Dr. David Fajgenbaum

We train the algorithm over and over again on all the known treatments, and it gets really good at picking up the pattern of connections.

Okay, if that drug treats that disease and that drug treats that disease, and then we say, okay, now that you know what a good treatment looks like in this map of knowledge, now give me a score from zero to one for how likely every other drug is to treat every other disease.

There's 4,000 drugs, there's 18,000 diseases.

We're basically asking it to calculate 75 million times, the 4,000 drugs against 18,000 diseases, 75 million times how likely every drug is to treat every disease.

And it's between zero and one.

So then our medical team, we look at the things at the very top and say, what are the 0.99s?

What's machine learning, artificial intelligence saying is the highest, most likely drug to treat a disease?

And that's where we start.

Wow.

So it sifts through everything.

It sifts through everything.

It goes through the world's knowledge.

And actually, when we ran it for the first time, Sean, we built the nonprofit three years ago, spent a lot of our first year fundraising.

But we finally hired our first data scientist.

We finally ran the algorithm for the first time about two and a half years ago.

And it took 100 days to run the algorithm.

to like literally scan everything the world knows about everything, 100 days.

And we didn't know it would take 100 days, but like just every day it's like slow and we don't have the results back.

Finally, the results 100 days.

Sean, now it takes 17 hours to scan the world's knowledge of everything to get a score, 75 million scores in 17 hours, which the reason that's really important is because then you can actually improve the algorithm and say, that's actually not the way to do it.