Tech Review: Clips is like a GoPro with a mind of its own. It’s the first camera that seeks to use AI to replace the main job of photographers: pressing the shutter button. It also tells us a lot about where technology is heading, for better and worse.

Share story

The very idea of the first camera from Google causes people to freak out.

The tiny Clips camera, which arrived in stores Tuesday, decides on its own what to record. Whenever it’s on, artificial intelligence inside is watching. “This doesn’t even *seem* innocent,” tweeted Elon Musk after Google announced Clips in October.

So in the interests of science and averting dystopia, I’ve been using a $250 Clips for the past two weeks. I turned its eye on a toddler playgroup. I wore it in the subway. I clipped it onto a feather toy at a cat cafe.

And I learned Google’s candid camera isn’t an evil all-seeing eye. In fact, it’s not even good enough to replace the need for your camera on your phone. But it tells us a lot about where technology is heading, for better and worse.

Clips is like a GoPro with a mind of its own. It’s the first camera that seeks to use AI to replace the main job of photographers: pressing the shutter button. It’s not hard to imagine the appeal to helicopter parents and anyone who spends too much time operating a smartphone camera. You set this thing near your baby (human, feline or canine) and it will sift through all the boring moments to record 7-second video bursts of some of the interesting ones. Many of us record far more than we ever edit, or even look at. To Google that looks like a problem AI might solve.

What’s just as remarkable as Google’s AI advances is that the tech giant seems to have understood it makes people uncomfortable. In a number of notably un-Googly ways, Clips plays it extra careful with how it handles data that could easily become yet another privacy incursion.

At least on this first version. And Clips is definitely just the beginning for AI in cameras and all sorts of everyday things.

Getting the shot — sometimes

From the moment you twist the lens on the Clips to turn it on, the camera starts hunting for interesting shots. Its wide-angle lens can see a bit more than most phones, though it captures scenes at a lower quality and without any sound.

But how does Clips know what’s interesting? Google says it spent three years studying what people do with cameras. Then it hired three professionals — a documentary filmmaker, a photojournalist and a fine-arts photographer — to train its camera how to pick good shots when they happen. Good: Stability with a little bit of motion. Bad: Fingers obscuring the lens. They taught it to look for joyous faces, especially of children, and to recognize dogs and cats.

When I took it into the real world, Clips captured about half of what I would have chosen — and some of what I would not. Google says AI isn’t about finding the needle in a haystack as much as it is clearing away some of the hay.

I brought one to a San Francisco cat cafe to see if it would suffice in a setting literally bursting with cute. It got some Like-worthy shots, but didn’t replace my phone. I had to keep moving Clips to where the action was in view and in focus, ideally within 3 to 8 feet.

With Clips, in AI you must trust. There is a button you can use to force it to capture, but if you’re using that you might as well get out your phone. There’s no screen on the back to show you what it sees. It can learn which faces are important if you take a 3-second close up or give it permission to learn from shots in an existing Google Photos library.

If you’re in the market for a camera to record anything other than kids or pets — from recitals to scuba trips — you’re better off with an equally tiny $400 GoPro Hero 6 Black.

But is it evil?

Even with a narrow focus on kids and pets, Clips raises thorny questions. The first one that came to my mind: Does it find black and brown children as appealing as white ones?

I tested Clips with a half dozen kids who were Asian, African American and white, and the software seemed equally interested in them all. Google says it is now more sensitive to building inclusive AI, but its algorithms are still a black box and are only going to get more complicated from here.

And forget missed shots: If I’m going to let a product peer into my life, who gets to see what’s going on? Just because Clips wasn’t trained to recognize images other than faces and pets doesn’t mean it won’t witness some more eyebrow-raising life moments.

Here’s where Google surprised me with some unusual — at least for Google — privacy concessions. Clips doesn’t connect to the internet: Its intelligence lives on the camera hardware, not in the cloud. Google already uses AI to look at and analyze people’s shots in a far more aggressive way with its popular Photos service — but Clips doesn’t rely on that capability.

Clips shots stay physically on the camera until you use a companion Android or­­ iPhone app to either transfer a shot over to your phone’s camera roll or delete it. (Clips has 16 GB of storage, which I found hard to fill because its battery only lasts about three hours.) Google says it also isn’t using our data or our choices about which clips to keep or delete to improve its algorithms.

Other issues remain: While you may have control over your own camera, what kind of consent do you need from the people you’re recording? Clips does have a blinking light, but it is small.

This camera is also an additional beachhead in our playpens and cat palaces for a company that already has way too much data about us.

Clips may be a first for a point-and-shoot camera, but it tells us a lot about how AI is arriving in our lives. I guarantee that improving your cat videos is just the beginning of Google’s ambitions.

Google deserves a high-five for showing there’s a way to design AI-powered products that doesn’t require beaming data about our lives over the internet. The risk is that could make us a little more numb if (or when) Google decides to update its terms of service and connect Clips to the cloud. Or puts out a new version that acts more like a surveillance camera.