On many occasions I have imagined a day when the smart glasses on a stranger's face would quietly listen to our conversations. After reading MIT News this week, I think that day is coming sooner than I expected.
The article is "Making machines that can listen" (MIT News, 28 October 2026). It profiles Professor Paris Smaragdis, who joined MIT in 2025. He describes his work this way:
"I like to say that I do the same thing the computer vision folks do, but instead I do it for hearing."
Regular readers know I have spent years worrying about machines that see us. Now the same science is being applied to machines that hear us.
What this research makes possible
According to the article, Prof. Smaragdis's group works in three main areas:
- Sound source separation: neural networks that pull individual sounds out of a mixture of audio
- Smart devices that can "augment a person's audio surroundings"
- Large language models applied to speech recognition and audio enhancement
His team also helped clean up the old, muddy Beatles recordings for The Beatles: Get Back, and that work helped win an Emmy in 2022.
Why does this matter for smart glasses? Turning speech into text is no longer the hard part. The hard part is a crowded restaurant or a busy street, where many voices overlap. Source separation removes that obstacle. It can turn a noisy room into one clean track for each speaker.
The article also mentions algorithms that do not need fixed sampling rates or frame sizes, which could lead to more efficient computing. That is exactly what a small, battery-powered device worn on the face needs. (This last point is my own inference, not the article's.)
To be fair, the article says nothing about glasses, surveillance or privacy. Prof. Smaragdis says his aim is to "enhance the toolbox a musician has." But technology seldom stays where its inventors put it.
What I have been saying
In 2018, in Parekh's Law of Privacy ?, I proposed:
"CONCERN FOR PRIVACY INCREASES IN DIRECT PROPORTION TO THE AMOUNT OF PERSONAL INFORMATION"
In the same post, I imagined a time when "each and every piece of furniture / device / gadget / artefact in my home, will transmit 24 hours" of my activity to computers around the world.
That same year, in Close Your Eyes ?, I worried about iris scanners and asked:
"Now, who but the very dead, can keep their eyes closed for ever?"
Today I must ask the same question about our ears. Who can stop talking forever?
In September 2025, in When Glasses Become Gateways, I wrote:
"Augmented devices with forward‑looking cameras and always‑on sensors change the rules of public and private spaces."
"Who owns those memory streams? How long are they stored? Who can query them?"
And in May 2026, in Pronto Sparks Privacy Alarm, I warned that cameras in a home "will inevitably record family members, neighbors (through windows), and visitors who never consented."
Replace "cameras" with "microphones" and "home" with "café", and you have the problem in front of us now.
Why separation changes everything for bystanders
Until now, background noise has given us accidental privacy. The couple at the next table was "protected" because their words were lost in the clatter. Good source separation takes that protection away. The people in the background become people whose words can be pulled out and recorded, and none of them agreed to it.
It also works backwards in time. Recordings that are too noisy to use today could be mined again with tomorrow's models. And glasses are far less noticeable than a phone held up to record.
What we can do
As I wrote in an earlier post, there is no way law can outsmart technology, but we must try. Here is what I suggest:
- Process audio on the device and keep nothing. Glasses should transcribe locally and delete the raw audio immediately.
- Use separation to protect privacy. The same technology that can isolate a stranger's voice can isolate only the wearer's voice and erase everyone else's. This should be the default.
- Opt in, not opt out. Voices of other people should be kept only after they clearly agree, as two-party consent laws already require in many places.
- Make recording visible. Show a clear indicator whenever the device is listening.
- Treat voiceprints as biometric data. Our new data protection laws must cover them.
The same research could also give us much better hearing aids. The science is neutral. Our design choices and laws are not. We must make those choices before the glasses reach our faces, and our dinner tables.
With regards,
Hemen Parekh
www.Iam-Immortal.ai / www.IndiaAGI.ai / www.YourContentCreator.in
No comments:
Post a Comment