The Deparment of the Air Force is planning to give Alphabet‘s [GOOG] Google subsidiary a first hand view of the department’s operations under a new program, U.S. Air Force Chief of Staff Gen. Charles Q. Brown said on May 9.
“One of the things that we’re just on the cusp of is bringing industry into the Air Force,” Brown told the Ash Carter Exchange on Innovation and National Security in Washington, D.C. “We have a program of education with industry where we’ll send our officers and NCOs [non-commissioned officers] to go spend a year with industry, and so I’ve been thinking, ‘Well, how do we bring industry into the Department of the Air Force so they see how we work?’ So, we’re working with Google, and we’re real close. The lawyers have to do the last few checks, but the goal is to bring someone in for three to six months so they can see how we operate. We want to make sure we put them in a place that doesn’t crush their soul, where they can see the real value and opportunities and then take that back, but also build a relationship wth those of us in uniform and our civilians that can be long-lasting because we can get things done.'”
“A lot of things get done based on relationships, including figuring out how to get around the process and the policies,” Brown said. “Instead of having an exception to the policy, sometimes I think, ‘Just change the policy so the exception becomes the new rule.’ That’s the way we’ve gotta operate in the future to make sure we can move as fast as we need to.”
Google is one of four awardees under DoD’s $9 billion Joint Warfighter Cloud Capability (JWCC) effort and was the prime contractor for Project Maven before the company dropped out in 2018 after receiving pushback from employees about the company’s tools being used for an artificial intelligence (AI) drone imaging effort. (Defense Daily, Dec. 7, 2022).
The other JWCC awardees are Oracle [ORCL], Amazon Web Services [AMZN] and Microsoft [MSFT].
The Pentagon established JWCC in July 2021 to replace the JEDI cloud program, which was shuttered after the department said it determined that “due to evolving requirements, increased cloud conversancy, and industry advances” the program no longer met its needs (Defense Daily, July 6 2021).
Microsoft originally beat out Amazon Web Services for the JEDI cloud contract in October 2019, but the program stalled following a series of legal challenges.
The Special Competitive Studies Project (SCSP), an Arlington, Va.-based foundation established by former Google CEO and billionaire Eric Schmidt in 2021, held the May 9th Ash Carter Exchange. SCSP aims to forge a national consensus to strengthen U.S. capabilities in AI and other emerging technologies.
“The Trump and Biden adminitrations did a good job in slowing down China,” Schmidt said at the May 9th event on the topic of AI. “At the moment, the leadership in this area is occurring in the United States. The Europeans are hopelessly behind because they start with regulation, which never works so, effectively, it’s ours to play with. We as a nation have control over this technology right now.”
Google, OpenAI, Meta Platforms, Inc. [META], and San Francisco-based Anthropic are the major U.S. builders of powerful AI language models.
During his remarks, Schmidt referenced a recently leaked memo from a Google engineer that suggested that Google and OpenAI’s fence around AI is falling, as the memo said that the open source community had gotten a hold of LLaMA, a large language model developed by Meta, and had used LLaMA to move toward the rapid creation of smaller, low-cost Bard and ChatGPT clones.
OpenAI developed ChatGPT, and in February Google said that it was rolling out its Bard AI chatbot technology as a ChatGPT competitor.
Regarding AI, Schmidt said on May 9th that he is most concerned that the “diffusion of this knowledge from these very expensive and specialized models is occurring much quicker than I thought, and open source has the property that it moves us along very quickly, but it gives all of our advantages away to our competitors.”
“Why in the world would you invent something this powerful and then choose to leak it to the Chinese, right?” Schmidt asked. “Not even I am that stupid, and yet, in fact, that’s what we’re busy doing, and this is a problem that as a nation we have to address this year.”
Industry may propose a rule that requires companies to notify the federal government if they are developing language models able to improve their functionality autonomously. “I think the industry is going to propose roughly the following rule: If you’re doing something of this nature, if you’re doing something which involves either recursive self-development, which is obviously dangerous, or some kind of fine tuning in these critical areas, you have to nofity the government,” Schmidt said.
Brown, for his part, suggested that AI and machine learning (ML) are not a military cure all.
“I don’t know that we fully understand AI and machine learning,” he said. “I think we’ve made some progress in certain areas as an Air Force just based on our being, in some cases, a more technological service than others, but also based on our relationship with the Space Force. But I think too often, for those of us that work at the Pentagon, you’ll see AI and ML on a bunch of Power Point slides, as if it’s the panacea that’s gonna solve all our problems, and we don’t fully understand it.”