Let me start out by saying that Li Deng is a genius. This is undeniable. Everyone in the AI world knows this. When he was recruited by Citadel from Microsoft there was a renewed curiosity within the technology industry that Wall Street was finally beginning to wake up to the very real power of AI in its truest form and would begin using AI to solve really interesting problems. At the AI Frontiers conference Li gave a great presentation which I could quote for pages but below are three core challenges that AI specialists have the potential to solve, and in doing so will give whatever firm they do it for a miles long competitive advantage.
On an interesting sidenote, I wonder if Li Deng will be the first of a flood of AI experts to migrate to Wall Street or will he be a unique unicorn? Over the past handful of years many hedge funds claim to be doing AI but they are really engaged in having their existing data scientists apply publicly accessible AI techniques to their data and are NOT doing original R&D. Here are a few things that Li Deng thinks could interest AI scientists:
Very low signal to noise ratio
This is defined as having a very low level of quality signals in relation to the amount of noise gathered
Strong non-stationarity with adversarial nature
In AI there are many examples of data with non-stationarity but few in the AI world where there is also an adversary - in other words, someone else trying to act using the same data and in a competitive manner. I would argue that two other examples would be intelligence work and cyber-security. But his point is well taken. In other words, Google is the only “player” in its system.
Heterogeneity of big data
The sheer amount of diverse data available, combined with the earlier notion of low signal to noise ratio, is an exciting hurdle to overcome.
Again, it will be interesting to see if Wall Street firms will begin reaching out to the leaders in AI. Up to this point they largely have not.