11 ms·

How does one "train" an AI with a flood of random toilet pictures and no corresponding medical data to match it with?

imglorp 11mo ago
"potty training". Sorry. Anyway a chemical or biological sensor in the bowl might be more useful. Optical could be useful if it's doing spectrographic analysis: the color of poo and urine is sometimes informative.
hackernudes 11mo ago
They probably do clinical trials (or at least something like that) where they get baseline data from participants through other means.
- joezydeco 11mo ago
  I'm talking about sold units in the field.
  - themafia 11mo ago
    The same thing we always do. Pay some citizens of an African nation a pitiful wage to just make up annotations. Then you can incorporate this into a "health care product" and charge insurance companies insane rates on personal toilet cameras.
  - hackernudes 11mo ago
    I meant that they train their system on pictures where they have the underlying medical data. Their system might still be total crap (teehee), but I'm guessing that they at least try to make it predictive/generalized.
captainkrtek 11mo ago
I think the obvious things are: - Deviation in consistency/texture/color/etc. - Obvious signs related to the above (eg: diarrhea, dehydration, blood in stool). Ultimately though, you can get the same results by just looking down yourself and being curious if things look off... tldr: this feels like literal internet-of-shit IoT stuff.
g-b-r 11mo ago
They probably do match it, with data collected from other sources
venturecruelty 11mo ago
You pay someone in a developing nation $1.00 per day to look at thousands of photos of shit. Like, how do people think Facebook moderation and semantic labeling happen? Cheap labor in places with no labor laws. It was ever thus.
- xp84 11mo ago
  Appropriate username. And oh dear, that’s all too realistic. Imagine responding to the job posting and finding out these are the images you’ll be classifying.

"potty training". Sorry. Anyway a chemical or biological sensor in the bowl might be more useful. Optical could be useful if it's doing spectrographic analysis: the color of poo and urine is sometimes informative.

hackernudes 11mo ago

They probably do clinical trials (or at least something like that) where they get baseline data from participants through other means.

joezydeco 11mo ago
I'm talking about sold units in the field.
- themafia 11mo ago
  The same thing we always do. Pay some citizens of an African nation a pitiful wage to just make up annotations. Then you can incorporate this into a "health care product" and charge insurance companies insane rates on personal toilet cameras.
- hackernudes 11mo ago
  I meant that they train their system on pictures where they have the underlying medical data. Their system might still be total crap (teehee), but I'm guessing that they at least try to make it predictive/generalized.

captainkrtek 11mo ago

I think the obvious things are: - Deviation in consistency/texture/color/etc. - Obvious signs related to the above (eg: diarrhea, dehydration, blood in stool). Ultimately though, you can get the same results by just looking down yourself and being curious if things look off... tldr: this feels like literal internet-of-shit IoT stuff.

g-b-r 11mo ago

They probably do match it, with data collected from other sources

venturecruelty 11mo ago

You pay someone in a developing nation $1.00 per day to look at thousands of photos of shit. Like, how do people think Facebook moderation and semantic labeling happen? Cheap labor in places with no labor laws. It was ever thus.

xp84 11mo ago
Appropriate username. And oh dear, that’s all too realistic. Imagine responding to the job posting and finding out these are the images you’ll be classifying.