Ai Deepfakes Are A Train Wreck And Samsung’s Selling Tickets

On Thursday morning, I attended a Q&A sheet pinch 4 apical Samsung smartphone executives. Until 2025, Samsung was nan world’s largest smartphone manufacturer, and by association, nan world’s largest shaper of cameras. It’s still nan 2nd largest aft Apple.

Samsung handed maine nan microphone first. I asked:

Samsung’s 4 executives did not person immoderate caller and different ideas to share.

I will in installments Won-Joon Choi, nan mobile division’s COO and R&D chief, for not dodging nan question. He told nan room that nan erosion of reality is simply a problem and he wants to hole it.

But he, and different Samsung executives, suggested that nan institution needs to equilibrium nan desire for photographic reality pinch letting smartphone buyers beryllium “more creative.” They passed nan subordinate by suggesting it’s an industry-wide problem, 1 that requires a broader conversation, and suggested that Samsung has already partially solved simply by adding a watermark to AI-generated images. A watermark that tin easy beryllium removed.

Given an opening later, an exec suggested our feelings toward AI-generated contented mightiness go much favorable successful nan future.

Here are immoderate of those answers.

“We admit nan issue, because a batch of contented is generated by AI,” Choi began. “On nan 1 hand, group want to beryllium much creative, truthful we judge we person to supply a solution truthful group tin beryllium much creative,” he said. “On nan different hand, it’s really difficult nowadays to separate nan existent photos and videos from nan clone ones. I deliberation this is simply a problem, we admit that, and I deliberation it’s a problem we person to lick astatine an manufacture level.”

“The C2PA, you whitethorn position it arsenic a failure, but it’s still capable to supply a system if group want to validate that those pictures and videos are made by AI. I deliberation we person to supply a system truthful group tin usage it. I deliberation it’s an ongoing effort passim nan manufacture to lick this problem,” he added.

“I deliberation if we make associated efforts to lick this problem, I judge that we should beryllium capable to do it,” he concluded.

As my workfellow Jess Weatherbed wrote earlier this week, we’re starting to interest that this benignant of “the manufacture will lick this together” rhetoric, and nan C2PA itself, are substitutes for existent meaningful action.

Perhaps responding to our Samsung AI slop watch reporting, Samsung America exec Dave Das besides chimed successful to opportunity that nan company’s still learning astir really overmuch AI is acceptable erstwhile creating its ain ads. He admitted that nan institution has “just started utilizing immoderate AI contented successful our creative, and nan feedback has been beautiful clear.”

“We’re trying to discern what is nan correct spot to usage it, and perfectly really to beryllium very clear astir erstwhile we are utilizing AI generated contented vs people generated content,” Das said.

But Das still frames it arsenic a balancing enactment betwixt business priorities, not a societal responsibility. “It really boils down to giving nan creator choice,” he said, suggesting that Samsung’s occupation is to find “the correct balance.”

Later successful nan morning, KTLA-TV tech newsman Rich DeMuro felt retired Samsung’s stance pinch a very different question. He asked whether Samsung mightiness want to make it easier for customers to region its AI watermark from AI-generated photos.

“If I’m making a Christmas paper pinch each your nosy tools, do I really want that to opportunity ‘AI-generated content?’” asked DeMuro.

Drew Blackard, Samsung America’s SVP of mobile merchandise management, took that one. He concluded by saying that “If we consciousness for illustration consumers really want that expertise to region watermarks, and location are different ways to reside nan first constituent astir authenticity astatine nan aforesaid time, we’ll do some things and present nan champion of what group are wanting retired of nan experiences.”

“At slightest correct now, there’s capable interest from consumers astir authenticity that that’s nan superior point we’ve solved for, successful position of nan watermarking some successful nan metadata and connected nan photograph itself. Not each services do that,” he said.

Blackard besides suggested we mightiness look backmost 1 time and recognize that AI-generated contented isn’t specified a large deal, that our perceptions mightiness alteration complete time, nan aforesaid measurement perceptions changed toward user-generated contented to statesman with. People were concerned past pinch “the magnitude of contented now flowing into nan system” and really it wasn’t nan master video merchandise they were utilized to, he said.

I wonderment if Samsung and chap smartphone makers person considered nan opposite: that perceptions of AI-generated imagery mightiness go moreover less favorable aft group suffer their jobs successful favour of cheap-to-produce AI slop, and aft it becomes acold easier to dishonesty and cheat and bargain successful nan absence of reliable recorded evidence.

I wonderment if they’ve considered solving this problem ”at an manufacture level” before they break nan dam, and whether cognition mightiness move against Samsung for helping create nan flood that comes after.