4 ms·

The wikipedia article seemed to indicate that these sorts of systems don't see much use outside of limited fields in medicine. Do you know why that is? It see

by ricree 18y ago

The wikipedia article seemed to indicate that these sorts of systems don't see much use outside of limited fields in medicine. Do you know why that is? It seems to me that the problem should be very similar to that faced by those 20 questions games, yet they seem to do fairly well on a wide variety of topics.

jerf 18y ago
The larger the domain, the weaker they get. Also, to differentiate diagnoses will frequently take tests and things that normal people don't have access to. A lot of things cause headaches, a lot of things cause nausea (damn near everything does, it seems like), etc. From an information theoretic point of view, it's hard to get enough bits of information from a normal person to diagnose anything. From a Bayesian point of view, it's hard to get enough evidence to diagnose a rare disease starting with low priors, but a system that only diagnoses common diseases is of limited value. In a limited domain it can do well, or even outperform a doctor (who is also limited by having a large domain, that problem is actually fundamental to the problem space not just the intelligence applied), in general it can't. I've half-self-diagnosed myself with Celiac disease (wikipedia article is the best on it I've found), which is in the running for one of the most under-diagnosed diseases currently known, precisely because the symptoms are relatively vague and quite variable ("Symptoms include chronic diarrhoea, failure to thrive (in children), and fatigue, but these may be absent and symptoms in all other organ systems have been described." - can it get any more vague? yes, but only barely.). This is an extreme, but there's enough medical conditions like this to make the general case very difficult to deal with.
- a-priori 18y ago
  Self-diagnosis is a whole other ball game, and it's a dangerous game to play. While a good expert system could help doctors significantly, in the hands of the general public it could also cause great harm. Self-diagnosis delays consultation with an actual expert. In addition, many symptoms (especially neurological ones) can only be seen or elicited by a trained third-party. PS: I have a friend with Coeliac disease, so I know something about it. It is hard-to-diagnose conclusively, but that's often not necessary. A key feature of the disease is that treatment, elimination of dietary wheat gluten, is harmless and usually causes drastic improvement. In my friends' case, it didn't, but that's an exception. I strongly recommend talking to a doctor and dietician about it.
  - jerf 18y ago
    I mean no disrespect, but that "half-self-diagnosed" isn't "self-diagnosed" for a reason. I don't need your approval of my decisions. I was simply using that as an example of something hard to diagnose. As you say, it's even hard to diagnose with medical resources, so it's not as if there's a choice. The whole "don't self-diagnose" speech is misplaced when diagnosis is not reliably available!
    - a-priori 18y ago
      I apologize for coming across harshly. I also didn't mean any disrespect.

by ricree 18y ago

jerf 18y ago

The larger the domain, the weaker they get. Also, to differentiate diagnoses will frequently take tests and things that normal people don't have access to. A lot of things cause headaches, a lot of things cause nausea (damn near everything does, it seems like), etc. From an information theoretic point of view, it's hard to get enough bits of information from a normal person to diagnose anything. From a Bayesian point of view, it's hard to get enough evidence to diagnose a rare disease starting with low priors, but a system that only diagnoses common diseases is of limited value. In a limited domain it can do well, or even outperform a doctor (who is also limited by having a large domain, that problem is actually fundamental to the problem space not just the intelligence applied), in general it can't. I've half-self-diagnosed myself with Celiac disease (wikipedia article is the best on it I've found), which is in the running for one of the most under-diagnosed diseases currently known, precisely because the symptoms are relatively vague and quite variable ("Symptoms include chronic diarrhoea, failure to thrive (in children), and fatigue, but these may be absent and symptoms in all other organ systems have been described." - can it get any more vague? yes, but only barely.). This is an extreme, but there's enough medical conditions like this to make the general case very difficult to deal with.

a-priori 18y ago
Self-diagnosis is a whole other ball game, and it's a dangerous game to play. While a good expert system could help doctors significantly, in the hands of the general public it could also cause great harm. Self-diagnosis delays consultation with an actual expert. In addition, many symptoms (especially neurological ones) can only be seen or elicited by a trained third-party. PS: I have a friend with Coeliac disease, so I know something about it. It is hard-to-diagnose conclusively, but that's often not necessary. A key feature of the disease is that treatment, elimination of dietary wheat gluten, is harmless and usually causes drastic improvement. In my friends' case, it didn't, but that's an exception. I strongly recommend talking to a doctor and dietician about it.
- jerf 18y ago
  I mean no disrespect, but that "half-self-diagnosed" isn't "self-diagnosed" for a reason. I don't need your approval of my decisions. I was simply using that as an example of something hard to diagnose. As you say, it's even hard to diagnose with medical resources, so it's not as if there's a choice. The whole "don't self-diagnose" speech is misplaced when diagnosis is not reliably available!
  - a-priori 18y ago
    I apologize for coming across harshly. I also didn't mean any disrespect.