Big Tech Forced To Retract AI Claims After Accusations Of "Overstating Capabilities"

To absolutely no one's surprise, the Silicon Valley AI hype machine has gotten ahead of itself. To the point where an industry group - nominally expected to wink and nod at 'puffery' - became so concerned at outlandish advertising claims to consumers about AI capabilities that it has called a number of them out - and from big names like Apple, Google, Samsung and Microsoft.

The key issue here appears to be that as corporations invest more than they anticipated in AI, but with disappointing results, they have taken to advertising to consumers and, to put it politely, 'overstated' capabilities. That was ok when they made the claims to financial investors who are expected to be sophisticated enough to do research. But having been caught inflating claims to the public, Big Tech firms are now being forced to retract or revise those exaggerations. JL

Patrick Coffee reports in the Wall Street Journal:

Apple, Google, Microsoft and Samsung are revising or retracting claims about their newest AI products in response to an ad-industry self-regulatory group about whether marketers are overstating capabilities of AI features. Microsoft added a disclaimer to surveys that 75% reported greater productivity after using Copilot for 10 weeks, clarifying it reflects perception not objective measurements. Samsung dropped assertions its AI-powered refrigerator recognizes what’s in your fridge, so you know what’s inside.” As AI worms its way into every industry, customers should expect a surge in questionable marketing. "When you market AI like magic, you’re going to invite scrutiny. It’s a performance claim - you need to back that up." 

Apple, Google, Microsoft and Samsung have been revising or retracting certain claims about their newest, hottest artificial intelligence products.

The tech giants made the changes over the past year in response to a probe by an ad-industry self-regulatory group into whether marketers are overstating the capabilities or availability of AI features.

The Federal Trade Commission has also been studying the sector’s advertising, in September announcing several legal complaints as part of a crackdown it called Operation AI Comply.

 

The scrutiny comes as the biggest names in tech seek returns on their massive investments in the AI race, holding splashy events, publishing demonstration videos and running advertising to bring in customers. AI developers including OpenAI and Anthropic ran commercials during the Super Bowl, the biggest television event of the year. But it can be challenging to describe AI features quickly and compellingly if also giving all the caveats.

“That is the struggle in a lot of these cases around AI tools and products,” said Laura Protzmann, an attorney at the National Advertising Division of BBB National Programs, the industry group that has probed the issue.

NAD typically referees complaints that companies lodge against competitors in categories like light beer and cellphone service. But it began proactively examining AI marketing when more companies started promoting claims about their new tools to the public, according to Protzmann.

“We thought that it was an important area to focus on, in part because these are hard claims for consumers to evaluate themselves,” she said.

While NAD found that some of the marketers’ claims had support, it identified notable exceptions.

NAD found in April that Apple’s website had prominently advertised unreleased AI features for the newest iPhone as “available now,” for example, and that its small-print disclosures weren’t clear or conspicuous enough.

Most of those features have since arrived, but Apple’s site now more visibly warns that new capabilities for its Siri digital assistant remain in development. Apple also pulled an ad, released last year, that depicted Siri checking the iPhone calendar of “The Last of Us” star Bella Ramsey to summon the name of an acquaintance from a meeting months earlier. That ability has yet to see the light of day. Although Apple disagreed with the conclusions, according to NAD, it pledged to follow the group’s recommendations.

The company didn’t respond to requests for comment. An Apple executive in June told The Wall Street Journal, however, that the features it promoted had worked enough to demonstrate but didn’t “converge in the way quality-wise that we needed.” Siri is now getting a new underlying architecture, the executive said.

Google last year unlisted a YouTube video showcasing the Gemini assistant—including a brief disclaimer that the depicted actions of Gemini had been “shortened throughout”— after NAD began examining its depiction of performance and speed. The 2023 video, which shows Gemini doing everything from identifying a picture being drawn to composing music, now lives mainly on a company blog post discussing the kinds of prompts used to assemble it. The video shouldn’t be seen as an accurate depiction of how the product works, according to a Google spokeswoman. Microsoft removed a page promoting the Business Chat function of its Copilot assistant that claimed the product works “seamlessly across all your data.” An NAD decision released last month concluded that the phrase might lead consumers to think Copilot could perform tasks like switching from one Microsoft application to another and generating a new document there. Users still have to take manual steps to do that, NAD said.

Microsoft also added a disclaimer to survey results, including a finding that 75% of respondents reported greater productivity after using Copilot for 10 weeks, clarifying that the numbers reflect people’s perception rather than objective measurements.

Microsoft disagrees with some conclusions about misleading language but agreed to follow the NAD’s recommendations, said Jared Spataro, chief marketing officer of Microsoft’s AI at Work division.

The removal of the Business Chat page preceded the release of the NAD decision, according to a Microsoft spokeswoman. Microsoft created a new page for the renamed Copilot Chat in its place.

A Samsung AI refrigerator was displayed at an event in Seoul in March. Photo: SeongJoon Cho/Bloomberg News

And Samsung last month agreed to drop assertions that its AI-powered refrigerator “automatically recognizes what’s in your fridge, so you always know what’s inside” after NAD began an inquiry. Protzmann said the NAD found that the camera behind Samsung’s “AI Vision Inside” product can recognize just 33 specific food items and only if they are clearly visible, but the group didn’t issue an official judgment because the language in question had been abandoned already.

The FTC has used its own power to pursue penalties over allegedly false AI claims.

A proposed court order against a company called Ascend Ecom, whose ads promised that customers could make $10,000 a month by purchasing an AI-powered e-commerce operation, would permanently ban its founders from marketing similar services and require them to forfeit related assets as part of a $25 million monetary judgment.

Ascend Ecom’s attorney declined to comment.

As AI worms its way into almost every industry, customers should expect a surge in questionable marketing, according to George Heudorfer, adjunct professor of marketing at the University of New Haven’s Pompea College of Business.

“When you market AI like magic, you’re going to invite scrutiny,” said Heudorfer. “Sometimes that exaggeration isn’t just puffery. It’s a performance claim, and you need to back that up with something more than a keynote and a catchphrase.”