Hacker News new | past | comments | ask | show | jobs | submit login
Ask HN: Would a theranos-style device be possible using recent AI breakthroughs?
2 points by bobosha on Feb 5, 2023 | hide | past | favorite | 7 comments



No. Clever AI algorithms cannot manufacture signal out of noise. The fundamental problem is that the Theranos device relied on small samples containing only a few drops of capillary blood mixed with interstitial fluid extracted from just under the skin. Those samples are not representative of arterial blood.

Such devices can work well enough for certain assays that only look for positive / negative results like antibody screening. That's known chemistry and doesn't require any AI.


As others have said, no, because what they wanted to do was not physically possible.

But what part of recent AI breakthroughs would plausibly be able to help improve an ultra low volume blood test anyway? A more life-like chatbot? Being able to draw a picture based on a description? What possible connection is there? We're right back to peak AI hype again where people just completely imaging stuff and assume that AI is the answer, without the slightest understanding of what it actually is

(Edit: given the bio of the poster, I'm even more confused about why they asked this question)


Hi, yes I am sorry I should have disclosed I work in the field of AI. But my question was posed to more biomedical scientists if lots more can be done with less using AI-based techniques.


>if lots more can be done with less using AI

I think a more realistic challenge for AI is to do almost as much as can be accomplished using natural intelligence.

Then sensibly progress to doing "just as much" in those cases where that's actually within reason.

Before you can correctly identify those opportunities where an AI solution can not be beat.

Instead of trying to do more with less, I think one of the strong points of AI could instead be reaching the desired outcome based on overwhelmingly more data, not less. Regardless of how much data and how many lower-intelligence helper apps are needed to keep everything in line.

Remember with a human programmer you're going to get human-based logic as the foundation of everything they do even though that includes human fallability.

With a machine you start out with its own kind of inhuman unsuitability, which really does take a lot to be carefully molded into humanly recognizable functions.

The bold promise of AI would be if top decision-making performance can be achieved, and for blood testing I would think you would want to work with vast amounts of data from a wide variety of FDA-approved equipment, to make progress toward insightful diagnosis. That might be one of the most intelligent high-stakes challenges of all. I would expect some leading doctors are involved in this already.

I don't think trying to get deeper insights from lesser material would be as promising.


Not sure about the bio stuff, but AI could be used very effectively to modulate a woman's voice so that it sounds deeper and more manly. This would allow a company to raise more venture capital.


What is this device you're talking about?


Probably something that can do extensive biomedical testing on very little bio matter. The answer is no, the problems with Theranos weren't technical, they were physical, the box they tried to build was impossible based on fluid dynamics and the viscosity of blood.

The fraud part came in when they kept hoping for a miracle and sold the thing to Walgreens before it even existed




Consider applying for YC's W25 batch! Applications are open till Nov 12.

Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: