Graava founder Bruno Gregory opens our interview by showing me what, in a parallel universe, could have been a snuff film. He and a friend are biking through Berkeley, Calif., when — wham! — a hit-and-run driver slams into him.
“Today, it’s funny, but then, it was not,” Gregory said.
He recovered, and the driver was soon caught by the police. The incident was being casually filmed from an action camera on Gregory’s friend’s bike, behind him, and the duo was able to give the authorities that recording. It’s a sideways angle into his current company, which hopes to use machine learning to automatically edit video recorded from a GoPro-like camera. But his point was that sometimes, people wish they had been recording when unexpected things happen.
“In two hours of video, there might be one good moment,” he said. “But if you don’t film anything, you don’t get anything.”
Graava’s preorder campaign, including the video embedded above, is focused on the idea that recording everything and outsourcing the editing to a computer will “make memories,” which sounds like something everyone wants to do. But the most interesting question it raises, one that no one will have a good answer to until the product has shipped and regular people are actually using it, is just how good these software-made memories are.
My colleague James Temple, who directs Re/code’s video efforts, was a bit pessimistic: “That will work about as well as robot text editors divided by five,” he said.
What James means is that editing often requires contextual understanding, something AI has historically struggled to recreate; algorithmic news services have instead been targeted at data-friendly stories like sports scores and companies’ earnings reports. Depending on how you count it, we have something like 10 human editors at Re/code, and our parent company Vox has many, many more across several sites.
But the promise of deep learning is that, with enough time, machines will become more capable on their own — and that it’s okay if they sometimes need help from a human. That’s why there’s no way of knowing how good Graava’s tech is until it’s out in the wild.
Gregory explained that the editing algorithms judge based on several criteria: The camera’s raw image, to identify when something dramatic or out of the ordinary happens; an accelerometer and GPS, to measure the user’s motion; a microphone, which lets the user manually command that a certain part of the video be preserved by saying “Graava”; and, with additional third-party hardware, the user’s heart rate, which can be wirelessly interlinked with the video.
In addition to talking into the microphone in the moment, users will also be able to manually drag and drop moments from their recording into an “advanced editing mode,” Gregory noted.
The company is asking preorderers for $249 starting today, though the camera they get in “early 2016” will not have any accessories, such as a stand or case for attaching to equipment like sports helmets. The eventual retail version is expected to cost $399 and will include some or all of those accessories.
This article originally appeared on Recode.net.