can regain the impactful position that they had 100 years ago in a world that is so drastically different today,” she said. “And so I’m thankful that I’m not a university president, because, boy, do they have a lot to think about right now. But I really think there is an incredible opportunity for the more forward-thinking (person) to do something very different.”

I’ve already done a lot of quoting here, partly because I like the articulation with which Koller addresses these questions, but I want to include a good bit of her closing quote, which is, I think, about the greater challenges of personal achievement in any real pursuit:

“I think that there is a tremendous wealth of opportunities for those of us who are willing to adopt the unease and lack of comfort, the discomfort of being outside of our own discipline,” she said. “So, being at the boundary between a discipline where you feel like you’re swimming like a fish in water, and one where you feel like you’re asking really dumb questions all the time because you don’t understand the basics.”

“I think that (above scenario) is really uncomfortable for people, but it’s also where the greatest impact is to be had, because that is a place where you can take ideas that are very natural to you, and do something that is completely transformative in the other discipline. And whether that is AI for, you know, medicine, like what I’m doing, or AI for agriculture, or AI for government, or - I can think of probably a dozen, two dozen, maybe more disciplines where AI hasn’t really permeated - the impact is so huge. And do you want to do the latest twiddle on the next, you know, slight variant of a transformer model, or do you want to do something that is truly impactful? I would think the latter offers a lot of opportunity. So I would encourage people to leave your comfort zone and do a little bit of that.”

So there you have it: quite a lot of input from someone who has done so much in these related fields. We have education, and entrepreneurship, and administration, and private sector growth… all centered around our ability to make progress, together.