Why America Needs a National Data Strategy

Don’t forget the data.

As America races with China to lead the world in artificial intelligence (AI), the U.S. government may actually be missing an even greater problem.

AI is the science of training computers to perform tasks with near-human levels of insight. But if we’re training computers to think more like humans, we should acknowledge that human decision making sometimes goes wrong. What happens to the human brain when it is fed bad data? If someone was to make a decision based only on fake news, could they be expected to make the best choice?

That’s why the U.S. may be falling into a trap. We are focusing on AI but not the data that AI needs to function.

The stakes could not be higher. Whichever nation wins the AI race will reap enormous benefits. Not only in military capabilities, but numerous other fields, from intelligence analysis and medical research to self-driving cars and cybersecurity.

While the Biden administration has launched a National Artificial Intelligence Research Resource Task Force that will help develop national resources to support AI innovation and there are numerous government and private initiatives to develop cutting-edge AI, this is still only a piece of the puzzle.

As the old saying goes in computer science, “garbage in, garbage out.” Just like automobiles and airplanes require high-quality fuel for peak performance, AI requires a diet of data that is accurate, consistent and accessible.

Unfortunately, the U.S. government’s AI data ecosystem is a mess. For example, there is no common set of standards that covers data for AI. Nor is there a common database so data can be shared between different agencies and AI systems. Good data can’t guarantee good AI, but poor or incomplete data will certainly result in bad AI.

As CEO of Scale AI, and as co-author of a recent report by America’s leading technology experts on the U.S-China technology race, I have seen how dangerous the data problem is. One of the reasons that I founded Scale AI was the realization that data was becoming a bottleneck that would choke AI innovation and application. As more of the world’s software is written by AI, and machine learning – where AI learns by analyzing vast amounts of data – becomes more common, then data quality becomes paramount.

What happens if the U.S. can’t fix its data problem? The dangers of a data gap for U.S. national security are frightening. From hypersonic missiles traveling at Mach 5-plus, to elaborate kill chains that instantly relay targeting data from sensors to weapons, the speed of combat is becoming too fast for the human mind to comprehend and control without assistance. AI is the future of warfare. AI will be needed to operate autonomous combat aircraft and tanks, sift through huge amounts of drone imagery, recommend the optimum weapon to destroy a target, and analyze enemy battleplans and how U.S. forces can foil them. Computers cannot replace human decision-making, but a nation who falls behind in AI will fight with one hand tied behind its back.

Ironically, the problem isn’t lack of data. Ever since World War II, the U.S. government has devoted enormous resources to creating the finest intelligence collection system in the world. Staggering quantities of information are scooped up by spy satellites, drones and electronic intelligence. But how is this crushing amount of video, audio and Internet traffic to be analyzed and then distributed to those who need it? Data analysis has become so overwhelming that the system is almost paralyzed, resulting in American leaders missing vital clues about world events and the intentions of our adversaries.

Take the Department of Defense’s huge Joint All-Domain Command and Control (JADC2) initiative, a cloud-based system that collects data from all the U.S. military services, and then uses AI to select the best available weapon to destroy targets. Such a real-time kill chain, which can strike time-sensitive targets with incredible speed, is devastating. But JADC2 can only be as good as its underlying AI, and that AI can only be as good as the data that feeds its algorithms.

China already sees the power of artificial intelligence: AI is now a fundamental component of Chinese military technology and strategy, from autonomous weapons to information warfare aimed at disrupting U.S. command and control systems. In pursuit of this goal, China has adopted a disciplined, coherent AI data strategy that exploits the strengths of a big authoritarian state: a lack of separation between military and private research, a huge population that generates vast amounts of data for AI research, and minimal privacy safeguards that enable the government to collect and share that enormous trove of data with AI researchers. And new investigations show that Chinese media, police, military and other agencies are planning to invest even more into sophisticated systems to gather data.

But America has many strengths as well. The U.S. can lead the world in artificial intelligence -if we can devise a national data strategy that harnesses American innovation. To that end, I would propose three key areas on which to focus these efforts:

1. Making the data we're already gathering AI-ready: We’re already sitting on troves of data, but it’s not being put to use. We need to invest the necessary resources to properly process this data and make it usable fuel for future AI applications.

2. Break down data silos: Today too much of the U.S. government data lives in disparate silos, meaning it’s caught up in layers of bureaucracy rather than being utilized to its potential. The U.S. needs a holistic approach to leverage our wealth of valuable data assets in a coherent way and share it across agencies for the purposes of AI for national security.

3. Invest in common standards for the future: To maintain a position of global leadership in AI, we can’t be stuck in a cycle of catch-up where we’re continually cleaning up the data we have – we need to build more sustainable systems for the future. The U.S. must develop common standards around how to produce foundational data assets for our long-term AI efforts, making it simple to put new data to work.

Ultimately, this strategy would require input and buy-in from actors across the board, including Congress, U.S. military and intelligence agencies, defense contractors and the AI industry. And this buy-in must be achieved with the long-term in mind. We can’t just start then stop and expect meaningful results, we must ensure there's a continual investment into a centralized data strategy so that we don't atrophy in our capabilities. The goal would be to make our data significantly more valuable, and ultimately a competitive advantage, in the race for sophisticated AI capability, rather than an impediment.

Without a national strategy for data, America’s AI systems – no matter how technically sophisticated – will come off second-best against China and other potential adversaries.

The stakes are too high to allow that.

Written by: Alexandr Wang, Founder & CEO, Scale AI

Alexandr Wang is the founder and CEO of Scale AI, the data platform accelerating the development of artificial intelligence. Alex founded Scale as a student at MIT at the age of 19 to help companies build long-term AI strategies with the right data and infrastructure. Under Alex's leadership, Scale has grown to a $7bn valuation serving hundreds of customers across industries from finance to e-commerce to U.S. government agencies.