Link Search Menu Expand Document

Medical Diagnoses and Arguments of Support (Writing Practice)

Date: 2020-08-24

Speed writing exercise, goal for splits: 3 min, 10 min, 2 min

Note: I asked ET about this in tutorial 31, and in response to some of his feedback sort of outlined/summarised it in a follow post to FI. Here is my outline:

Two philosophers: Aro and El. Aro believes in arguments of support, El doesn’t.

I label explanations with capital letters, A..Z. The explanations are diagnoses in this case. We eliminate them via evidence that contradicts the explanation.

There are 25 explanations under consideration, A..Y.

Situation 1; (case 1) eliminate Y, A..X remain. Aro and El basically agree on things, here.

Situation 1; (case 2) eliminate A..X, Y remains. Aro and El basically agree on things, here, too.

Situation 1.1.: Start with the second case in Situation 1 and introduce a new option, Z.

What do Aro and El think given Z is introduced?

El doesn’t think much of it; he’s just not done when he otherwise might be.

Aro needs to think about whether the same evidence still supports Y, to what extent, and how much it supports Z, etc.

Aro needs to re-evaluate the evidence’s support for Y and Z.

Evidence’s relationship to Y changes due to the introduction of Z. The epistemic status of that supporting relationship changes w/ the introduction of Z. So the epistemic status of supporting relationships for Y depend on other ideas you know. Those other ideas are not refutations of Y, however, they influence how much Y is supported. So Y’s support from some evidence is not dependent on the evidence as much as it is on alternatives to Y (which are things we can make up, almost at will).

The mistake here (leading to the need to recalc the amount of support) is different from other mistakes. With other mistakes: the correction can both be decisive and involve learning something new.

With supporting arguments, the mistake of incorrectly judging support is one where:

  • the correction is not decisive
  • we don’t ever learn something new
    • (note: a new option like Z is what causes one to realise their estimate of support is wrong, so it can’t be a thing we learn as a result of fixing that)

When irrelevant* new knowledge was introduced, Aro has to change his measurement of the amount of support an option, Y, had, but there was no associated error correction in a substantial way. For any idea that gets introduced he has to do that. But shouldn’t the relationship between evidence and ideas be something we can make enduring, objectively reasonable statements about?

[*]: Irrelevant to the evidence<->idea supporting relationship

A way of resolving this issue is to find equilibrium points in one’s measurement of support. These points should be such that: the amount of support provided by some evidence for some idea doesn’t change as not-directly-relevant ideas are introduced.

The only such equilibrium point for supporting arguments/evidence is 0. Therefore arguments/evidence always provide zero support. This is equivalent to the idea that supporting arguments don’t exist. This is a contradiction with the premise that they do exist. Therefore they don’t; QED.


  • diagnosis has lots of options (hypotheses)
  • symptoms eliminate options
  • reactions to stuff eliminate opts
  • you can narrow down by testing
  • testing a thing might seem to “support” a diagnosis,
  • but what it really does is provide more evidence that eliminates other options, even if you don’t know exactly what the alts are
    • just knowing there are alts is enough (and if they’re eliminated) – can work on classes of theories
  • house - ‘you’re treating with X looking for support for a diagnosis of Y’
  • when you come up with an idea to test
  • if a test or reaction or evidence would be particularly unexpected we feel like it supports some particular idea
  • but that’s just b/c lots of OTHER things, more ordinary things, eliminated
  • so like ‘antivenom making you get better supports snake-bite’
  • but antivenom making you get better and not having a snake-bite doesn’t make sense, so lots of other stuff eliminated
  • how does the idea of diagnosing make it more obvious arguments of support don’t exist
    • b/c
      • evidence (results) are theory laden in 2 ways - diff evidence supports different stuff depending on your hypothesis of what it could be, e.g. ibuprofen reducing inflammation - if you try that b/c you think it might be X and it gets better, all you’ve done is eliminate all the A,B,Cs that wouldn’t get better with that treatment, not eliminated any Ys or Zs that also would get better w/ the treatment
        • but evidence against results are theory laden in the normal way only - understanding of human body
        • arguments of support are either meaningless b/c they don’t eliminate any options, or irrelevant b/c they are compatible with all the options (e.g. X,Y,Z)
  • imagine you’re a doctor and having trouble diagnosing a patient
  • there are 1000 possible things but you eliminate 900 immediately b/c they are way different from initial conditions
  • come up with tests for lots of the 100 remaining by comparing and finding places to look for points of difference - eliminate 75
  • have to progressively eliminate the remainder
  • what happens with “supporting” arguments

(total time planning: 10 min)

how does diagnosing a medical condition explain that arguments of support don’t exist?

Pretend for a moment (if you have to) that you’re a general doctor and trying to diagnose a patient - your goal is to find exactly one explanation for the patient’s condition, and have no reasons it couldn’t be that diagnosis. When the patient walks in the present with a set of symptoms you can see plainly. You know very quickly that you can eliminate most medical conditions; they don’t fit the symptoms, so you can eliminate 900 of the 1000 possible medical issues. Technically you should include non-medical issues too, but we won’t worry about that now.

Of the 100 options left, you eliminate 75 of them with ‘easy’ tests – you can compare all 100 options, loosely and quickly in your mind, to find some tests that will necessarily contradict lots of options. So you do some things like checking the patients throat or something. But now you’re stuck; there are 25 conditions left and you don’t have easy things to try. You need to do more cumbersome tests, or get more data/evidence in some way.

Let’s consider arguments of support from the PoV of 2 philosophers: Aro and El. (new-school, arguments of support don’t exist). Aro thinks arguments of support exist, and that support can be quantified on the interval [0, 1]. El thinks arguments of support don’t exist; equivalent to the belief that all support is quantified as exactly 0.

Situation 1

Let’s say there is one option out of the 25 (call it Y) which is common (like the flu), and (Fact 1:) there is some treatment for it which does not have side-effects – regardless of whether the patient has the condition or not. Furthermore, (Fact 2:) this treatment does not help any of the other possible conditions (call them A through X). You decide to administer this treatment. What happens in each of the cases?

If the patient does not get better then we’ve eliminated Y, so we have 24 options left. This “seems” to support all other options A..X, but we’ve also reduced the size of the set of options from 25 to 24. I don’t think Aro will have an issue accepting that this outcome doesn’t really “support” any of A..X. If they did, esp if they tried to calculate it, the “amount” of support would be low. Aro’s going to die on that hill. El doesn’t need to do anything, this is fine and expected.

If the patient does get better then we have not eliminated Y, however, we have eliminated A..X – did we find supporting evidence for Y? Let’s call this ‘getting better’ part Evidence 1 or E1 for short. This is the sort of situation where Aro will claim that E1 supports Y. El will claim E1 does not support Y, it just refutes A..X. However, this situation doesn’t help us determine if Aro or El are wrong; because in both cases they arrive at the same result unambiguously.

Situation 1.1

Let’s modify situation 1 as follows:

  • exactly 1 of the initial 900 options eliminated was done so for a particular reason – call this option Z.
  • (Fact 3:) Z is rare and especially in someone like the patient who is outside the normal age bracket, which is why it was eliminated as a candidate, but the symptoms otherwise match.
  • (Fact 4:) when someone with Z is given the treatment you expect the condition to get better.

So, you give the patient the treatment and they start to get better; and you soon realise that Z could be an option and actually shouldn’t have been eliminated. What now happens with Aro and El?

Let’s presume Aro and El agree that option Z was always wrongly eliminated, but it’s understandable considering how exceedingly rare it would be for Z to be the cause (perhaps a world first) – what will they think next?

El doesn’t have much of a problem here, E1 never supported Y so there’s no need to make new conclusions, we just aren’t done yet. El’s theory is entirely compatible with reality.

Aro does have a problem though: does E1 still support Y? Does it support Z too? (Logically, it must.) Do the amounts of support change? Aro has some thinking to do.

How does the introduction of new knowledge change things? Well, it doesn’t change anything about reality - the same options were always available and the same causal forces are at work, etc. It doesn’t change anything in El’s mind besides adding the new knowledge. It does change a lot in Aro’s mind, though, because Aro needs to retroactively re-evaluate E1’s support for Y and judge E1’s support for Z too.

Aro thinks arguments of support exist thus Y must still be supported by E1. However, the introduction of new knowledge (that Z is an option) changes E1’s relationship to Y (call that supporting relationship RS1). This means RS1 cannot be an objectively sound thing – it’s epistemic status changes depending on new knowledge, so that epistemic status wasn’t objectively true. Consider the relationship where evidence refutes an option: this type of relationship (removing/excluding options) does not change with new knowledge (unless that knowledge is about a mistake you made, which is fine, but not an important case here).

Aro’s need to recalculate is not a mistake like most other mistakes - here Aro is treating RS1 as an epistemic truth, but it needs to be changed for any relevant new knowledge. El doesn’t need to recalculate anything; the normal process is fine to continue.

Even though Aro realised something wasn’t objectively true, the solution was just to update RS1 instead of asking why RS1 needed to be updated. When Aro thinks about why RS1 needs to be updated the conclusion is that the need was due to the introduction of new knowledge. But we’ll always be creating and introducing new knowledge (sometimes we can’t even tell if it’s relevant or not straight away, which technically makes it relevant). Aro thinks the mistake was in the measuring (the amount of support), but the method for calculating that support hasn’t changed; neither has Aro changed any epistemic beliefs (besides those about RS1).

Aro thinks about this, and decides one approach might be to try and factor in new knowledge. Aro thinks he’ll be right if - for any new knowledge introduced - his updated estimate of the “amount” of support doesn’t change, regardless of what new knowledge is introduced and in which order. That is, RS1’s “support” should be constant regardless of the set of current knowledge and the set of new knowledge that will or could be created. Aro knows the set of new knowledge that will or could be created is infinite. So the only logical conclusion is that the “amount” of support must be 0 because that’s the only way it’ll be constant in all future cases. Aro doesn’t know how to deal with this contradiction – he arrived at the same position as El!

El doesn’t have any issues with the above events; everything fits in neatly.


Since Aro came to the same conclusion as El started with, it either means the contradiction needs to be explained or Aro was wrong. Arguments of support do not exist (or, equivalently, arguments always provide exactly 0 support).


While the diagnosing aspect didn’t end up playing as significant a role as I thought it would, I think the argument is good, so I’m leaving it as it is. Also, I lost 5 min to a discord reply (which really wasn’t necessary, but I got distracted anyway).


planning 00:09:55
writing 01:06:19
editing: 00:00:57

Well, it took longer than I expected.

You can leave a comment anonymously. No sign up or login is required. Use a junk email if not your own; email is only for notifications—though, FYI, I will be able to see it.

Comments powered by Talkyard.