Project Linchpin Aims to Be the Keystone for Army Artificial Intelligence
The U.S. Army is marching forward with its vision to build an artificial intelligence (AI) solutions pipeline for its sensing, targeting and intelligence, surveillance and reconnaissance (ISR) environment.
Project Linchpin, rolled out last fall by the Program Executive Office for Intelligence, Electronic Warfare and Surveillance (PEO IEW&S), aims to broaden the use of AI in tactical environments, while ensuring data security, openness and AI modularity with “maximum flexibility,” Army leaders explained.
Young Bang, principal deputy, assistant secretary of the Army, Acquisition, Logistics and Technology, (ASA ALT); Mark Kitz, program executive officer, IEW&S; Alex Miller, senior advisor for Science and Technology, vice chief of staff of Army; and Project Linchpin Program Manager Col. Christopher Anderson, Intelligence Systems and Analytics at PEO IEW&S, briefed the 1,200 industry and Army attendees yesterday during the Program Executive Office for Command, Control and Communications-Tactical’s (PEO C3T’s) Technical Exchange Meeting—TEM-X—in Philadelphia held May 24-25.
The Army leaders expect to award contracts for the AI/machine learning operations pipeline in March or April 2024, after senior leaders provide decisional guidance in November 2023.
“We are going to be the biggest consumers of artificial intelligence and machine learning in the services, given the density of our force and soldiers,” said Bang, a former Army captain and West Point graduate who joined ASA (ALT) Douglas Bush’s office last spring and is an AI specialist, having led many groundbreaking projects across industry.
The groups plans to issue a request for white papers in September or October and before that will hold a Project Linchpin Industry Day in August or September. To prepare for the event, the Army is drafting industry specifications by June or July.
“The initial prototypes that we've done have been on very clean data sets,” Kitz said. “But the data sets that we'll be collecting from are the [various PEO IEW&S] systems, with data that is very ‘dirty.’ And so how do we define an ecosystem, embracing the fact that no one model is going to be able to apply to all these problem sets. In fact, it will be many different models that will have very different performance across the different data sets.”
In addition, the service is in the process of adding office space and personnel to the project, Col. Anderson reported. The officials have seen great support from Army leaders in standing up what will hopefully be a long-term acquisition program. “My problem statement as PM [project manager] is how do you turn that vision into a program of record,” Col Anderson said. “The first challenge was to figure out some requirements. There are no standalone requirements documents for AI and machine learning in the Army.”
“Linchpin came about because we believe that the Army needs to have an ecosystem to serve as data to model developers so that we can have a trusted environment and we can assess the ability for that model to work against our data, and scoped initially towards sensor data for IEW&S,” Kitz added.
How do we define an ecosystem, embracing the fact that no one model is going to be able to apply to all these problem sets. In fact, it will be many different models that will have very different performance across the different data sets.
And although the goal for Project Linchpin is to deliver “trusted AI” within IEW&S, Principal Deputy Bang has a bolder vision that the artificial intelligence pipeline could expand to serve the greater Army.
In addition, the leaders emphasized that for Project Linchpin they are “looking at the acquisition cycle differently,” with Kitz calling it a nontraditional approach to acquisition. They are aiming to create an environment for “continuous competition inside an ecosystem of industry partners.” Additionally, they are trying to incentivize the use of small business solutions and are working closely with ASA (ALT) Bush to include specific language to use small business in the coming request for proposals.
To meet their AI-related objectives, the solicitation will be a phased contracting approach. The initial contract award next spring will most likely be a standalone other transaction authority (OTA), designed to expedite prototypes and target AI-related technology services, products and solutions. After the OTA, the Army is planning to offer follow on contract opportunities.
We are going to be the biggest consumers of artificial intelligence and machine learning in the services.
Miller warned industry they are looking at artificial intelligence models as “disposable” and the service will avoid locking in one industry solution or algorithm. “We hope to ‘rip and replace,’ algorithms,” he said. And despite the very recent inroads of AI solutions such as ChatGPT, Miller cautioned that those type of capabilities are far from what the service needs. “These generative AI models, they generate the exact response that we already know,” he noted. “That's not telling us anything new. They are telling us what is statistically probable to be right. ... And people keep saying that this is going to replace analysts, and maybe one day, it will, when we've actually trained on those 40 terabytes of real data from our real operation. But for right now, it’s still a statistical bundle of words that is spitting out the things that it thinks we want to hear.”
Lastly, Bang encouraged industry to bring specific, innovative solutions. “We all know that the whole space around AI is pretty big, whether it's pipelines, the front end, data management, data wrangling, labeling, training, deployment,” Bang said. “And industry, you all have expertise in a lot of those areas. But if you are a vendor that comes in and says, ‘I can do everything for you,’ I am literally going to kick you out the door. If you are a vendor that says, ‘We can do this piece extremely well and we can actually help accelerate some of these other areas,’ absolutely. That way we can accelerate speed with the enterprise in the Army to really achieve machine learning at scale. That is what we are talking about.”