AI That Gives Everyone a Voice
16 August 2026
There’s a version of AI progress that doesn’t get talked about enough.
Most of the AI conversation right now is about speed, productivity and who ships the next model first. Every so often, a story shows up that’s really about helping someone do something they simply couldn’t do before.
Google DeepMind recently built a model that reads sign language through a phone camera and turns it into typed text. According to its announcement, the technology is available on the Pixel 11 through keyboard and live-transcription features.
That means someone who uses sign language can search the web, write a text or interact with an AI assistant by signing to their camera, without typing.
For anyone who’s never thought about this before, phones and keyboards have traditionally been built around typing or talking.
If you sign as your first language, every digital interaction has quietly asked you to translate yourself first.
This new model, called SL2T, is Google’s attempt to close that gap.
Privacy and accessibility by design
As DeepMind describes it, rather than streaming full video to a server, SL2T tracks around 130 points on a person’s face, hands and body on the device, sending movement data for translation.
Privacy was part of the design from day one.
The model was trained on more than 100,000 hours of sign language across over 50 sign languages. The initial release supports American Sign Language, with the broader training suggesting room to expand.
Even with strong benchmark results, Google worked with Deaf community organisations and restricted the technology to casual, low-stakes use.
It is not approved for medical visits, legal situations, school or job interviews, where a mistranslation could have serious consequences.
More power to AI for good.
Not the version racing to be first, but the version that sat down with the people it was meant to help and built something for them.
Reference
