The Military Is Doing New Things With AI

It’s probably no surprise to people who understand the ramifications of new AI tools that Uncle Sam is quickly working to integrate these into military operations.

But what does this look like on the ground?

A few days ago, I wrote about a presentation by Colonel Tucker Hamilton about what warfighters need in terms of technology, and where we discussed the landmark of an AI running a military drone for the first time.

I especially liked his phrase “battleground of things.”

Today there’s more from an interview that I did with Sean Batir, CTO of Maven, a company that’s helping the military to deploy new approaches to artificial intelligence.

Three Insights

During the interview, Batir broke down some of the principles that experts are using in innovating in these ways.

One is the enhancement of accuracy:

“Essentially, many of our models now are able to actually meet or exceed human detection, classification and (standards for) performance,” he said.

Another has to do with interconnectedness and systems integration.

“No system, kinetic or non-kinetic, is an island,” he said. “In other words, different types of systems are typically horizontally integrated, and not always so vertically integrated. So interoperability of interfaces to share data, models, targets and detections – that has improved, because now we’re actually having more stable and standardized methods to pass that data between different types of modern systems and infrastructure.”

The third is something you might call “location visibility.”

From Batir: “We’re increasing the speed of being able to locate these types of dynamic and moving objects on the battlefield. … I think what we’re observing right now is that as AI capabilities are going to be integrated into various military flows, these types of triggers are actually getting integrated into how imagery and video are exploited. And essentially, if you have to think about military operations, the reality of it is like, if you can’t see it, you can’t perform an operation. And so we’re noticing that the adversaries are now able to also discern an object from a non object, which is also accelerating their ability to, what we would term it as, execute the speed of conflict.”

MIT to DoD

Later, Batir talked about that cultural change, of going from a place like MIT, into the Pentagon, where a vastly different culture holds sway.

Still, he suggested, some things carry over.

“That critical thinking that you learn in undergrad, I think, is so translatable,” he said.

Culture of Comms

I asked Batir what it’s like to be in meetings at the Pentagon, what his drive is like, what his day looks like. He said phones are not allowed in meetings, to promote attentiveness.

“(The fact that) you’re forced to be super-present probably has its advantages, given the fact that if you’re making, you know, national-level decisions, you want everyone to be fully present and (feel) like they are with you at all times,” he said.

Central Command

Here’s a through line from the interview to something that I’ve been hearing about a lot, and that I think just makes intuitive sense when it comes to technology.

In a way, it goes back to that term: “battleground of things” or more generally, “IaaS?”

In a smart system, connecting elements is key.

“The idea is that our data and our detections are fed into other platforms,” Batir said, describing Maven solutions. “Those other platforms are not always our own agencies. They belong to those other groups that you just mentioned. The idea is that through our program, we’re actually able to build a sort of a tapestry of connected sensors from all those branches of both the armed forces and the intelligence agencies that participate in this, into a unified AI network. And so normally, a lot of the systems are Balkanized because of their history, of how they’ve evolved. And so our goal, right, is that we’re kind of basically bringing those together.”

When asked whether the firm focuses on computer vision or LLMs, Batir said Maven does both, in a sort of “multi-model” approach.

As for risk:

On one hand, Batir pointed out, it’s not always beneficial to wait for perfection.

“Our philosophy is that perfection is the enemy of progress,” he said. “And so, as long as we can ensure that the risks are very upfront, that we are clear about metrics, we don’t want to prevent people from being able to use a capability that’s available, because it’s not 100%.”

However, there are other standards for mission-critical systems.

“In terms of real-world situations where lives are at stake, there is an entire risk acceptance framework,” Batir added. “That’s when people outside of maybe the pure technology space come in. We have things called, you know, disclosure officers. We have legal review. There is still a very critical manual process, especially when you’re talking about potential loss of life. And so there is a process that exists to bring in that element of careful adjudication.”

I thought all of this illuminates what’s going on in DoD right now, as we take part in the fruits of AI. Watch the video for questions from the audience about Ukraine and other concerns.