Very thorough report, reminds of my thesis. Sadly, I wrote that thesis just as enthusiastically a decade and a half ago. (I want to clarify this comment is only referring to the parts of this document that discuss central nervous system BCI being used outside of medical contexts.)
I started my career in central nervus system invasive and non-invasive BCI almost a decade and a half ago. I left the field to go into VC and PE when I realized how stuck it was. I read this report excitedly to see if any new technology had broken through the core barriers that explain why the field has moved so slowly and my take away is that at least in those areas (central nervous system recreational BCI) we are at a complete stand still. Most of the tech you list as new in central nervous system BCI was around or at least being talked about regularly when I was in the field so while it is “new” in neuroscience terms it is nowhere close to AI timelines.
Why is the central nervous system BCI so stuck? The non-invasive stuff can’t get around the stuff that surrounds our brain acting as a low pass filter, (ECOG solves for this but no one is going to get a BCI put under their scull for recreational use). As for the invasive stuff astrocytic scar formation is still the major block of long-term use (basically your brain starts to build scar tissue around the input device and the device needs to create “louder signals” to get through the scar tissue causing yet more scar tissue). This is solvable with (and has been for a while) with immunosuppressants but again that largely rules out recreational use (which was the only thing that interested me).
Still, I am glad to see people still pushing ahead at the field. If everyone allowed themselves to become as dejected as I did it just because it moves so much slower than other EA relevant fields nothing would ever advance.
I think I did a bad job highlighting the recent successes in my attempt to be comprehensive. And I probably shouldn’t have listed the iBCIs first, b/c readers will think that’s the state-of-the-art.
Definitely agree that non-medical BCI will have a slow adoption curve. As I discuss, though, they don’t need a fast adoption curve to be relevant to AI safety.
Very thorough report, reminds of my thesis. Sadly, I wrote that thesis just as enthusiastically a decade and a half ago. (I want to clarify this comment is only referring to the parts of this document that discuss central nervous system BCI being used outside of medical contexts.)
I started my career in central nervus system invasive and non-invasive BCI almost a decade and a half ago. I left the field to go into VC and PE when I realized how stuck it was. I read this report excitedly to see if any new technology had broken through the core barriers that explain why the field has moved so slowly and my take away is that at least in those areas (central nervous system recreational BCI) we are at a complete stand still. Most of the tech you list as new in central nervous system BCI was around or at least being talked about regularly when I was in the field so while it is “new” in neuroscience terms it is nowhere close to AI timelines.
Why is the central nervous system BCI so stuck? The non-invasive stuff can’t get around the stuff that surrounds our brain acting as a low pass filter, (ECOG solves for this but no one is going to get a BCI put under their scull for recreational use). As for the invasive stuff astrocytic scar formation is still the major block of long-term use (basically your brain starts to build scar tissue around the input device and the device needs to create “louder signals” to get through the scar tissue causing yet more scar tissue). This is solvable with (and has been for a while) with immunosuppressants but again that largely rules out recreational use (which was the only thing that interested me).
Still, I am glad to see people still pushing ahead at the field. If everyone allowed themselves to become as dejected as I did it just because it moves so much slower than other EA relevant fields nothing would ever advance.
Seems like minimally invasive ultrasound, endovascular BCI, and optogenetic cell therapy all get around the “core barriers” you cite to CNS BCI.
None were around 15 years ago.
https://pubmed.ncbi.nlm.nih.gov/33756104/
https://clinicaltrials.gov/ct2/show/NCT03834857?term=synchron&draw=2 (endovascular BCI in humans)
https://www.biorxiv.org/content/10.1101/333526v1.full
I think I did a bad job highlighting the recent successes in my attempt to be comprehensive. And I probably shouldn’t have listed the iBCIs first, b/c readers will think that’s the state-of-the-art.
Definitely agree that non-medical BCI will have a slow adoption curve. As I discuss, though, they don’t need a fast adoption curve to be relevant to AI safety.