We pointed an ultrasound probe at the scent-processing region of the brain to obtain different sensations. Different focal spots corresponded to different smells, which we’ve replicated first-try on two people and validated with a blind trial. The sensations we obtained are:

Here is a video from our blind tasting:

The Setup

Smells are processed in the olfactory bulb. We decided to try to stimulate it with focused ultrasound through the skull. As far as we know, no one seems to have done this kind of olfactory stimulation before - even in animals.
However, after being able to induce sensations of motion the previous week, it seemed promising to try the same for olfactory.

The Anatomy

The olfactory bulb, our target, is tucked behind the top of the nose. That turns out to be a pretty inconvenient location for a couple of reasons:

1. The nose doesn’t provide a flat surface for mounting a transducer for stimulation.
2. It's mostly filled with air, which interferes with ultrasound. Ultrasound needs a continuous medium to travel through, and filling the nose with gel seemed rather unappealing.

Instead, we found that you can place the transducer on the forehead and aim the ultrasound downward towards the olfactory bulb. While this isn’t a perfect solution because the frontal sinuses can weaken the signal, careful device positioning above the sinuses still allows us to reach our general target region.

The ultrasound

We got our first effects using just a handheld probe and some gel, but it quickly became obvious that holding a probe steady by hand makes it nearly impossible to keep the focal spot in the same place. To improve stability, we improvised a makeshift headset, allowing for more reliable positioning. We switched from gel to a solid, jello-like pad for stability and general comfort. In the end, our headset got a bit hacky:

It ended up having a knife taped to the probe for mechanical support At the time, all of our headsets had a knife taped to the probe, as untaping the knife lead to software errors.. At some point we thought of using a mouthguard for fixing the probe relative to the brain. This was a great idea considering the teeth are the only exposed part of the skull, except it turns out you can’t talk about smells while wearing a mouthguard.

Both smells and sensations are strongest on a light in-breath, so we tested by sitting there, with a probe to the forehead, mildly sniffing. Sometimes there is a slight waft of a smell that comes on over a few breaths, and sometimes it just hits you. The first time Albert smelled the garbage, he jerked his eyes open thinking a garbage truck just drove in! This was indoors.

Many of these scents correspond not to specific receptor types but rather combinations of receptors. We think this is because the focal spot is pretty large—300kHz ultrasound in tissue has a wavelength of 5mm, while the adult human olfactory bulb is roughly 6-14mm in length The olfactory bulb can vary in size by up to 3x, depending on "age and olfactory experience", so perhaps (we're making this up) with more usage your olfactory bulb might actually get bigger, leading to better resolution stimulation!.

We found different scents by steering the beam over ~14 mm (20 degrees at 4 cm radius). The distance between freshness and burning was ~3.5 mm. We ensured that the effect was not placebo with an auditory mask (blasting music through airpods) so you don’t hear the probe, though you cannot distinguish the different focal spots through sound anyways. We then tested discrimination in a trial where Thomas selected the focal spots, and Lev was naming the scents. You can check out the full video here.

It is remarkable that we could induce different scents with such little steering (40% of the diffraction-limited focal spot size And potentially even higher, because there was some dead space in between the focal spots, where you don't feel anything.). This suggests that the resolution we have access to is much higher than the spatial resolution of the ultrasound (a kind of super-resolution for neurostim!) In particular, we do not need single-neuron resolution to find an independent basis of scents, upon which we can construct our latent space. To improve this system, the next steps are a more stable setup, increased frequency, more play with focal location, spot size, and stimulus waveform.

The reason stimulating olfactory sensations is interesting is not just "VR for smells", as one might initially assume. The nose has 400 distinct receptor types, and we can distinguish subtle combinations of their activations, so they could serve as a channel of writing directly into the brain, as a means of non-invasive neuromodulation.

The olfactory system potentially allows writing up to 400, if not 800 due to two nostrils, dimensions into the brain. That is comparable to the dimensionality of latent spaces of LLMs, which implies you could reasonably encode the meaning of a paragraph into a 400-dimensional vector. If you had a device which allows for this kind of writing, you could learn to associate the input patterns with their corresponding meanings. After that, you could directly smell the latent space. A bit of ultrasound, a breath in - and you understood a paragraph.

People are able to develop synesthesia - being able to hear colors and see smells, and it might be possible to extend that to semantics. However, at this stage it is speculative.

One could try to make a similar argument for the eyes: take 400 cones on the retina, hijack them, and you've got yourself a 400-dimensional channel. But we think the nose is better. The olfactory system is much simpler and more directly interfaces with core brain regions, like the hippocampus. The signal through the olfactory system is simply less filtered and processed. If you tried to write arbitrary light intensities into a patch of cones, the next step of the processing would be a convolutional neural network-like structure in the visual cortex, and the signal would get averaged out. The embeddings you'd write would never make it into the higher levels of processing in the brain. You can try to encode the information in a more easily perceptible way, such as Chernoff faces, but it would reduce the bandwidth, and learning the remapping would still be very difficult.

In contrast, only a few synapses separate the olfactory receptors from the hippocampus This is why certain smells bring up such strong memories!, which is responsible for memory, as well as from the amygdala, which does emotional regulation.

Finally, personally speaking, the authors use their eyes and ears more than their noses during office work Raphael Hotter noted that this is in fact a general statement, as the usage of eyes and ears extends beyond office work.. The nose is an underutilized channel that imposes fewer bad priors (spatial/tonal maps) than the visual, auditory, and somatosensory.