If you look at my blogger homepage, I list what my interests are in making this blog. One of them is epistemology. That is the study of how we know what we know. The biggest boost to my own epistemology, something that cleared up how I know what I know, is Bayes Theorem (BT).
Because BT is a simple formulation of how the scientific method is supposed to work, I should be accused of “scientism”. But since most people who accuse people of scientism don’t know about BT, they don’t know to level that charge against me. But I don’t even think I’m a proponent of just scientism. As Eliezer Yudkowsky says, [scientism] isn’t strict enough, bayesianism is stricter. As a bayesian, I’m much more critical of “other ways of knowing” than mere scientism.
What is the most important part of BT? If you read my most recent post on starting with different priors you’ll know that the most important part of BT is Bayes Factor, or dividing the conditional probabilities. This shows you how strongly or weakly the evidence favors or disfavors your hypothesis. For any binary test (true or false), the conditionals can be called the Success Rate (SR) and the False Positive Rate (FPR). This is real important to note, since it’s a blow against “other ways of knowing”.
At work, I’ve helped develop a program that can automatically generate100% valid messages according to a military standard. Other software developers use this program to test their systems. Now, this program was started before I became a project lead, so my input for how the program would work didn’t start until earlier this year.
In the language of BT, this means that the SR of this program is 100%. But what about the FPR? This program doesn’t automatically generate invalid messages, so it is actually of very little use to other software engineers who want to test their systems to make sure they are correctly validating messages. It’s actually very easy to make a system that will successfully validate 100% of the valid messages it receives: Just make it say that every message it receives is valid!
What would really make this automatic message generating application strong is being able to test both the SR and the FPR. This would help to ensure that other programs are successfully validating actual valid messages and correctly identifying invalid messages. So not only should the SR of this program be 100%, but the FPR of the program should be pretty close to zero. So for this next development cycle, I snuck in some BT language in the requirement 😉
If this isn’t clear enough, I’ll use a more hypothetical example. Let’s say that the TSA implements some new computer system that identifies 100% of the trial-terrorists during development. If this system is put into the field and identifies someone as a terrorist, does this mean that they actually are a terrorist? No. You would have to also know the system’s FPR (and also the prior probability / base rate) in order to guage how likely it is that a person caught is a terrorist. If the FPR for this TSA system is 100% or close to 100%, then this is either very weak or no evidence the person caught is a terrorist.
That actually seems to be how the current TSA system works, so it’s not as hypothetical as I thought…
Anyway, what about these so-called “other ways of knowing”? What is the success rate of these other ways of knowing? The false positive rate? If some other way of knowing had a 100% SR does this mean that you actually know something? No. It would be just like my automatic message generator.
One way that proponents of other ways of knowing would attempt to get out of it would be to claim that BT doesn’t apply. But that would be like Christian apologists claiming logic doesn’t apply when talking about the existence of god. Sure, maybe logic doesn’t apply, but then you wouldn’t be able to talk about god in any meaningful way. Apologists in this case are confusing ontology with epistemology: Logic is a function of epistemology, not ontology. The less something can fit into a logical framework, the less we can know and understand about it. Since BT is just an extention of the laws of logic the same argument would be applicable. The less something fits in with the laws of probability (BT), the less we can know about it.
But there’s another angle for where the charges of scientism, and appeals to “other ways of knowing”, might be coming from:
“Traditional Rationality” [i.e. scientism] refers to the memes of rationality which are in general circulation. It is good enough to get Science done, but lacks the technical precision of Bayesian probability. Unsharpened by curiosity, it demands only the least effort needed to push forward human knowledge.
Traditional Rationality is phrased in terms of social rules, with violations interpretable as cheating – as defections from cooperative norms. If you want me to accept a belief from you, you are obligated to provide me with a certain amount of evidence. If you try to get out of it, we all know you’re cheating on your obligation. A theory is obligated to make bold predictions for itself, not just steal predictions that other theories have labored to make. A theory is obligated to expose itself to falsification – if it tries to duck out, that’s like trying to duck out of a fearsome initiation ritual; you must pay your dues.
This, I would bet, is where the majority of charges of scientism come from. They are talking about the social rules of the scientific method. You can violate social rules (because this means you are no longer participating in that social ritual), but you can’t violate the rules of probability. Well you can, but if you are willing to put money on breaking those rules, you will lose money. This also seems to explain another phenomenon: That of religious scientists. Again, they consider the scientific method a social contract, only applicable when “doing science”. Because it is only a social contract, we have Non-Overlapping Magisteria between religion and science; religion is one social setting, doing science is yet another.
So I’ll make a bold statement: There aren’t any “other ways of knowing”. The only ways of knowing that we have are according to the laws of logic and the laws of probability. Break those rules, and you simply don’t know what you claim to know.