Over the past year, Microsoft, Sony, and other companies have filed a number of patents involving generative AI that changes content in real time. Patents often never become products, but the steady stream of filings suggests real interest in the area. Microsoft states that its most recent patent is limited to cleaning up and translating broadcasts. The broader concept, however, leads to concerns about accuracy and misinformation.
Ever since generative AI blew up, videos made or edited by AI have been some of the biggest crowd-pleasers. Now Microsoft wants in on the live side of things. Patentlyze recently found a Microsoft patent filing from late last month. It describes a method for improving how AI edits real-time streaming.
The first claim is where it gets interesting. Basically, the AI listens to a broadcast’s audio, picks out certain words and swaps them for different ones, depending on who’s watching. It looks at things like your age, where you live, and what language you grew up speaking. Microsoft says the point is to tidy things up by cutting pauses and filler words. But the same system could also censor swearing or change words based on what it finds online.
The claim could support instant fact-checking, but it doesn’t say whether users would be told which edits were made or why. It can also hide edits by adjusting lip movements in the frames in between, so viewers might end up watching AI-edited streams without realizing it. Scammers already use real-time AI face swapping. Altering what is said in a livestream could make misinformation and fraud worse.
There’s also a second, standalone claim, and this one is about translation. YouTube can already translate audio with AI, but Microsoft wants to do it for livestreams and make the lips match the new language too. One big difference from the first claim: the translator holds on to the original recording as a “ground truth.”
Microsoft’s livestream patents aren’t an isolated experiment. They echo a wider ambition to let AI reshape what people see and do while they’re still in the middle of it. In July, the company filed a claim that took the idea into gaming, imagining a game that rewrites its own code and adds new content as it learns how you play.
Sony has also recently filed multiple patents. Some use AI to predict upcoming frames and the next button a user will press, with the goal of improving game performance and reducing input latency. Another would use AI to control game time progression, difficulty, and other aspects, or to fill in for players who temporarily leave online sessions.
Maybe you would like other interesting articles?