Leader of China’s genomics powerhouse steps down to pursue AI research

Deep learning (or more powerful machine learning techniques generally) will be important for making CRISPR useful.

For instance, we'll need to be able to more accurately predict the result of highly complex interactions that would arise from modifying DNA. We do have the ability to generate tons of data, but we have less ability to analyze it and build predictive models.

Deep learning is the best bet today of being able to do that on huge quantities of data where the subject matter experts really don't know what "high level" features they're looking for. But they know the "result" and want to tell "good result" from "bad result" (or from 10,000 other results). Deep learning is good at (and getting better at) filling in the blanks in between, the high level features.

I've been interested in deep learning for several years, and CRISPR is something that's recently come on my radar that I'm becoming convinced will be similarly impactful. And to be clear about "impactful", I mean we're in 1992 if you compare The Internet to advances in machine learning (which, it is hoped, will somehow lead us to a breakthrough in AGI, or at least a better definition of intelligence). So the technology is here and will continue to advance (probably exponentially) as it's productized, refined, and ultimately commoditized over the next two decades.