Past technologies have never offered access to people’s thoughts. Instead, machines such as the fMRI allow us to analyze mind activities to identify the simple thoughts and effects.
For example, we can show people pictures of objects and read their reaction under the fMRI brain scanner. But these machines are massive, and as a result, impractical for daily usage.
A more efficient system would be to create a system that connects directly with a patient’s brain. Such technology could potentially help people with paralyzing conditions like ALS and spinal injuries communicate more freely.
A team of scientists from Columbia University’s Neural Acoustic Processing Lab thought of this and created a temporary AI-powered implant that connects directly with a patient’s cortex. Here is how it works.
Read More: DARPA Create Brain Implants to fix Disabilities
How the AI-Powered Implant Works
According to the researchers, their goal is to create a permanent implant. That way, people that communicate through keyboards can synthesize speech on the go.
“Speech is much faster than we type. We want people to talk to their families again,” one of the researchers, Nima Mesgrani told Science Daily.
So, how does the implant work?
The device captures the patient’s neural signal from their auditory cortex and feeds it to an artificial intelligence network. Here is a sample of the reconstructed speech.
Before then, the AI was trained by reading 30 minutes of a continuous speech to five patients. This not only allowed it decode new words (digits zero to nine), but it also allows the machine to synthesize them.
As a result, independent listeners could recognize three out of every four numbers.
While the technology is far from perfect, it’s an excellent start for a first version. As the implants and dedicated AI evolves, we expect that the tech will get better.
That means a future where people can regain full speech by simply thinking. Furthermore, this could serve as a foundation for a more significant technology; brain-to-computer interfacing.
Comments (0)
Least Recent