Australian researchers have developed AI technology that can read minds, although the system is a few years away from being rolled out commercially for uses like helping stroke patients.

The study by the University of Technology Sydney team is the spotlight presentation at a US conference being held this week.

The team invited 29 participants to read while monitoring their brain activity using an electroencephalogram (EEG).

The participants wore a “cap” that recorded their brain signals.

The researchers used an artificial intelligence model called DeWave to translate the signals to words and sentences after training it on large quantities of EEG data.

UTS professor Chin-Teng Lin, who led the research, said the work was a pioneering effort in translating raw brainwaves into language.

Combining neural decoding with large language models was opening new frontiers in neuroscience and AI, he said.

Elon Musk has championed technology called Neuralink that would implant electrodes in the brain to achieve similar results, however the Sydney-based project does not require the invasive process.

But the work still has limitations.

The use of a cap with participants rather than implanting electrodes meant the signal was noisier.

In the study, the accuracy of translating participants’ brain activity to words was about 40 per cent.

However, the study paper outlining the research was completed about six months ago

Prof Lin said the team had since improved accuracy to about 60 to 70 per cent, although he cautioned that rate only applied after training the system for a specific application.

“For its general application, we still have some way to go,” he said.

“The accuracy is still not high enough.”

The study was also limited to participants reading set texts, rather than thinking freely.

Prof Lin said commercialisation of the technology could take two to three years.

 

Steven Deare
(Australian Associated Press)