Ask HN: Using AI to repair spinal cord injury

[h_tbob]

This is a tech forum, not a medical one, but I'll ask anyway.

In the future, robots will obviously be doing surgery. Is it feasible to think that a robot with sufficient number of limbs and microscopes could actually repair a spinal cord injury, instead of doing something like neuralink to fix it?

[yawpitch]

Autonomous robotic surgery, even for relatively simple and usually routine procedures, is way off. Robots are currently in use in numerous surgeries as highly precise and repeatable assistants / instruments, but the state of the art on robot safety doesn’t come close to the point of even allowing minimal autonomy, and these devices are all carefully and precisely controlled by direct human operators. Human bodies are just not that consistent or predictable an operating (excuse pun) environment to allow much else without significant development from today’s state of the art.

On the level of spinal surgery, I’d doubt it, unless we’re talking nanoscale machines capable of manipulating tissues as small as single synaptic junctions at the terminal ends of axons and dendrites. Maybe if we developed some chemical or other material that allowed direct neuron regrowth and reattachment and all you needed was precise alignment of the spinal cord itself then there’d be some positioning limit that robots could reach that humans couldn’t, but it’s all very much in the realm of science fiction at the moment.

[h_tbob → yawpitch]

Honestly dude, I was thinking that nanoscale would be the only way to do it. But I don't know enough about biology. But my (super basic) understanding was that it's just too small for a human surgeon. So I was thinking maybe a robot could do it.

But it sounds like, from what you're saying, this is a super difficult task. But let's hope somebody invents it!

What do you think about this idea: Humans operate using Apple vision pro, and wear special tracking gloves, or better yet, some camera specifically can tell using vision algos the 3d position of everything. Then using thousands of surgeries, combined with notes and stuff, do you think AI could be trained on it, like GPT?

[yawpitch → h_tbob]

I wouldn’t think there’s any conceivable way of training anything we currently see as the state of the art in machine vision ML/AI (and definitely nothing in the area of LLMs) to meaningfully interact with the human spinal cord.

Think of the spinal cord as basically an undersea cable, itself composed of an extremely complex outer insulating structure surrounding thousands of individual microscopic cables, each requiring their own insulation (in the case of long nerves something called the myelinated sheath) to function. Nick the insulation and you change the flow of data… sever the cable and you’ve got to reconnect both that cable and its insulation to get it functioning again. Now, understand that all of this is grown by the human body, rather than lain down according to some plan, and further that every fiber is connected at randomly distributed points along its length to an unknown and likely unknowable number of other fibers and junction boxes and such. Oh, and all of this relies on both chemical effects and quantum physics to work, so really you should throw out the cable metaphor because these nerves can rewire themselves dynamically while working.

Having and repairing a spinal cord injury is pretty much like taking one of those enormous industrial chainsaws that get attached to front end loaders and using it to carve through a room full of computers and all their network gear, then expecting a human being (or a robot helping them) using only another set of huge industrial manipulators to make sense of putting all those chips and wires and infrastructure crudely back together and hoping the results will do the same job.

Honestly the best hope in this area is and always has been stem cells, and specifically embryonic stem cells, which could lead to treatments that allow the specialized cells of these nerves to heal themselves by growing uninjured replacements rather than trying to reconnect the injured cells. If you’re going to use AI in any way to make a difference here, get it to figure out better ways of convincing people to abandon their fundamentally ignorant opposition to research involving embryonic stem cell lines (which means getting them to abandon their religious-based thinking on abortion or at least the idiot idea that budding zygotes that could eventually be people are in fact already people).

[gcheong]

Maybe but depending on the time frame something like this might be more feasible: https://news.northwestern.edu/stories/2021/11/dancing-molecu...

[h_tbob → gcheong]

that's amazing!

[reify]

And there will be form to sign with two questions:

1. Do you want a fully qualifed spinal cord human surgeon who has spent many years studying and practicing his art to operate on you?

2. Do you want some fantasy shiny robot made of titaniun, driven by AI, which uses the combined knowledge of spinal cord surgeons to operate on you?

its a 1 for me...

[h_tbob → reify]

Right now... 1. In ten years... 2!

But make sure to ask me first :)

[duped → reify]

Technology itself is the invention of machines that exceed human capability. That's why even though surgeons still sometimes cut with obsidian blades, those blades are honed by other machines and not human hands.

Robotics are already used in surgery and exceed the capabilities of the hands of the surgical team that control it. It's not a stretch to see where that goes in the future.

What's troubling is that AI is not better than human brains today. It's especially good at being wrong, and making shit up. We don't want whatever tech bros call "AI" guiding surgical robots making up their own procedures when they see a spinal column that looks different from their training set.

[yawpitch → duped]

Uh, just so you know, you cannot “hone” obsidian, in the sense of making it any sharper, the edge is produced by a stress fracture, not by sharpening. Also quite a lot of the world’s obsidian scalpels are still made via what is essentially just Stone Age flint knapping.

[duped]

Why do you need AI to work at a scale smaller than surgical teams already can, instead of better robotics?