Dr. John Bates on AI, Document Intelligence, and Trusted Automation | Ep 1247

95.131 - 113.334 Brian Thomas

So thank you again. And John, if you don't mind, I'm jumping into your first question. You've built an impressive career leading multiple innovative software companies from founding Appama to now serving as a CEO of DOCSIS. What key experiences shaped your journey to where you are today?

113.814 - 132.672 Dr. John Bates

Well, I'm really a product guy. And so I always look at things through a product lens, I guess. I'm a former professor of computer science. I'm a PhD in computer science. And Appama, which was the first company, was when I cut the umbilical cord and went over there to

132.652 - 149.994 Dr. John Bates

from the crossed over to the dark side, if you like, and decided, okay, I'm going to commercialize some of my research, which was a company that used machine learning techniques to find patterns in fast moving data and act on them in sub microsecond latency.

150.475 - 167.715 Dr. John Bates

But so my first experience, I guess, was building everything myself from the product to the company, to the go to market, to then doing crossing the chasm. Boy, that's a lot of effort. So we applied that technology in the area of algorithmic trading.

167.735 - 189.237 Dr. John Bates

So I had a lot of early experience with autonomous agents kind of running around trading automatically and processing millions of events a second and fighting against each other. So I think I've got a little bit of a harbinger for what's going to happen with the agents now that we're rolling out of AI. But I learned really from that that

601.988 - 630.51 Dr. John Bates

So you can get things that aren't actually there in your documents. You can get non-determinism. So your processes will behave differently on... The same process will behave differently on different days, different moments. And you can generate code that can't scale and might run into serious bottlenecks. So, I mean, for me... AI is clearly a massive game changer.

630.53 - 658.507 Dr. John Bates

It's a new industrial revolution, but you've got to manage that trust piece. So there's amazing promise out there. You can ask questions in natural language. You can generate new applications quickly. You can use agents to replace or augment people, approving steps. But you need to add things around this. So you need governance. You need one security model, one authorization model.

658.868 - 686.61 Dr. John Bates

You need determinism. You need to make sure your automation behaves the same way every time with the same input. So you need determinism in automation. You need determinism in search. So you want to make sure that there's no hallucinations in your search. And the way that we do that is with a technique called retrieval augmented generation combined with vectorization so that you can add

686.59 - 708.241 Dr. John Bates

real data from documents back into your search. So you actually zero in on the right information rather than it cooking something up for you. You need a highly scalable system. So you're not going to just vibe code the next generation document management system. And it needs to be future-proof.

708.261 - 727.693 Dr. John Bates

So you need to be able to have what I call composable AI, the ability to map into different large language models, even dynamically. So you might send particular analyses, requests to a different L&M depending on the specialization. And then finally, I like to think of sort of you need a checkpoint Charlie.