I agree that Occam’s razor a very important heuristic, perhaps even with normative force, in scientific modeling. I agree with this, and we formalize this mathematically many ways, and rely up on it reasoning too.
However, I assert that Occam’s razor is almost guaranteed to misguide us in understanding reality itself. This seems to be clear from even an atheist scientist point of view. It is not an appeal to God or the supernatural, but an insistence that Occam’s razor is merely a useful convenience for modeling. Without a heretofore unarticulated limiting principle, it leads to epistemological absurdity.
Occam’s razor (also Ockham’s razor or Ocham’s razor ; Latin: lex parsimoniae " law of parsimony ") is the problem-solving principle that the simplest solution tends to be the right one. When presented with competing hypotheses to solve a problem, one should select the solution with the fewest assumptions https://en.wikipedia.org/wiki/Occam’s_razor
Who says it has no connection to reality? As with any heuristic, it isn’t perfect. The simplest hypothesis that fits all current data may be overturned by future data, though this becomes less likely as data accumulate without contradiction. Still, we would be unable to function without some intuitive equivalent. If you want to make a sandwich, and you find what appears to be a jar of peanut butter in your pantry, you should suppose it’s the jar you bought last Thursday rather than a clever substitute placed there by demons that actually contains steel ball bearings elaborately camouflaged as peanut butter. I can’t think of any way to distinguish between the two hypotheses by empirical test, because the ball bearings are disguised so well. So what would you do? Make the sandwich anyway?
I’m pretty sure I agree. The way I’d put it is that Occam’s razor helps us develop good hypotheses and models, however, it’s not that great at actually determining if the hypothesis/model is correct. This is plainly true because we actually have to actually do experiments …
Another common ways to phrase Occam’s razor is that a theory should be no more complex than necessary. The trick of course is that we most often don’t know what’s necessary until after we’ve done all the work of testing the theory. So what often happens is we start with a simple theory and keep getting more complex as needed to explain the data.
In general chemistry classes we do this with bonding theories. I teach freshman science majors 4 different bonding theories from Lewis theory as the simplest up to molecular orbital theory as the most complex. No chemist would say Lewis bond theory is the best because it is the simplest (but Occam’s razor!), because it only gives an accurate picture of some properties (what’s bonded to what) in a limited set of cases.
There are several problems with the razor. Much of this can be summed up in the admonition: absence of evidence is not evidence of absence.
Most directly, there is no guarantee that the simplest equation is actually the true explanation of any given set of data. A great example of this in your field, phylogeny, is reversion mutations. They might happen, but they will not usually appear in a parsimonious tree. In another example, looking my CV you could build a mathematical model of my collaborations and members of my group over time. You would not need to posit any members of my group that have not yet published to explain the data. This, however, does not mean that there are not in-fact members of my group that have not (or never had) published a paper with me. They are sliced out by the razor, even though they do in fact exist.
Just as important, every inquiry has a bounded scope of inquiry, perhaps at times defined its dataset. Expand that scope, and what is simple in one domain may not be sufficient as the scope is increased. Very often more data requires more complexity. For example, the syllogism fails: “I do not need to @John_Harshman to explain the metabolism data I’m studying, therefore @John_Harshman most likely does not exist.” It better to understand this as, “I do not need to @John_Harshman to explain the metabolism data I’m studying, therefore I’m not including @John_Harshman in my model.” That shift is important one, which recognizes the value of the razor in modeling, while also remembering it is poor way of determining truth.
We know things exist that leave no evidence of their existence. For example, we know that every person has parents. I have no evidence or records of who are my great-great-great-great grandparents. I do not need to posit their existence to explain any evidence, because I have no evidence concerning them. However, we can and should infer that they did in fact exist. A great example of how this works is coalescence simulation (and inference), where we just ignore most individuals in a population in the past. For explaining the data, this is the most parsimonious model. However, it clearly does not match with reality, because it does not include all the contemporaries of an ancestral sequence.
There are beliefs that are best understood as proper basic beliefs. This recent Veritas Forum article explains this well: http://www.veritas.org/positivism-burden-proof/ . These are true beliefs that are not formed by simplest explanation of the evidence.
The criteria of “most simple” is not well-defined, and is subjective in many important cases. What is more simple, God or the multiverse? There is no effective way of settling that question.
Occam’s Razor requires a tradeoff deciding fit to data and model complexity that is subjective, and even shifting with context.
Here is a good article that goes into this in relations with science:
Never liked that buzzphrase. Because absence of evidence often is evidence of absence. If you know that given X you expect to see Y, and you don’t see Y, that actually is good evidence that there is no X.
Of course not. That’s why it’s a heuristic. But I don’t think your example works very well; in most cases, if we see two states that look identical by descent, they are identical by descent. Or to put it another way, the contribution to the likelihood of seeing A and A at adjacent nodes of 0 changes is almost always very much greater than the contribution of 2 or more changes. Given enough sites, this will be wrong some of the time, but it’s still the way to bet.
I see your example here as deeply flawed, because the problems with the syllogism have nothing to do with Occam’s razor but with the poor formulation of the conclusion. A proper formulation would be “therefore John Harshman had no effect on the metabolism data”. And that would be a correct conclusion. I also presume you meant to say "therefore I am not including John Harshman in my model. I would say that your lack of a need to include me in your model would most plausibly be explained by me having no influence on the data. Occam wins.
I don’t see this as having anything to do with Occam’s razor either.
I will have to read the article, which I haven’t yet done. But I don’t yet see how the existence of basic beliefs is relevant to Occam’s razor. It doesn’t have to be applicable to everything in order to be applicable within its proper scope.
True. But in the example case, this comes from the ambiguity of “God”.
If it was observed that the hypothesis required to explain phenomenon became increasingly simpler as more data is uncovered, then your statement would be true.
However, this is not the case in the vast majority of phenomenon. As more data is uncovered, the explanatory hypothesis becomes increasingly complex.
Hence Parsimony is not a principle supported by real life experience.The final explanation is almost always far more complex/unparsimonius than the first explanation.
This is true. However, what you are explaining seems more a bayesian process than simple parsimony. The reason people dont assume that demons didnt substitute the peanut butter is because over decades,its their experience that such things dont happen.
If they lived in a universe where demons regularly substituted peanut butter with something else, they would suspect the same. This kind of thinking is more Byesian than parsimonius.