Research challenges

Although BMIs are rapidly maturing, several challenges remain before they could come into widespread use:

• Better technologies for measuring neural signals while remaining non-invasive. Ideally, one would like to measure outputs of thousands of neurons with a high signal-to-noise ratio. One alternative to fMRI that is attracting significant attention in recent years is functional near-infrared spectroscopy (fNIRS). Such signals can be used in combination with EEG to enhance measurement [135,231].
• Improved bandwidth in terms of bits-per-second that can be commanded by the user so that there are clear advantages over using body movements or controllers. VR systems with non-invasive BMI typically offer up to one bit per second, which is woefully inadequate [177].
• Better classification techniques that can recognize the intentions and decisions of the user with higher accuracy and detail. Modern machine learning methods may help advance this.
• Dramatic reduction in the amount of training that is required before using an interface. If it requires more work than learning how to type, then widespread adoption would be unlikely.
• A better understanding what kinds of body schemas can be learned through the feedback provided by VR systems so that the brain accepts the virtual body as being natural.

Thus, with the exception of helping people with motor disabilities, BMI has a long way to go before reaching levels of mind control that are expected from science fiction.