Sramana Mitra: I’ll just make one point. I agree with you that there is this whole category of invisible AI that is already prevalent in some domain. In cyber security and ad tech, that invisible AI is very much present already. There are a lot of functions that cannot be done without that kind of AI. You can’t really do automated bidding of ads in real time without AI. How the hell do you do it?
John Roese: You’re absolutely right. The third category is, you do it because no amount of human beings thrown at the problem could achieve the outcome. Back to the topic today, that third one is a massive opportunity for innovators and entrepreneurs.
Sramana Mitra: Exactly.
John Roese: In domains where that opportunity exists, it functions the same way. If you understand the domain, then identify an opportunity which can really be impacted by this kind of invisible AI, it’s another level of sophistication that you can introduce into a business.
On a more serious note, given the audience, I think what we just said is very obvious that there are different domains and some of them are visible and some are invisible. The ones that are invisible tend to be massively disruptive and are great entrepreneurial opportunities, but we have this inverse relationship where the most visible ones are attracting the attention of not just the press but also a lot of the technology community.
A lot of people think the skill set needed to play at AI is to learn how to create an Amazon skill in Alexa. That is not the skill set that is going to create a new business for you. The real skill sets are those second and third ones. It’s great to experiment with the likes of Google Voice. It’s the equivalent of learning HTML5. It isn’t going to build the actual product at the end of the day.
What we have to do is invest in learning the frameworks, the algorithms, and how we take advantage of accelerators and infrastructure optimizations. These are harder problems to solve and we just need to give more attention to them because we’re going to need a massive skill shift of people who don’t just understand how to use the higher-order experience pieces but also understand all the way to how do I build a piece of technology that’s in a deeply embedded system that completely allows something like an automobile or an ad tech system to scale orders of magnitude beyond what a human being could possibly do.
It’s harder work but it’s much higher reward and there’s a lot more activity there. Unfortunately, we’ve created a bit of sensationalization around the first category because it’s natural for humans to understand. To me, it’s the least important opportunity. The other two are far more significant.
Sramana Mitra: I fully agree with you. What visibility do you have into the education process of building this high-order skill set? Right now, people who have these skill sets are like football players. They’re getting compensated at incredible levels and they’re rare. The talent war for accessing this level of talent is incredible. How do you expand that pool of trained professionals who can tackle these kinds of problems?
John Roese: Given your focus on One Million by One Million, let me tell you a story. We have a lot of activity to build up our skill set. We build our skill set by not judt training our people but also hiring people. If you go to Silicon Valley and try to find a world-class AI team that can do that second or third category, you’re going to pay a fortune. It’s going to be very difficult to find them. There’s just not enough of them. It’s not thst they’re not being produced.
The top-tier universities have a great program to produce people that have modern expertise in tensor flow and data conditioning but there’s just not a lot of them. Because we’re a global company, we’ve been looking to see if that is just a phenomenon limited to the Stanford’s and the MIT’s or is that something that a Computer Science curriculum at any good university around the world has already begun the journey on as the start to dabble with big data and analytics.
I’m very optimistic. We’re finding talent pools in Russia, Brazil, and Egypt – places that you wouldn’t expect to be hot beds. The one thing that they have in common is that they have a very strong university that has a strong computer science curriculum. That computer science curriculum has already made it through the evolution of thinking about big data and analytics, which is the precursor to what we are talking about.
It’s not to find people who have done this before because there aren’t many of them. It’s the places where there’s a strong computer science culture. Secondly, that computer science culture has already pivoted to start to think about processing data by using non-real time machine tools like Hadoop and Spark. If you have those two conditions, the jump into using TensorFlow or these other next-generation frameworks is very easy.
Thinking about a data-centric model to improve outcomes of data, the only real dimension that has changed is you went from non-real-time to real-time. You’re just using more compute and more acceleration and some new software framework that are a little better at doing this type of work.
I’m very happy with the fact that we have found that there are pockets emerging everywhere. Consider them seeds because we haven’t grown trees yet. It isn’t just centralized on two or three universities. It gives me confidence that given the amount of work that we’re going to have to do, we probably have a workforce that’s forming. We have to cultivate and encourage it.