Transport hubs in Los Angeles, Denver and Washington are soon to trial Total Recall-style high-speed body scanners

A startup bankrolled by Bill Gates is about to conduct the first public trials of high-speed body scanners powered by artificial intelligence (AI), the Guardian can reveal.

According to documents filed with the US Federal Communications Commission (FCC), Boston-based Evolv Technology is planning to test its system at Union Station in Washington DC, in Los Angeless Union Station metro and at Denver international airport.

Evolv uses the same millimetre-wave radio frequencies as the controversial, and painfully slow, body scanners now found at many airport security checkpoints. However, the new device can complete its scan in a fraction of second, using computer vision and machine learning to spot guns and bombs.

This means passengers can simply walk through a scanning gate without stopping or even slowing down like the hi-tech scanners seen in the 1990 sci-fi film Total Recall. A nearby security guard with a tablet is then shown either an all-clear sign, or a photo of the person with suspicious areas highlighted. Evolv says the system can scan 800 people an hour, without anyone having to remove their keys, coins or cellphones.

The importance of it being fast cant be overstated, said Aaron Elkins, an AI professor at San Diego State University, who is developing his own AI security technology. Any place you currently see a metal detector would probably consider an upgrade.

Millimetre-wave scanners are useful because they can identify both metallic and non-metallic items, such as 3D printed guns and explosives. But they have also attracted criticism because scanners were able to capture realistic images of peoples bodies beneath their clothes.

We never build an image that would enable anyone to see anatomical details, so theres no naked peepshow in the first place, says Michael Ellenbogen, Evolvs CEO. None of the raw data is stored and none of the data we do keep is traceable to an individual.

Evolv claims that no human sees what the scanner is looking at. The system uses solid state micro-antennas to steer radar beams over anyone walking through the gate, and to pick up the reflections. That data is then fed to an AI system that has been trained to spot distinctive scattering patterns from all kinds of objects, including firearms, suicide bomber vests and even knives. The scanner also has a camera that takes a photo of each person passing through, enabling facial recognition.

Because it promises to be faster and cheaper than existing millimetre-wave scanners, the new device could bring airport-level screening to venues that were previously difficult to secure.

Transportation is a very soft and attractive target, said Alex Wiggins, the executive in charge of security for Los Angeles County Metropolitan Transportation Authority. Given the recent large-scale attacks at transit facilities in Europe, we need to see if there is technology that can screen large number of peoples and focus in on weapons and explosives.

The three locations named in Evolvs FCC application process upwards of 300,000 people every day, although Denver airport said its pilot project with Evolv had yet to be finalised. The test in LA is due to run in November for three or four days, and will involve thousands of members of the public.

We want to see how finely these [scanners] can be tuned, and what size of weapons and explosives they can detect, says Wiggins.

Evolv was spun out in 2013 from the Seattle-based firm of Intellectual Ventures, considered by some a patent troll but by its founder Nathan Myhrvold as a hi-tech invention factory.

Evolv received $11.8m in funding from investors, including a sizable chunk from Bill Gates, the ex-Microsoft billionaire. The company recently raised another $8m.

Evolv has also benefited from close links with the US government, perfecting its technology with the help of a contract from the US Department of Homeland Security (DHS). According to documents previously filed with the FCC, Evolvs scanner was tested last year at the DHSs Transportation Security Laboratory in New Jersey, and the FBIs headquarters in Quantico, Virginia.

But public tests will be invaluable, said Elkins. What was a really high-performing detection algorithm in the lab can get tripped up in the chaos of the real world, he said. People will go through with ice cream cones and sunglasses and backpacks.

Evolvs Ellenbogen admits that luggage could be a problem. Were looking at what people can carry through the system versus what has to be separately inspected, he said. Our intent is for people to be able to walk through while carrying all the things they would normally carry.

Another issue that worries both Wilkins and Elkins is false positives when people are incorrectly flagged as carrying something dangerous. If too many people are falsely checked, they have to turn down the sensitivity and just flush people through, Elkins said. Then it starts to not really be useful.

The biggest problem, however, could be the very artificial intelligence that makes Evolvs scanner so attractive.

One of the hazards of algorithmic, as opposed to human, detection, is that an attacker who can reverse-engineer a machine can almost certainly find a way to make dangerous objects appear benign, said Ari Juels, a computer science professor at Cornell University.

Countermeasures may be possible, but the research community does not yet have a good sense of how to construct them.

Read more: www.theguardian.com