Background
It is hard to pick up a magazine or newspaper today without seeing something about the amazing things artificial intelligence and/or machine learning (AI/ML) are doing to change our lives for the better. Most people enjoy the benefits of talking to our computers, cars, and home specialty devices like Google Home and Amazon Alexa but don’t think about or care that these technologies are enabled by natural language processing (NLP), one of the today’s most advanced forms of AI/ML. It is even possible today to get real time language translation earbuds to help us more easily explore visits to foreign countries, bringing the Star Trek universal translator that much closer to reality (https://www.startrek.com/database_article/universal-translator).
While these commercial uses for AI/ML are enriching our lives, the potential military use of AI/ML has ushered in an “unspoken” new cold war between China and the United States. Although AI rapid development and deployment initiatives are being supported and coordinated at the highest levels of Navy, the traditional Naval research & development (R&D) model is highly unlikely to pace the Chinese Navy’s rate of AI adoption, given China’s national drive for AI superiority. Something different must be done to bring Naval AI deployment timelines closer to the nuclear ballistic missile submarine deployment rates achieved in the post WWII years!
Artificial Intelligence (AI) and Machine Learning (ML) research has been funded at Defense Advanced Research Projects Agency (DARPA) for over thirty years, however, it wasn’t until the last few years that AI emerged to global commercial prominence in this big data era. To outpace the growth of AI in the U.S., in 2017 China’s central government announced ambitious AI plans to keep pace with the U.S. by 2020, achieve AI breakthroughs by 2025, and be the world leader in AI by 2030. Author Hemant Taneja, in his new book Unscaled, makes the case that, “Artificial Intelligence is this century’s electricity,” because as he states, “AI is powering all sorts of profound new technologies and companies that will change work and life.”
AI for Naval Warfare
Since the dawn of warfare, superior technology has provided a significant edge to military powers and consistently been used as strategic leverage for U.S. National Security. AI is now being considered one of DoD’s future advantages, similar to the advantage stealth technology provided the U.S. in the 1980’s. However, unlike stealth technology where the U.S. held a commanding lead, DoD may not be able to leverage AI technology faster than China. For DoD and particularly the Navy, the question becomes, what is the fastest way to deliver effective AI into Navy operations.
Current Navy AI development activities are following the traditional technology development model which includes early development work by the Office of Naval Research (ONR) with transition to Program of Record (POR) Program Management (PM) organizations for system integration, certification, deployment, and sustainment within the Navy’s Systems Commands. These AI efforts are focused on man-on-the-loop weapon coordination & control, man-in-the-loop command & control (C2) and Intelligence, Surveillance, Reconnaissance (ISR) planning and decision aids, and a Development Security Operations (DevSecOps) environment to support AI development and training infrastructure.
While these new AI initiatives are being worked to shorten the transition from ONR to POR deployment, they are still operating with traditional acquisition constraints. ONR constraints include: no or minimal use of licensed software; open source software if the licenses are DoD friendly; and, the research must be ONR-hard, meaning developing new AI capabilities not available in the commercial market. POR’s on the other side of the technology divide, prefer subscription or licensed open source products such as RedHat Linux to help maintain and leverage commercial software advances. In addition, new POR capabilities must be integrated, Information Assurance (IA) certified, and weapons certified where applicable.
Although these rapid AI development and deployment initiatives are being supported and coordinated at the highest levels of Navy, the traditional Naval development model is highly unlikely to pace the Chinese rate of AI adoption given their national drive for AI superiority. If the Navy is going to prevail, AI development and deployment timelines must become similar to the sub-decade nuclear ballistic missile submarine deployment timelines that were achieved in the post WWII years!
Related Commercial AI Activity
In his new best-selling book, AI Superpowers: China, Silicon Valley, and the New World Order, Kai-Fu Lee (former U.S. citizen, now a resident of Taiwan) provides a detailed overview of the AI relationship between the U.S. and China, based upon his background as an AI speech recognition expert, leader of Microsoft’s Chinese research division, and later head of Google’s China division. While Lee does not comment on the military implications of AI, he does break down his estimates on where China and the U.S. are with respect to the four major waves of AI. In his book AI Super-Powers, Kai-Fu Lee compares AI development between the two countries, but argues for cooperation to avoid the cold war winner-take-all mentality.
Lee characterizes AI development areas as:
- Internet AI – AI powered across the internet such as Google Maps.
- Business AI – AI used to improve business knowledge, processes, and competition.
- Perception AI – AI used to track people, patterns, and perceive the world around us.
- Autonomous AI – AI that enables fully autonomous self-driving cars, factory robots, and delivery drones.
As shown in this chart, Lee predicts that China is currently at parity or slightly ahead of the U.S. in Internet AI and Perception AI with the U.S. holding a significant lead over China in Business AI and Autonomous AI. His five-year out estimate, however, shows China leading or at parity with the U.S. in all but business AI. No matter how we look at this, with respect to the military adoption of AI, China is an AI competitor.
When reviewing the current U.S. commercial work with several big companies racing for AI dominance, one finds many commercial proprietary AI platforms where various industries are working to adopt the value of AI. These commercial AI platforms are focused in different directions for different functional areas, but all have a vibrant circle of developers and researchers working to improve their products and capabilities as shown in this Predictive Analytics Today article on the Top 15 Artificial Intelligence Platforms. Topping the list of AI platforms are offerings from well know companies like Amazon, Google, Microsoft, and IBM. In addition, several young AI platform companies are focused on different AI markets such as Infosys Nia, Wipro HOLMES, API.ai, Rainbird, Ayasdi, Mindmeld, KAI, Receptivity, and Meya. Open-source AI libraries are also available, some being used by ONR project managers to support their AI projects.
Adopting and Adapting
Because the Navy’s AI ONR development activities are assuming commercial licensed AI platforms are generally not useful starting points for these efforts, Navy is failing to fully explore the possible adoption and/or adaption of commercial AI R&D. Given the yearly multi-billion-dollar U.S. investment in AI, the comparatively small amount of Navy AI investment will be hard pressed to outlearn or outperform these commercial activities.
This leaves the unanswered question of whether evolving commercial AI platforms, and the associated startup developments, have anything to offer Navy’s AI efforts, or whether useful commercial AI could be more quickly adapted or adopted for Naval use. ONR’s traditional technology development model could put Navy significantly behind the Chinese Navy AI development given China’s strategic practice of cross connecting industry, finance, and military activities to gain greater advantage.
A viable alternative could be to structure a Navy AI program with features similar to the integrated DARPA/Navy/Air Force Long Range Anti-Ship Missile (LRASM) project. After demonstrating successful capability in the four years between 2009-2013, LRASM was authorized for limited Navy production in February 2014 as an urgent capability stop-gap. By integrating the DARPA and Navy NAVAIR team from the beginning this project eliminated all valley of death transition issues while taking advantage of Other Transaction Authority (OTA) for rapid contracting.
However, unlike LRASM, which is primarily military technology where little commercial technology can be leveraged, AI to support C2, ISR, and weapon control, can potentially be leveraged or repurposed from the many commercial AI platforms and startup activities that are competing with China’s commercial companies. Because AI is an urgent Navy need, it could take advantage of the Congressionally authorized Section 804, Middle-Tier Acquisition process and Other Transaction Agreements, to explore rapid AI capabilities in short six-month rapid turnaround prototype phases.
Such a Middle-Tier Acquisition could be structured with a small select government leadership team working together with industry led OTA teams to explore all possible unconstrained ways to adapt, adopt, repurpose, or develop effective AI warfare functionality. By establishing three or more teams with short phased Statement of Work (SOW) goals, teams could be continued, turned-off, or new teams added in six-month cycles. Where useful prototype functionality is demonstrated, follow-on agreements could be awarded to push the capabilities into POR production as appropriate.
Funding such an effort could be accomplished using Middle Tier Section 804 authorization and critical Fleet requirements. This effort would not require the disruption or elimination of the current ONR/POR project efforts, but rather, provide an alternate skunk-works activity applied to this emerging Naval warfare need.
Part 2 – will discuss the challenges of black box AI, explainable AI, AI spoofing, and brittle AI solutions. All are AI challenges that must also be addressed if AI is going to successfully augment Navy Command & Control!
AI talent today is in high demand, with newly trained engineers earning well over $100k per year. The only reasonable way for the Navy to enlist such talent is with some well-funded efforts to achieve some exciting results. The Navy, DoD, and USG should invest in a well-funded glamorous center of excellence aimed at making intelligent systems ubiquitous throughout the government. Employees will be attracted to the good causes, and they wouldn’t need to sacrifice to help the nation.
My $.02
Good idea Rick… I am being told an AI coder with a Masters and some experience is getting $500K
Marv,
Another fine analysis and template for change. Is anyone listening? Keep on nudging and carrying Diogenes lamp!
Thanks for the historical reference… even more interesting is Victor Davis Hanson’s writings on how ancient wars have valuable lessons for us modern people… among his thoughts are that wars are test laboratories for the often erroneous beliefs rulers use to carry their constituents into war!! But I have been around long enough to be a Diogenes cynic because bureaucracies rarely change from within… hence the hope we all place in our shared power system of governance…
Industry leaders such as yourself have been recommending acquisition reform since I entered the defense workforce in 1982. When reforms are enacted, they miss the mark or are misused such as LPTA. Most recently, we’ve watched our dawdling and complicated acquisition system fail to keep pace with IT innovations or manage end-of-life which would reduce system costs and significantly improve security. A specific program could be established using OTAs to speed development of AI for defense uses. A broader problems exists whereby program managers are discouraged from using OTAs as they don’t fund local contracting offices. And what about solving development delays for other technologies? I believe that most of us in the industry (military, civil service, and industry) want to do the right thing. However, can we even agree on what should be done given the myriad, and often arcane, interests? Can we start at the top and get our elected officials out of the loop of deciding what our military buys?
The US Navy has no chance in a war against China. The reasons are manifest and none are easy to fix: systemic lethargy of US DoD procurement process, inability to recruit and retain top engineering and scientific talent, dysfunctional policy-making and governance, incompetent leadership. The list could go on. We will lose not because China is smarter or better, but because our DoD’s incompetence renders any conflict into a contest of size. And China has more people and soon will have more capital.
The way out is a radical transformation of how we think about A.I. for the military. We must stop buying technologies that follow the pattern of the F-35 fiasco and return to the models that delivered the SR-71, USS Merrimac, and 30.06 bullet. That is shorter specifications, more numerous suppliers, and less restrictions on who can participate in the DoD procurement process. The current system is so burdensome that many of the best high-tech companies simply refuse to sell to the US Government. Companies like Intel, Google, Facebook, and even startups like mine, cannot be bothered to deal with the endless bullshit from government employees whom we would never hire, much less want to depend upon as customers.
The only way for the DoD to change might be to lose a war. It’s hard to imagine anything less would inspire the self-awareness and radical change necessary to compete in the world that A.I. is now creating.
All good points from a new company that applied AI to help other companies win business faster… Thanks Olin
I can relate to the frustration Olin, but there is a third way, which we are pursuing– off-the-shelf systems. It is much more resource intensive in every respect — talent, capital, computing, but I’ve concluded from our systems perspective that it’s the only way to deal with the government in a non-fatal manner. Otherwise, the bureaucracy will copy or steal anything they can’t block or kill. I’ve personally tested over decades with several companies. Waiting for the U.S. to lose a major war will be too late in the current and foreseeable future as the winner would likely be global totalitarianism, meaning most of us would disappear in short order. We can’t allow the bureaucracies to destroy the U.S. The Constitution is too rare and brilliant even if the institutions have become corrupt. We must overcome these problems. Sometimes losing isn’t an option, whether internal or external adversaries, or some combination thereof.
Good points Mark… I agree that we can’t afford to lose a war to improve the system… In today’s world, peer competitor wars will likely be very short with a high probability that they escalate to the unacceptable!!
The Constitution is not rare. Many countries have brilliant constitutions (just read the Netherlands for example). The US is exceptional because our culture fosters creative destruction. Yes, the genesis is our adherence to an 18th-century political dogma. But we now live in a world filled with more than 90 democracies where technology evolves globally. Off-the-shelf is a laughable strategy if the top producers refuse to play by the DoD rules. And yes, we are at risk of losing the next war precisely because we have funded DoD bureaucracy to the point where it no longer pays a price for its own stupidity. Even the hawkist Heritage Foundation (well known for fawning over the Constitution) gives the military poor grades on readiness. See: https://www.businessinsider.com/new-study-gives-us-military-c-for-strength-and-readiness-2019-10?nr_email_referer=1&utm_source=Sailthru&utm_medium=email&utm_content=BIPrime_Strategy&utm_campaign=BI%20Prime%20-%20Strategy%202019-11-01&utm_term=BI%20Prime%20-%20Strategy%20Insider
Great points Olin but I don’t see the DoD changing anytime soon…
Hey Marv
This is a very timely blog. I’m fighting like he_ _ to change the culture here at our Agency and we’re making some headway but, it’s not fast enough. I think we’ll get there but, again, when you’re dealing with a workforce that’s been FAR based for decades, it’s hard to change their habits. It’d be nice to have a commercial company that specializes in this new way of thinking with regards to turning around contracts. We’re working on a 60 working/90 calendar day turn I(or shorter) on getting vendors on contract. We’re about half way there. So, it’s getting better but, as I stated above, we have a ways to go to get over some of those FAR based processes/mentality. As usual, you’re spot on. Keep it coming. I plan to share this blog with Seniors across the Agency. Thx Rik-i
Rik-i, thanks for the good words and I applaud you working it… We need good leaders like yourself to make the difference… It doesn’t happen on it’s own:-)
With the best U.S. AI application developers eschewing military work, whether for commercial or personal reasons, and the best Chinese AI developers suffering lesser constraints, how can the U.S. Navy maintain AI superiority, irrespective of whether the applications are based on ONR or licensed commercial technology.
Another challenge may be the level of willingness of US commercial AI companies to partner with the DoD.
Great point Michael, Silicon Valley seems to be setting new rules for helping National Security…
Hi Marv,
Good effort. It’s not just the USN, but essentially all of the older, post war research organizations. We’ve engaged with most in an attempt to accelerate R&D, but we run into stalling, blocking and attempts to redirect. Unlike any other technology, it is the very adoption of properly designed AI systems that starts the voyage. I recently presented first public talk on our 23 years of T&D in AI systems at a NM conference, so we recorded. If you haven’t viewed yet, take a look when you have 47 minutes –www.kyield.com.
You make some very good points and also provide the “artifacts” of why you are making your statements. There aren’t many that do that! Great Work!!! In working with ONR and PORs for many years, the constraints that not only ONR but all that propose solutions to PORs are very much out of line with how industry and the all others outside the Navy or DoD in order for the Navy to stay ahead of China or any other adversary. As you know the PORs aren’t willing to take the risk of taking ONR “code” due to many factors you have previously stated in your Blog. Your idea of 6-month OTA Teams is something I wish the Navy’s senior leadership would implement! Your statement about DevSecOps is another aspect of all of this that needs to be looked at seriously. Many talk about it but “many” need an education of what it means. Thank You for saying what many of us think!!!
Thanks Ref, great comments from a person that knows what you are talking about from the operational trenches:-)
Until we stop training Chinese students in AI at the best US universities so they can go back to China to compete direclty against us, we will never be able to compete and win decisively.
Interesting point Andy… Perhaps there should be an education equivalent to ITAR on technology exporting:-)
Marv, an excellent blog, and I am in complete agreement with the points you make. The challenges are many to create supported teams to adapt, adopt, repurpose, or develop effective AI warfare functionality using the best AI technolgies from around the globe. A starting point, were in not for the ill-timed abolishment of the NRAC, could be tasking a Navy Science Advisory Board to create multiple teams to explore different OTA approaches for creating paths to achieve greater use of AI to ensure the preservation of naval dominance. Doing this would be a very low cost way to get this initiative started.
Thanks Bill, it is interesting that in this critical time with new peer competitors in National Security that the wisdom of the senior leaders is that they don’t need any wisdom from historic organizations like the Congressionally authorized Navy Research Advisory Committee (NRAC)…
Marv the problem we’re having with IA is the same problem we’re having with quantum computing and 5G and precision manufacturing and everything else that requires a high level of expertise -> lack of funding for strategic technology development which allows China to buy into new ideas before the DoD even understands them (by which time its too late). Sadly the last Administration that clearly saw the value of a strategic technology investment was President Reagan. America is now suffering from decades of underinvestment.
Being a world leader takes real work. We need to invest in people to develop the new technologies which in turn can create the solutions companies can sell to the DoD. Moreover we need to re-establish the NATO coalition around technology development and just not weapons systems deployment.
Congress, the White House and the Penatagon have to realize there’s no “easy” button to fight China.
Good points Junaid… I think our Nation is suffering on all fronts from thinking that everything has an easy button…
Marv
– Appreciate the article and the book-of-the-month recommendation.
– For arguments sake, I’d like to bar-napkin a different approach to the AI dialog. I’ll buy the first round.
– I’m a big fan of AI/ML and automation. But AI won’t fix stupid. It may even scale stupid, automate it and help it achieve new levels of precision (e.g., precisely wrong).
– My favorite example of this is Microsoft’s Chatbot Tay. (https://www.theverge.com/2016/3/24/11297050/tay-microsoft-chatbot-racist)
– I once took a class called Ignorance Management. It was an niche sub-set of Knowledge Management focused identifying the “unknowns” that were the root cause of dumb organizational SOPs.
– It might be useful to revisit Middle Tier as a case-study in Ignorance Management and IM as a lens to facilitate the adoption of AI operationally.
– It might be that the POR and ACQ constructs are a “precisely wrong” approach to funding AI in support of military capability. We may find that it is not even possible to define AI as a system or in terms of pre-defined system requirements.
– My recent engagements with major industry partners working defensive cyberops tend to imply this. They are focused on network “behaviors”. In the case of AI, they are using it (e.g., algos) to identifying ephemeral, “emergent behaviors”, categorizing them as acceptable or unacceptable and specifying response options (digital, physical and carbon).
– That sounds a whole lot more like a Top Gun approach to Readiness than it does as a requirements generator for a POR.
– Why not one for Mathletes? How would we fund it? Maybe we could look to the Flight Hour Program as an example of a multi-billion dollar funding stream that accommodates new operational TTPs, training and desirable outcomes that are not “Acquisition”.
– We can then concurrently pursue a systems approach to ML adoption via middle tier and traditional ACQ processes and their deliverables.
Thanks Dan, you are on for the bar-napkin session. Good comments and we are aligned in the idea that this has to be moved faster and better…
Thanks Dan, always thoughtful comments and you are on for the bar napkin discussion:-)