In many science fiction books, subvocal recognition is a basic enabling technology. It means using micro movements in the muscles in your throat to identify speech. Researchers have been investigating this for decades but no one has made a system that really works. The navy was interested for scuba divers. NASA wanted to use it for astronauts. Presumably there are other use cases, but it seems no one had a strong enough need to pour money into it. In March 2018 an MIT Media Lab group published their research showing that with electrodes all over your face and a multilayer perception you could reliably recognize a vocabulary of ~30 words. That’s pretty bad, but their method was also naive. It suggests that with a little sophistication, you could make something that really works.
Research shows that people can be trained to subvocalize all their thoughts in a single day[citation needed]. This means with a real system, you could start building apps on people’s thoughts.
According to an otolaryngologist:
“I don't think there's that much specific information available at the throat level. Articulation is in the mouth and lips. Ventriloquists manage to not use lips, replacing v's with vocalized th's, and replacing p's with t's. But it's all going on in the mouth. I have cancer patients who have had large portions of tongue removed… They're almost impossible to understand.” This suggests having electrodes on the face is necessary.
Use modern pre-trained models to make a high quality, fully general subvocal recognition system. Use a dry electrode array in a collar on the throat. Sensors are cheap and it’s important for placement to not matter too much. Electrode readings would go through a battery-powered bluetooth module paired to a smartphone, perhaps with a little preprocessing to reduce bandwidth needs.
Almost certainly would need to N-shot calibrate the device every time you put it on or move it. Maybe it would auto-calibrate whenever you speak out loud.
The initial killer app would be a CBT bot on top of your thought stream. For example:"I should go to the gym”->“I noticed you thought,‘should.’ Can you replace that with‘I want… because…’?”
Compared to BCI, the abuse potential for this technology is minimal since subvocal cloaking techniques are effective and simple(see wikipedia articles).
Audience
Uber/Lyft drivers who don’t want to disturb their passengers
Passengers on public transit etc who don’t want to disturb fellow riders
Any time you want to give a command to your phone in a public setting
Anyone who could gain value from some kind of recording or analysis of their thoughts
Military/law enforcement — lets teams communicate hands free, with local privacy
Risks
Some of the application areas of this technology are more‘vitamin’ than‘painkiller’.
Live-CBT efficacy can’t be clinically demonstrated
Hard to get sufficient recognition quality
Hard to generalize across people
Placement of electrodes is essential to quality
Electrode arrays might be
hard to manufacture
not durable
expensive
Moat
Recognition quality - latency, accuracy(though this is likely temporary)
Device quality - comfort, cost, aesthetic
App ecosystem - may not produce lock-in
Developers can put apps on the Telepath store. Apps have hooks into the processed text stream and can do things like CBT(described above), help you take notes, control devices in your home such as lights and music, send and receive text messages, etc.
Brand
Patent
Might not be possible due to prior art, but perhaps with the addition of language modeling it would be sufficiently novel
Opportunity
Audience
Risks
Moat