This spring, when the teams submitted their results to IARPA, the evaluator teams rated how well they did. In June, the teams learned who advanced to the second phase of Smart, which will last 18 months: AFS, BlackSky, Kitware, Systems & Technology Research, Applied Research Associates and Intelligent Automation, now part of defense company Blue Halo.
This time, the teams need to make their algorithms applicable to different usage scenarios. After all, Cooper points out, “It’s too slow and too expensive to design new AI solutions from scratch for every activity we might want to look for.” Can an algorithm built to find construction find crop growth now? That’s a big switch because it’s swapping slow-moving, man-made changes for natural, cyclic, ecological changes, he says. And in the third phase, which begins in early 2024, the remaining competitors will attempt to do their job in what Cooper calls “a robust capability” — something that can detect and monitor both natural and man-made changes.
None of these sentences are strict “elimination” rounds – and there won’t be one winner per se. As with similar DARPA programs, the goal of IARPA is to transfer promising technology to intelligence agencies that can use it in the real world. “IARPA makes stage decisions based on performance against our metrics, diversity of approaches, available funds, and the analysis of our independent testing and evaluation,” said Cooper. “At the end of phase 3, there could be no more teams or more than one team – the best solution could even combine parts from several teams. Alternatively, there could be no teams that make it to Phase 3.”
IARPA’s investments also often leak outside of the programs themselves, sometimes steering science and technology trails as science goes where the money goes. “Whatever problem IARPA chooses, it will get a lot of attention from the research community,” says Hoogs. The Smart teams are further allowed to use the algorithms for civilian and civilian purposes, and the datasets IARPA creates for its programs (such as those labeled troves or satellite imagery) often become publicly available for other researchers to use.
Satellite technologies are often referred to as “dual-use” because they have military and civilian applications. In Hoogs’s view, the lessons learned from the software that Kitware develops for Smart can be applied to environmental science. His company already does environmental science work for organizations like the National Oceanic and Atmospheric Administration; his team has helped the Marine Fisheries Service locate seals and sea lions, among other things, in satellite images. He envisions applying Kitware’s Smart software to something that is already a primary use of Landsat imagery: signaling deforestation. “How much of the rainforest in Brazil has been turned into artificial areas, cultivated areas?” asks Hoog.
Auto-interpretation of landscape change has clear implications for studying climate change, Bosch Ruiz says, for example where ice melts, coral die, vegetation shifts and land becomes desert. By spotting new construction, it can be shown where people touch areas of the natural landscape, forest turns into farmland or farmland gives way to houses.
Those environmental applications, and their spinout into the scientific world, are some of the reasons Smart sought out the United States Geological Survey as a testing and evaluation partner. But the IARPA cohort is also interested in the findings for themselves. “Some environmental issues are of great concern to the intelligence community, especially with regard to climate change,” Cooper says. It is an area where the second application of a dual-use technology is much the same as the first.
This post Spotting objects from space is easy. This challenge is more difficult
was original published at “https://www.wired.com/story/spotting-objects-from-space-is-easy-this-challenge-is-harder/”