What nighttime fighting can teach us about using AI in companies.
(I wrote this piece with my friend and colleague Richard Dunks. He is the founder of Datapolitan, a data science consultancy that empowers public sector professionals to make data meaningful and actionable. He is also a US Army veteran and decorated military intelligence professional.)
In war, being able to see when your enemy cannot is an advantage. The Army has coined the phrase "to own the night" to express when only one side's combatants can effectively operate under the cover of darkness.
The Viet Minh held this title during the Viet Nam war, in the 1960s and 1970s, being able to move effectively under the cover of darkness to wage their insurgency against a militarily superior force. Fast-forward to the early twenty-first century, and thanks to gear such as night-vision goggles (NVGs), laser-sight designators, and other advanced optical technologies, US forces have a clear edge over comparatively ill-equipped fighters in Iraq and Afghanistan who are unable to shoot, move, or communicate in the dark.
The authors of a recent Modern War Institute piece, however, suggest this perceived edge isn't what many think it is. They point to the indiscriminate and uncoordinated use of these technologies as if they were simple augmentations that allowed fighting at night to be just like fighting during the day. For them, really owning the night means sorting out how night missions differ from daytime missions, beyond just the lack of plentiful ambient light. It means getting groups of combatants to coordinate moving around in the dark in ways that adapt the technologies to the needs of warfighting, minimizing risk while enhancing operations and improving the likelihood of mission success. It means developing improved ways to discern friend from foe, to coordinate with pilots who fly support missions, and more generally to approach nighttime fighting as its own experience separate from daytime operations.
In short, truly "owning the night" means weaving the use of night-vision techniques into practice across the organization, via doctrine, such that troops can hone tactics, techniques, and procedures (TTP) around what it means to operate in the dark. To achieve mastery of this subject matter is to know what is special about that environment and developing special mission types just for it, rather than just opting always to fight at night because of this perceived advantage.
Based on our experience in the data space, we can tell a similar story about how companies employ ML/AI. In the same way that having night vision gear doesn't automatically grant you nighttime superiority, merely hiring data scientists doesn't guarantee that your business will suddenly become effective in its use of data. Data scientists, operating under a clear data strategy and with influence over the product road map, won't just augment your business model; they'll provide a fundamentally different way to achieve your business goals and realize value for your customers.
Borrowing ideas from the military's nighttime operations, we offer concrete steps for companies to do more than simply build a data team. These will help you to be effective in your use of ML/AI:
This doesn't mean that every executive needs to be able to build a neural network; but everyone, from the most junior employee to the CEO, needs to be comfortable asking for the data to back up assumptions upon which key decisions are made. Even the best data scientists won't stand a chance if they're the only data-minded people in the organization. They need to be an extension of a culture that sees and recognizes the value of data for making informed decisions that deliver results.
Furthermore, everyone who will interact with the data team or who will have a role in building AI-based systems must learn what is truly possible with these technologies and where they are appropriate. When AI is it a fit, it can work wonders; and when it doesn't fit, it's terrible. Knowing the difference ahead of time can save you a lot of what we call TERM: time, effort, reputation, and money.
Too many companies have collected data and have hired data scientists, but they haven't weaved them into the business model and product strategy to make them effective. Worse still, they've spent millions on advanced technologies without understanding what advantage those technologies give their companies in achieving their goals.
This is similar to mapping out what makes nighttime missions special: determine which challenges in this company are specifically suited to an ML/AI solution and build the doctrine on how you want to use it for maximum effectiveness. How can you apply ML/AI to be more efficient internally, or to get a leg up on the competition?
A common refrain in the military is, "failure to train is training to fail." That's why people at every level of the organization spend so much time learning about their own equipment, and also rehearsing realistic scenarios as a combined force. Training cycles follow a "crawl, walk, run" approach, starting small and then building in complexity and difficulty such that teams master simple concepts before trying to tackle more difficult ones.
In the data sense, this means tackling basic problems to learn fundamental competencies and ensure foundational support before trying more complex tasks. One approach is to work through the entire lifecycle of a few ML/AI projects, from the initial research and planning on the one end, all the way to deployment and monitoring on the other. Starting with smaller, lower-risk projects gives you the chance to run through these steps when the stakes are low and the lessons learned are cheap. Don't worry if the first couple of rounds are rough; making a lot of mistakes early on will help your larger projects run more smoothly.
Training is an opportunity to practice the plan and to practice the coordination and logistics that make any plan work. This holds even when that plan undergoes rapid changes to meet emerging threats. The training you do now will get you thinking about what might go wrong later and how to adapt. Because something will go wrong at some point. There's a reason the phrase, "no plan survives first contact with the enemy" is repeated so often in the military. By developing your contingency planning muscle, you'll learn how to identify and then prepare for the inevitable problems. This will also make you more agile when handling surprises.
In the realm of night operations, your problems can range from "the adversary develops their own (or, outright steals your) NVGs" to "they learn how to use bright lights to thwart our night-vision capabilities." Both of these level the playing field, which is precisely what you do not want, but it can happen. The military hones its strategies and tactics in training centers with realistic opposing forces trained to act as our adversaries.
In the data world, the analogy is that all of your competitors develop ML/AI practices. Perhaps they find ways to be more effective with it than you are. And on the other end, people may learn how to game your systems, corrupt your data collection processes, or otherwise exploit your weak spots. Cultivating the ability to think like your competitors -- even undergoing "red teaming" exercises specifically designed to expose your weaknesses early on -- will help you craft a business strategy and approach to your market that not only looks good on paper, but delivers results.
In the early days of predictive modeling and advanced analytics, it was possible to gain an edge simply by hiring smart people and buying the right tools. This approach is no longer sufficient if you want to "own the data" in the same way military combatants can "own the night." The entire culture of your organization has to reorient to what these tools and techniques offer in a way that effectively leverages their strengths and minimizes their weaknesses.
If you see this as an opportunity to just "add a little AI" to what you're already doing, you'll quickly fall behind competitors who achieve tactical superiority and strategic gain. The only way for you to keep pace is to make the most of your ML/AI potential.
The Importance of Simulating Data
Why it's important to be able to simulate your own data.
The Lifecycle of an ML/AI Model
It's not just train-test-deploy