Hacker News Reader

Comment by Konnstann | Hacker News Reader

What does this even mean? Hacking an open-loop insulin pump and making it closed-loop doesn't eliminate the need for insulin, did you misread me as saying insulin prices are A-OK and diabetics should just eat healthy?

All power to these knowledgeable, individuals, for doing something dangerous but potentially great. This is HN, full of smart people who understand the risk of failure and will monitor the crap out of their system and adjust and tinker until it's ok. The problem is people making tutorials and publicizing their work, which leads to people not as smart following their tutorial and thinking it will be alright.

There are tons of papers on closed-loop systems, and probably clinical trials are happening as I write this comment. Medicine isn't one of the industries to "disrupt" though, as much as people on here want to believe.

View on HN

Smart insulin pump sounds awesome. Bring them on. If it really is a smart insulin pump. Trouble was the article author was talking about 'reading his emotions' etc. There's a red flag for bullshit. AFAIK no publicly disclosed technology can, in general, read emotions. Also, why whatever signal they are using as 'emotions' is a perfect predictor of insulin levels, I don't know.

I studied control engineering, and I bet he forgot some things. A few things come to mind. How does he recognize sensor failure (pretty important in closed-loop systems, like MCAS for example)? Did he use floating point math which may cause numerical issues? Does he use integrators and limit their output? If he does limit the output does he counteract windup? If he uses neural networks, did he gather enough data that we can be sure that this things works?

There's so many pitfalls here where teams of engineers with the whole part spec and from different disciplines have failed before. It sounds very dangerous.

Fuzzing seems like a really useful test for this.

Either way, I agree I'm not really subscribed to the idea that DIY life critical systems are a good idea beyond POC.

Maybe it isn't life critical but only quality of life critical?

Insulin pumps are both.

If it pumps too much you're dead - which is why I'm considering it life critical.

Not enough is also bad but hopefully they'll notice and can act accordingly - still a threat too though.

It's not that we're all saying this is an unsafe machine, we're saying that a random person DIY'ing and using such a solution is very risky, there's a reason why commercial medical equipment requires rigorous testing/certifications, etc.

Usage of cheaper slow acting insulin becomes viable if it is controlled by an algorithm that knows your insulin level.

You can already use slow-acting insulin, it just requires planning your day out really well, which is hard and why slow-acting stuff sucks. The amount of data required to predict someone's glucose consumption/insulin requirements is massive, and needs to come from that individual. Then you'd need to build in overrides because people change their habits/behave spontaneously which puts them in danger. It isn't a simple problem.

Reply on news.ycombinator.com