Hadean builds large language model for British Army virtual training site

A UK deep technology start-up, Hadean, has secured a contract with the Defense and Security Accelerator (DASA), an agency within the UK Ministry of Defense (MoD), to populate the UK Army’s virtual training space.

The company already has relationships with the Army, having built a cloud-based training platform under the Army’s Collective Training and Transformation program.

Having validated a cloud-delivered collective training platform, the Army now wants a more detailed virtual environment. So, for the next year Hadean will create a large language model that meets the requirements described in Call of the Ministry of Foreign Affairs for a “complex, representative human terrain”.

One that will allow the troops to swallow a “free thinking A3E [Audiences, Actors, Adversaries and enemies] ability [that] provides cues, stressors and frictions in the human, physical, environmental and information domains’.

The leading information consulting firm, GlobalDatafound that this artificial intelligence (AI) production tool has grown faster than any other technology, as the company expects it to “disrupt and transform businesses across sectors with lasting impact.”

In an exclusive interview, Hadean’s vice president of innovation, Chris Arthurs, spoke Army Technology to describe how the company plans to apply its own AI production product to the military’s virtual training space.

Get access to the most comprehensive Company Profiles on the market, powered by GlobalData. Save hours of research. Gain a competitive edge.

company profile unit

Company Profile – free sample

Your download email will arrive shortly

We are confident in the unique quality of our company profiles. However, we want you to make the most beneficial decision for your business, so we offer a free sample that you can download by submitting the form below

From GlobalData

This is an application, he noted, “that certainly has not been attempted with this technology before, and I would argue that it has never been achieved before.”

Chris Arthurs Hadean in text image
Chris Arthurs, vice president of innovation at Hadean. Credit: Hadean.

Using genetic artificial intelligence to populate a virtual environment

“The central position is that historically – and this has been a problem for a while, it’s been discussed before – virtual learning environments started [off] to be very empty,” remarked Arthurs.

“So they have a Red Force and a Blue Force, maybe some buildings, but then no people, civilians, sheep, cattle, birds or all those things that animate an environment.”

Powered by a large language model at its core, Hadean will use artificial intelligence to understand the evolving context during a live, virtual, immersive training exercise.

This enables the simulated A3ES to dynamically respond to the details of learner actions. These in turn will generate reactions and prompt educational staff to adapt their approach and tactics, effectively bridging the physical and virtual worlds through real-time interactions.

“One of the things we do using language models is create a social media feed, on something like X or Twitter. This is something we have the capability for in the Hadean product, and this is built using logical language models, but we want to improve it.

“That’s one of the key things as part of the project: the models are great at generating a small tweet-like snippet of text and being able to respond directly to things happening in the learning environment as it’s happening.”

Hadean’s social tool provides an updated representation of the “pattern of life entities” beyond the detailed parts of life during a conflict. It also provides a useful – and due to the nature of social media, perhaps intentionally unhelpful – information base that will, in effect, inform military operations in real time.

Arthurs went on to describe such a scenario: “So events that might happen, radio statements from soldiers about what they’ve seen can be fed into a language model.

“One can say that he is pretending to be a citizen who observes these things, but he cannot hear the radio messages, which issue tweets about it. So the soldiers say, “I just saw four red team tanks driving down this road, on the radio,” and then that feeds in and another source might start saying, “Oh, no, I’ve seen those tanks go down Central Road.'”

Delivery of three troops discussing logistics. Credit: Hadean.

Getting the right AI

Combined with AI-powered data exploitation capabilities, the platform will also create an After Action Review dashboard to help commanders identify weaknesses and improve future performance.

Although, managing the AI ​​tool to detect these weak points occasionally poses challenges:

“So if you run the language model twice against the same data, it can sometimes tell you exactly what went wrong in the function, and sometimes miss it completely. Thus, getting this right involves coordinating incentives for encouragement [the AI tool] to look for different types of failures.”

However, Arthurs went on to suggest a shift in perspective in terms of managing our own understanding of AI performance.

“Then you have to do the evaluation phase to prove that you are statistically likely to get the right answers. But this test evaluation part, it’s not going to be binary, yeah, it works. No, it doesn’t work like it could with a piece of hardware, for example.

“It will be more likely [that] we can configure it to work 99.9% of the time, but we understand and should all understand that it won’t work 0.1% of the time, and you have to figure out where that limit is going to be.

“Our brains are not wired enough as a society to keep track of the full statistics, they will come in individual stories.”

Chris Arthurs, vice president of innovation at Hadean.

“Whenever an autonomous vehicle crashes or runs into someone or makes a mistake and falls into the back of the truck, which no human would have ever done. Because it would say there’s a truck there, I’m not going to ride in the back. But the thing is, whether he does or not is of very little interest to me personally.

“What I want to know is whether it’s statistically fair enough to not crash where a human wouldn’t. Statistically, how many accidents were not avoided that would be caused by humans? But our brains aren’t wired enough as a society to keep track of the full statistics, they’ll come in individual stories.”

Read the original at Defence247.gr

Related Posts