Did you know that brain-computer interfaces can benefit those who have been impaired due to injuries like amyotrophic lateral sclerosis? They do so by converting brain activity into outputs including speech and movement. Recently, researchers from Stanford University have utilized artificial intelligence to transform mental handwriting into physical text.
More about the research
On the whole, research on BCI centralizes on gross motor skills involving large muscle movements i.e. reaching for a computer mouse. With the Stanford researchers, however, they believed that targeting dexterous behaviours like handwriting could greatly improve the rates of communication.
So, this study involved a man aged 65 who suffered a spinal cord injury, thus resulting in limited motion of the limbs. He was connected to several microelectrode intracortical arrays, in order to record neural signals in the motor cortex. Then, this was used to collect data on his brain activity as he tried to mentally hand write sentences, after which a decoder was trained.
In particular, a recurrent neural network approach was employed, converting brain activity into ‘a time series of character probabilities’. The RNN also accounted for system processing time by including a time delay of one second when predicting the probabilities.
What are RNNs?
Recurrent neural networks are used in natural language processing, or any application that uses previous data to influence the current output. So, RNNs function by using an internal state (memory), connecting all the inputs together.
Results
The results of this research study were spectacular; the participant had a typing speed of 90 characters/minute with 94% accuracy. These results exceed all other ones from BCI experiments and are quite similar to those among the participant’s age group. Hence, it has become evident that mental handwriting is most definitely possible!
Sources
Written by Amanda Y