I particularly happy to be part of this project so we continue to collaborate now that he has left O’Reilly and joined databricks. So I checked, and actually I met him for the first time back in 2013, when I did a one week trip to Silicon Valley. As it seems to happen, every person we met suggested other people to meet, and Chetan Conikee (thanks again!) suggested I talk to Ben. We met in some hipster coffee place in San Francisco (depicted above, pretty hipster for 2013) and talked for an hour, and for some reason he started inviting me to talk at Strata, which I eventually attended, more than once.
So this December, roughly 6.5 years later, we sat down at our respective desks and had a chat over Skype, following some notes we had put into a Google doc. He had told me beforehand that he always have concerns about this format, because there is no accountability. I hope it turned out fine nevertheless. So what were things we’d be expecting to hear about ML/AI in 2020?
- Reinforcement Learning. RL as a concept has been around for decades now. I actually made my first seminar talk when I was a student about RL, and Richard Sutton was just writing up his book. Deep Learning and AlphaGo and related works have definitely redefined what we know RL can do. I think we now also know much better how to bootstrap RL models through simulation, using human example data, having the model play against itself, and so on. Maybe it is now ready for widespread industry adoption? So far, my impression was that using data efficiently is the biggest issue.
- End-to-end ML platforms. If you believe the marketing we’re already there, but I think we’re still missing a few pieces. Cloud provider companies are working on products, and companies like Uber, Netflix, or AirBnB have been talking publicly about the infrastructure they have built to support ML, and patterns are starting to emerge.
- Machine Learning’s effect on society is pretty real by now, and especially in Europe we’ve started to see some legislation about this, like GDPR. We have very powerful NLP models now that can create realistic text. People have been worried that these will complete change the game around misinformation. What effect will this have on elections in 2020?