The U.S. Department of Commerce (DoC) Bureau of Industry and Security (BIS) surprised industry with an interim final rule posted this past weekend. Effective more-or-less immediately, on January 6, 2020, the rule, tediously titled “Addition of Software Specially Designed To Automate the Analysis of Geospatial Imagery to the Export Control Classification Number 0Y521 Series,” appears familiar to anyone who has worked with ITAR in the space industry.
And like the laws surrounding ITAR, the rule here is knee-jerk. Also, it’s akin to closing the barn door and setting fire to the barn after the whole herd of horses ran out.
We understand that many senior officials with the DoC were uninformed of this pending rule. However, someone (or more likely, many someones), decided that software created to automatically conduct analyses of geospatial imagery (and something called point clouds) falls under the bureau’s Export Administration Regulations (EAR). The two reasons provided are national security or foreign policy. Both are unhelpful and overused reasons backed by the BIS, Department of State, Department of Defense, and other agencies.
Specifically, they are concerned about “[g]eospatial imagery ‘software’ ‘specially designed’ for training a Deep Convolutional Neural Network [CNN] to automate the analysis of geospatial imagery and point clouds…” In other words, artificial intelligence technologies for computer vision relating to geospatial imagery and surface coordinate data (point clouds). Our understanding is the rule definition and interpretation are rather narrow – for example, the rule applies if the software also helps to train the network. But the rule’s side-effects (unintended consequences) probably impacts legitimate and ongoing internal R&D work by foreign nationals and related work in other sectors.
CNN has been around since the 1980s, developing rapidly for computer vision over the last decade (see here). Furthermore, the basic foundations for such technologies are widely available on an open-source basis (i.e., while DoC can regulate or restrain U.S. companies from selling such technologies, the crucial building blocks cannot be export-controlled in practice).
Many companies in the United States—Google, Facebook, and Microsoft, for example—are working on AI. They are also working on point clouds and imagery (just, um, Google them). In the space industry, Maxar and Planet talk about using this combination of technologies. Why? Because it’s useful to the billions of people around the world, not the few hundred thousand in the U.S. intelligence and defense communities.
However, U.S. companies aren’t the only ones working earnestly in AI, imagery, and point clouds. We believe China is a main target of the new rule. China has at least three big-name companies investing heavily in this field as well: Tencent, Alibaba, and Baidu. And they appear to be well along in their efforts.
We’ve seen this before. It never works. When the U.S. space industry was saddled with new ITAR restrictions, China continued on its way to build a robust space industry. Imagery satellites? China has more than quite a few. Launch? The country kicked ass in launch cadence in 2019. And as we’ve noted, China is already hard at work on CNNs—not just for imagery, but for the country’s social credit system.
AI, working in conjunction with point clouds for facial recognition. That scenario may sound nightmarish to the U.S. citizen, but it is the sad reality in China, and it serves as fuel to accelerate that nation’s technology development. In other words: (i) China doesn’t require the U.S. software in question and (ii) the U.S. should find ways to stimulate industry’s investment and monetization of AI technologies, not slow down industry’s progress in this very important domain by requiring vendors to wait for numerous waivers.
While the rule was evidently made in secret, it is theoretically at an interim stage. The rule can supposedly be modified, or even better, abolished. The BIS will stop accepting rule comments on March 6, 2020. Again, go here to read the rule (it’s pretty short) and comment. The rule will impact not only the space industry we love but potentially advancements in applications that we use day-to-day.