There are two competing nightmare scenarios for any company unveiling a new brand or logo. The first is that the internet will rise up in righteous indignation about how awful it is or what part of human anatomy it most resembles. (Or both.)
The second and significantly scarier nightmare is that the logo is infringing on an existing trademark. The phenomenon, which is today often called design plagiarism, is often intentional—but can just as frequently be accidental. And no wonder: Searching for similar logos and marks is nearly impossible in the warren of databases of American trademarks, much less in any other country. Now, AI is stepping into the fray—led by TrademarkVision, a company founded by computer vision researcher Sandra Mau. The startup bypasses old-fashioned keyword searching and uses deep learning to match your logo or trademark with millions of trademarks it may infringe upon.
It’s a bit like legal insurance for branding. Mau, who started out doing computer vision research on surveillance for the government, founded TrademarkVision in 2013 after realizing that the technology had much broader applications. "We found that brands have this need for protection," she says. TrademarkVision uses a combination of different image matching algorithms—akin to what Google might use in its reverse image search feature—beneath a higher-level "semantic understanding layer," which is in charge of correctly identifying actual objects in an image. "The combination is what gets returned to the user," Mau explains. So a search for a logo involving Apple might return thousands of similar logos, or many of them owned by Apple Inc. itself. Designers can even search for color combinations to find out if their palette is similar to other trademarks.
AI is evolving quickly, and the company only introduced its latest deep learning search this fall. Mau and her team are actively experimenting with new features, which they upload under their Labs site for anyone to try. One current experiment is an algorithm that will spit out a description of your logomark using the formal classification of trademarks, so I tried uploading one recent logo that raised the internet’s delicate hackles: Airbnb. The algorithm described it strongly as a symbol of infinity—along with other descriptions like a "series of lines forming a triangle" and the letter "A." Not bad. Co.Design’s own logo, a much more straightforward mark, registers as a single letter, white on a black background.
Another recent experiment, still in beta, applies the same technology to industrial design instead of just graphics. Rather than just searching for 2D images, designers will be able to upload multiple views of a 3D design and find out if it’s closely matched to an existing design patent.
It’s easy to see the outlines of the company’s future here: Design patent law is more in flux today than it has been in more than a century. The Supreme Court’s recent decision in the landmark Apple v. Samsung case will limit the power of design patents, but it’s still unclear how. Being able to see how your design—however detailed or general—might infringe on another patent? Or being able to understand how similar designs have approached gaining patent protection? That capability could change how both giants like Samsung and independent industrial designers operate.
Even further down the road, Mau says that the same tech could help companies bust counterfeiters—imagine being able to upload a photo of your product and find a knock-off being sold online. For now, TrademarkVision is expanding its reach and hoping to get more feedback from designers themselves. Though, when it comes to knowing what body part the internet will ultimately see in your logo, you’re still on your own.