This is the coolest part! Turns out, the powers you need are actually lower than what is used for imaging babies :) We measured our probe with a hydrophone on a computer-controlled scanner to get the pressure field, and made sure that it's below diagnostic levels (the generally accepted mechanical index limit is 1.9 and ours was 0.4 peak). We also made sure to avoid the eyes and keep thermals in check.
Totally! We think this is because the brain is hard-wired evolutionarily to interpret smells by danger level first. So maybe there's just more "bad smell" receptors, or maybe the brain treats unknown smells as "uh oh, danger". Lots of cool stuff to test!
Haha, luckily not! It's a very speculative link, so we didn't want to talk about "AI" too much in the main post. But we originally got interested in this concept because we are interested in other forms of input to the brain (other than the classic reading, listening, watching, etc). The nose is interesting because it seems to have many independent basis vectors and very sharp discrimination ability, so it might be a sensor into which you can pack many inputs. LLMs are just a proof-by-example that ~1k input dimensions is enough to really encode semantic meaning.
Woah, that would be wild! It seems like most neonatal ultrasound reaches peak internal pressures of few-hundred kPa to 2 MPa. We ranged from 150-250 kPa. So, a little lower than the lower end of prenatal diagnostic imaging.
So, the pressures are high enough to be stimulating them! But most diagnostic imaging happens at 1-20 MHz, while most neurostimulation seems to occur at few-hundred kHz (we were at 300 kHz, on the mid-high end). So I don't think it's likely that babies are being sent smells?
Sorry it's unclear in the post, they weren't exactly the same! The numbers reported were on Lev, and we swept them around that range for me (Albert). But we didn't take down the exact values, so unfortunately I don't know how similar the maps were. iirc they were pretty different.
Woah I didn't know about that theory, that's really interesting! If I understand correctly, it's that the ligand needs to both fit "physically" and also have the right vibrational mode to have high binding affinity / trigger the receptor? Sounds like the relevant frequencies would be in IR range, or roughly 10-100 terahertz. We're at 300 kHz, so 9 orders of magnitude lower, so we're likely not activating the receptors directly with that mechanism. But, maybe the acoustic radiation force from the ultrasound gives existing molecules in the air enough energy to increase the coupling? And nobody seems to really know how ultrasound neurostimulation really works, so who knows—maybe something similar even happens with neurotransmitters in cortex...
Certainly! We didn't get a chance to test it on more people before we had to take it apart, but we thought the result was too cool to share. Would love to see other folks run with the idea!
Super interesting! That would make sense, because a lot of the nose is presumably dedicated to smelling evolutionarily-relevant smells, most of which are "smells bad, avoid this". The method is very crude right now, but maybe with more fine-grained targetting we could better tune the smell profile.
That's a really interesting concept! The wiki page right now ends in "example of a combination of smells that neutralise each other", which makes me wonder if the "olfactory white" combinations are actually tuned to neutralize? I suspect what we're hitting is a bunch of receptors, and the brain is interpretting it as a common strong and evolutionarily important smell (which I assume has stronger pathways by default).