Regulator Accuses AI Video Firm of Deceptive Marketing

Facial recognition software maker IntelliVision reached a settlement with regulators over deceptive marketing claims, including that its tools worked with “zero gender or racial bias” and market-leading levels of accuracy.

See Also: Maximizing data utility in mission delivery, citizen services, and education

The U.S. Federal Trade Commission said in a complaint Tuesday that from late 2018 through early 2024, the company made numerous false claims, including that its software “can detect faces of all ethnicities, without racial bias, and recognize them from a database of images,” and that it trained its software using “millions of images from across the world.”

Commissioners voted 5-0 to issue the complaint and accept the settlement agreement with the San Jose, California, company – not to be confused with Intellivision, a Mattel Electronics’ games console first released in 1979.

“Companies shouldn’t be touting bias-free artificial intelligence systems unless they can back those claims up,” said Sam Levine, director of the FTC Bureau of Consumer Protection. “Those who develop and use AI systems are not exempt from basic deceptive advertising principles.”

IntelliVision, which sells facial recognition software used in home security system and smart home technology that’s powered by AI, says its software is used in more than 15 million cameras and other edge devices globally.

Unlike companies such as Ring, Arlo and SimpliSafe, the video AI vendor doesn’t sell direct to consumers, but to original equipment manufacturers, large integrators and consultancies, and large end users. Examples of where the company’s software gets used includes in equipment manufactured by its former parent company Nice North America, including its 2GIG Edge home security system, which enables users to scan their face to access the system, as well as the Elan Intelligent Touch Panel, via which users can scan their face to unlock the touch-based controls.

The FTC said IntelliVision falsely claimed its software included anti-spoofing “liveness tests” that ensured the tool couldn’t be spoofed using prerecorded videos or static photographs. When pitching its products to trade customers considering whether to build it into their own products, IntelliVision also “claimed its facial recognition technology has one of the highest accuracy rates on the market,” it said.

The company failed to substantiate these claims, including its supposed high levels of accuracy and being bias-free, or the efficacy of its anti-spoofing technology, the FTC said. The company also didn’t train its tool using millions of faces, but rather using “images of approximately 100,000 unique individuals,” after which it created “multiple variants of those same images and faces” for training purposes.

Test data demonstrated that IntelliVision’s tools fared very differently depending on an individual’s region of birth and sex, as is typical for such tools, according to the U.S. National Institute of Standards and Technology. “NIST testing has found that many facial recognition algorithms produce significantly more false positive ‘matches’ for certain demographics, including West and East African, East Asian and American Indian than for images of Eastern European faces,” the FTC said. “NIST also found false positives to be higher in women than men.”

IntelliVision submitted its facial recognition algorithms to NIST for testing multiple times in 2019, 2022 and 2023, and the results published by NIST showed that the software’s effectiveness exhibited both of those types of false positives, and also failed to distinguish itself against the competition, despite claims of being market-leading.

“In terms of false non-match rate IntelliVision’s algorithms were not among the top 100 best performing algorithms tested by NIST as of Dec. 19, 2023,” the FTC said.

IntelliVision didn’t immediately respond to a request for comment on the allegations or settlement agreement. As is typical with a company that agrees to an FTC settlement, it neither admits nor denies the allegations in the complaint. If an organization breaks its settlement agreement, the agency can sue them in federal court, seeking civil damages.

This marks the second major facial recognition settlement to be reached by the FTC. Last December, the agency banned U.S. drugstore chain Rite Aid from using facial recognition technology for security surveillance for five years. The regulator announced the ban after finding the retailer failed to implement safeguards to combat that risk posed by the tools producing inaccurate outputs based on race and gender, after rolling out such technology across hundreds of stores.

In a concurring statement issued alongside the complaint against IntelliVision on Tuesday, Republican FTC Commissioner Andrew N. Ferguson said the agency isn’t suggesting that all AI systems will be bias-free, in the sense that they produce equal false-positive and false-negative results across various race and sex groups.

Rather, he suggested IntelliVision’s own use of the word “bias” in its marketing materials failed to make clear what the company meant.

“If it intended to invoke a specific definition of ‘bias,’ it needed to say so. But it did not say so; it instead left the resolution of this ambiguity up to consumers,” he said. “IntelliVision must therefore bear the burden of substantiating all reasonable interpretations that consumers may have given its claim that its software had ‘zero gender or racial bias.'”

IntelliVision’s ownership has recently changed. Nice North America, a subsidiary of California-based smart home, residential and industrial technology vendor Nice, was acquired last month by Nipun Vision. The San Jose-based company, also known as NipunVision.AI, was founded in August by Vaidhi Nathan, past founder and CEO of IntelliVision until 2021, who’s now chair of the new company’s board.

In a statement, IntelliVision said it plans to focus on the security and surveillance market, as well as ADAS and DMS fleet dashcams, respectively referring to an advanced driver assistance system, and a driver monitoring system.