Data is the Hard Part | Building AI Tools for the Military

Bharat Patel (0:00)

Jordan, data, data, data. In my opinion, data will never be AI ready. Data will always have its weirdness about it. So what you see today in Ukraine is not how it started, what was being recorded as video. That data was getting back to a place, and then people started to realize, okay, I could start making things more autonomous. The way that we do our exercises may not be representative of what the next fight looks like. The enemy's gonna have a vote, the environment's gonna have a vote, the warfighter that's using the model is gonna have a vote.

Jordan Schneider (0:31)

The 4D chess of bureaucracy.

Bharat Patel (0:34)

Yeah, you have to be able to navigate the Pentagon. So the first thing that I would do is, hey, if it's a high priority, make sure my highest speed individuals, humans, are part of that acquisition team, and they will move mountains.

Jordan Schneider (0:49)

Bharat Patel currently Accenture's AI and data lead for its defense portfolio. A former enlisted Navy who had arcs in Mitre as well as various corners of the Army. We're here today to talk about why data is the hard part in making AI work to win wars for you, what standards integration process and governance you need to turn, all fun tech and cute little language models into things that can actually target things and blow stuff up.

We're going to talk about the sort of, I don't know, dark side of data as well, what poisoning can happen and how the enemy has a vote in you creating your datasets to find things on the battlefield, as well as what life is like as a civil servant in acquisitions, and how this administration is currently reshaping that. Bharat, welcome to the roof and welcome to ChinaTalk.

Bharat Patel (1:49)

Oh man, I love this place. I haven't, I love this place and I haven't been back into New York City in quite some time and holy crap, like last night I was walking around and it is just beautiful. I miss this, thank you.

Jordan Schneider (2:02)

Well, thank you. I mean, it's a real shame we closed down the nearest fort to New York City, otherwise we'd maybe get you up a little more often. Um, Bharat, why is data the hard part?

Bharat Patel (2:15)

Oh, Jordan, data, data, data. Let me, let me, so we talk about AI ready data.

In my opinion, data will never be AI ready. Data will always have its weirdness about it. And it's more about what the AI use case is, because the AI use case will drive the type of data that you need and the type of quality you need that data in. So, we often just focus sometimes on the wrong problem. So, just figuring out what AI use cases you want to go after will then help us determine, again, the data quality and going after that.

Jordan Schneider (2:54)

All right, great. So, why don't you pick a case study and walk us through how data can be used and fit for purpose? Right.

Bharat Patel (3:01)

So, I've had a ton of experience across the Department of War in different AI projects. One of the most famous ones was Project Maven, started in 2016

I would say that one of the first things that we all struggled with was with data itself. So, it was trying to understand building computer vision models against a particular set of targets, and then it's trying to get enough of that data that has the actual targets in it, that we needed to actually explore. So, at the first start of Project Maven, we were getting a ton of data that, again, didn't have relevant targets, so then your models weren't as performant once we put those models into operations. So, then we had to quickly learn that, oh, no kidding, we have to continuously collect data and relevant data, so then we could build some of those models. So, that was one of my first early inroads of kind of understanding the use case tied to the data, and then getting enough of it to build performant models.

Jordan Schneider (4:04)

How do you get an organization to start taking data seriously, and what is the infrastructure? Let's start with the cultural thing.

How do you spread the gospel on data first?

Bharat Patel (4:13)

How do you spread the gospel on data? First, you got to find your battle buddies that understand some of the gospel, and once you find your battle buddies that are spread across different parts of the organization, you have to kind of create that culture to bring them up, and they will advocate to leadership that this is important. But then you could only do that by evidence.