home bbs files messages ]

Forums before death by AOL, social media and spammers... "We can't have nice things"

   sci.electronics.repair      Fixing electronic equipment      124,925 messages   

[   << oldest   |   < older   |   list   |   newer >   |   newest >>   ]

   Message 123,788 of 124,925   
   Paul to micky   
   Re: Is it AI or not   
   10 Aug 23 15:47:39   
   
   XPost: alt.comp.os.windows-10, alt.home.repair   
   From: nospam@needed.invalid   
      
   On 8/10/2023 2:43 PM, micky wrote:   
   > No one in popular news talked about AI 6 months ago and all of sudden   
   > it's everywhere.   
   >   
   > The most recent discussion I heard was about "using AI to read X-rays   
   > and other medical imaging".   
   >   
   > They have computer programs that will "look" at, examine, x-rays etc.   
   > and find medical problems, sometimes ones that the radiologist misses.   
   >   
   > So it's good if both look them.   
   >   
   > But is it AI?   Seems to me it one slightly complicated algorith and   
   > comes nowhere close to AI.  The Turing test for example.   
   >   
   > And that lots of thigns they are calling AI these days are just slightly   
   > or moderately complicated computer programs, black boxes maybe, but not   
   > AI.   
   >   
   > What say you?   
   >   
      
   A radiologist assistant is not a Large Language Model.   
      
   I would expect to some extent, image analysis would be a   
   "module" on an LLM, and not a part of the main bit.   
      
   Bare minimum, it's a neural network, trained on images,   
   one at a time, that slosh around and train the neurons.   
      
   For example, something like YOLO_5 (You Only Look Once), can   
   be trained to identify animals in photos. It draws a box around   
   the presumed animal and names it (or whatever). That uses a lot   
   less hardware than a Large Language Model, and less storage.   
   The article had a picture with a bear in it, and indeed, the   
   bear had a square drawn around it.   
      
   But as for whether the "quality" is there, that is another   
   issue entirely. In my opinion, no radiologist would ever trust   
   something as sketchy as YOLO. Radiologists are very particular   
   about their jobs, as they hate getting sued. And I can imagine   
   the look on the judges face when you tell him "yer honor, I didn't   
   even bother to look at that film, the computer told me there was   
   nothing there". Some lawyers recently, learned about what happens   
   when you "phone it in". Professionals are still on the hook for the   
   whole bolt of goods. The computer isn't going to get sued for   
   "being stupid", because it is stupid.   
      
   It would take a *lot* of films, to train a radiologist assistant.   
   Who would have a collection, large enough for the job ? It would be   
   a violation of privacy law, for a bunch of hospitals to throw all   
   their films into a big vat, for NN training. It's not like crawling   
   the web and getting access to content that way.   
      
   While a lot of individuals and their jobs can be replaced,   
   the radiologist will be "the last to go". Regular doctors are   
   quite dependent on the radiologist taking the fall for mis-diagnosis.   
   The doctors would be scared shirtless, if the professional that   
   "has my back" was as stupid as a computer. The doctors would quit.   
   Doctors do not read films. They say stuff like "the radiologists   
   report says you have tits". And you can then take that to the bank.   
   They don't use their knowledge of Grays Anatomy to figure that out.   
   They identify the radiologist as the source of the information.   
   The radiologist is their "God".   
      
      Paul   
      
   --- SoupGate-Win32 v1.05   
    * Origin: you cannot sedate... all the things you hate (1:229/2)   

[   << oldest   |   < older   |   list   |   newer >   |   newest >>   ]


(c) 1994,  bbs@darkrealms.ca