On this episode, Julianne Pepitone sits down with Kim Branson, Global Head of Artificial Intelligence and Machine Learning at GSK to learn about using models to gather data and create stackable knowledge for future generations.
HIGHLIGHTS FROM THIS EPISODE:
Historically, drug development meant you as the research team, come up with your hypothesis, design studies, and then try to validate that hypothesis. But artificial intelligence changes that entirely. And as Kim would explain what he’s describing is a whole new approach to the concept of knowledge itself. It sounds like the AI approach just completely flips that traditional paradigm on its ear. It’s not just about creating a model to solve the problem, it’s how you think about the problem itself. It helps you find and solve questions you didn’t even know that you have.
The key thing is that nobody works at GSK forever. Unfortunately, it’s a 300 year old company. So how do we keep the memory of all the things we’ve done before? Or even the breadth of different things we’ve got? We can build AI systems that know all these types of things. So we think in terms of building: every time you do an experiment, you’re not only doing it for you, you’re doing it for future you or someone else. So we’d be very careful about the data you collect, the metadata, and making as useful as possible. Every piece stacks into this increasing body of knowledge that we can learn from.
And obviously not all the world’s science happens within GSK. This model not only knows about what Granddad did, but also knows about what Grandma might have done at a rival company 30 years ago as well. So we want to basically establish this continual corpus of knowledge. Every time we run a clinical trial, even if unfortunately it failed, we should be able to learn something, so we increase our probability of success for the next time we do something. So it’s this increasing sort of body of knowledge, isn’t it? What we call stackable knowledge is key to our thinking.