Economics and epistemology

There was some discussion of epistemology and economics on x a while ago. I'm going to discuss epistemology and how it is related to economics.

Epistemology is about knowledge: what constitutes knowledge and how it can be created and assessed. Knowledge is often discussed in terms of a philosophical definition: "justified true belief". This definition smuggles in a substantive theory of knowledge that is wrong in almost every important respect but is also accepted by almost everybody including many economists.

Knowledge isn't belief

The idea that knowledge is belief is common but falls apart as soon as you think seriously about almost any real example of knowledge. Consider almost any good book in the entire history of the human species, such as Capitalism by George Reisman. Consider a particular sentence in that book, such as this sentence on p. 1:

Procapitalist economic thought and economic thought as such are essentially synonymous.

The author chose each word in that sentence, the order of the words and the punctuation. He had some reason for each choice he made. He may or may not remember the reason for each choice. And the same is true for each sentence in the book. Those choices were made to convey the author's ideas So every sentence includes knowledge of how to convey the author's ideas. There is no particular reason to think that any human being currently knows the exact reason for all of those choices. So the book instantiates knowledge that nobody believes. So we have an example of knowledge that nobody believes and as such knowledge can't be belief.

There are many other examples. For example, if you write a computer program, you will have some reason for choosing each character you type in the files that constitute the program, but you may not remember the reason for every choice. In addition, before the program is run a compiler or interpreter will turn the contents of those files into lower level computer instructions. There may be several layers of such instructions. No human being will know the complete contents of all those lower layers but they all include some knowledge that makes the program work, such as checking the types of variables and that sort of thing.

Knowledge isn't belief, it is information that solves some problem. This broadens the scope of knowledge in some ways. There is information in the genes of some birds about how to fly in terms of behaviour, materials, physiology and anatomy. Human beings don't know everything about this information for all birds partly because we don't know all that information about all of the known species and also because there may be undiscovered species of bird.

The alternative to broadening the scope in this way is that you will end up making distinctions that don't have anything to do with explaining any feature of reality, such as whether somebody happens to recall a particular piece of information.

Justification and truth

Justification is supposedly a process that leads to an idea being true or good, or to have a greater measure of truth or goodness. A rational person will want his ideas to be true and so will want to justify them, or at least that is the story told by people who advocate justification: justificationists. But justificationism is actually dangerous nonsense.

Knowledge includes assumptions about some aspect of reality and rules that give you correct conclusions when applied to those assumptions. For example, there are well known rules of addition and if you apply those rules to adding up apples in different boxes in a market stall you will arrive at the correct number of apples. So if we count the apples in each box and apply the rules of addition to those numbers we have justified our knowledge of the total number of apples, right? Wrong.

There are various subtle and not so subtle problems with this story. Your input to the rules for adding up apples could be wrong. You could miscount the apples. You could also make a mistake in applying the rules of arithmetic. It is common for people to make arithmetical mistakes. And if you put the numbers into a computer to add them up that can also go wrong in various ways. You might store the number of apples as a u8 in the rust programming language, but a u8 can only hold numbers from 0 to 255 so if you have more than 255 apples you might have a problem. Some people might buy apples from the stall so that after you have counted the apples and added them up you get the wrong answer for the number of apples in the stall. Or apples could be taken by animals. And if you wait for long enough the apples will rot so you may no longer be able to distinguish the apples from one another and there won't be a well defined number of apples.

At the end of your attempt to add up the apples you will be either be right or wrong and there is no way to guarantee you are right. And since you are either right or wrong there is no probability of you being right other than 0 or 1, nor is there any other measure of rightness. You might try to define a probability or some other measure over the set of all occasions on which you counted apples, but since no particular count is guaranteed to be correct all you're doing at most is checking some measure of consistency of different estimates.

You could get more than one person to calculate the number of apples or use more than one technique for counting them and then compare the results. And you could correct flaws in your apple counting technique by counting the apples before the start of the business day so people won't buy them. But no set of steps you take will guarantee the correctness of the total.

There is another issue that is more subtle. There are two kinds of problems involved in justification: the problem of the correctness of your assumptions and the problem of the correctness of the rules about how to get from those assumptions to true conclusions. If you count the number of apples in the stall on a particular day and time you may get the correct number with whatever rules you use. But the rules you use for the counting are supposed to apply to some range of possible situations and there is no way of trying to apply those rules and checking them in all possible situations. For example, if there are 5406 apples on the stall at 5:30am on June 7 2025 and your technique gave the correct number of apples, the technique might have given the wrong answer if there were 5407 apples or if you had counted them ten minutes later or under some other situation that didn't happen to occur now, but could have occurred or will occur in the future or did occur in the past. So if you are in the business of trying to work out whether rules are correct that is flatly impossible. And if you can't guarantee the correctness of the rules then you can't guarantee their correctness in any given situation in reality.

There is also no such thing as the probability that some set of rules is correct. The rules are supposed to apply to all situations fitting some set of circumstances, e.g. - when you manage to count all the apples in the stall before they rot or are stolen by rats and so on. But if the rule is supposed to apply to all of those situations then it either works or it doesn't. And if you want to define a measure of goodness on sets of rules you get into problems of explaining what set of rules you're using and why. And how do you justify the rule you're using to specify a set of rules so you can justify them? This appears to lead to an infinite regress.

One common response to criticisms of justificationism is that we have knowledge so justification must be possible and this argument must be wrong somehow. But this presupposes that justification is required to create and assess knowledge. There are unrefuted non-justificationist accounts of epistemology such as the one I will give in the next section.

Epistemology without justification

Let's go back to the stall apple counting example. Suppose you have a two apple counting methods X and Y and somebody points out a situation in which X and Y produce different apple counts. At this point something has gone wrong with your apple counting knowledge.

One possibility is that X is wrong. If you have an explanation of why X would be wrong you can look at the consequences of that explanation in some other situation and check whether it holds. If you are holding the number of apples in a u8 in a computer and you have more than 255 apples, then your counting rule will fail to work for any situation in which there are more than 255 items to count. So if the stall has more than 255 pears you could count the pears to see if you get the wrong answer. Also there is an explanation of how u8 numbers are stored in a computer and if you understand that explanation you might not feel the need to check.

Another possibility is that your standard for judging apple counting rules is wrong. It might be the case that any method of counting apples you would actually want to use will produce numbers that have some level of inaccuracy. But the cost of eliminating that inaccuracy would be greater than anything you could hope to gain by doing the calculation correctly so you relax your apple counting standards.

The pattern used to settle this argument is to eliminate ideas by looking for flaws in those ideas. This allows you to judge ideas without justification. The standard by which you judge an idea is whether it has a known flaw for solving the problem you're interested in solving. If your idea doesn't have a known flaw then it meets this standard and justification or lack thereof is irrelevant. This was pointed out by Karl Popper who called his epistemological ideas critical rationalism.

People often object to critical rationalism by saying it is unjustified. By the standards of critical rationalism this is irrelevant because CR claims that justification is impossible. There are no known unanswered criticisms of CR and that is enough for CR to survive by its own standards. Justificationism doesn't meet its own standards because there is no justification for justificationism and there are objections to justificationism outlined above to seeking justification.

I should also note that criticism need not take the form of observations or experimental results. The argument above doesn't take that form and there are many other arguments that don't take that form. For example, theism isn't experimentally testable but it can still be criticised and refuted.

Criticisms of other Friedman's epistemology

Milton Friedman is well regarded by many people who describe themselves as free market economists. I think that criticisms of Friedman's economics from free market people more closely aligned with Austrian economics are correct in substance. Friedman's epistemology, as explained in his essay "The Methodology of Positive Economics" is also bad in ways that I haven't seen criticised elsewhere.

In Section I Friedman contrasts positive economics quoting Keynes as "systematized knowledge concerning what is" with normative economics "a system of rules for the attainment of a given end". Friedman argues that people often confuse them and that many normative issues could be settled by coming to agreement on positive issues.

In Section II Friedman starts explaining his epistemology. He writes:

The ultimate goal of a positive science is the development of a “theory” or “hypothesis” that yields valid and meaningful (i.e., not truistic) predictions about phenomena not yet observed. Such a theory is, in general, a complex intermixture of two elements. In part, it is a “language” designed to promote “systematic and organized methods of reasoning.” In part, it is a body of substantive hypotheses designed to abstract essential features of complex reality.

In general there is no tenable distinction between a language used for organized reasoning and a substantive hypothesis. For example, the Marxists may want to describe economics in terms of socially necessary labour hours to produce commodities. Austrian economists would correctly maintain that this entire description is nonsense. What counts as a description depends to some extent on what theory you're using. Between different branches of Marxism socially necessary labour hours to produce a sofa may be an uncontroversial description. In a dispute between an Austrian economist and a Marxist that description can't be taken for granted.

Friedman claims that economic theories can't be proved because there may be some circumstance we haven't discovered that doesn't match the theory. Friedman claims that in the social sciences it is rarely possible to conduct controlled experiments. But in practice situations often arise that can be used to test economic theories although this usually involves subtle and involved chains of reasoning that people often find unconvincing.

Friedman also claims that evidence is often used to construct and to test economic theories. He claims that the assumptions of a theory don't have to be realistic descriptions of how the world works. Rather, those assumptions are used to motivate a model and what matters is whether the model survives testing. Friedman claims the assumptions have to be unrealistic because they will neglect many other features of reality. Whether the assumption holds should be judged by the model's success not by their realism or lack thereof.

This may sound similar to my outline of critical rationalism in the previous part, but the resemblance is superficial and a lot of what Friedman wrote is wrong. One of the problems is that Friedman's position is instrumentalist: he claims to judge ideas solely by empirical success or failure not by judging ideas as accounts of what is happening in reality. This makes no sense when considered carefully.

Suppose that I look at economic data like the exchange rate between various currencies over several decades. I then pick a bunch of cases in which the exchange rate between currency A and currency B was 1.5 for various times and different currencies, e.g. - the pound and the dollar in April 1978, the yen and the euro in 2005 and so on, and say that my theory is that the exchange rate between currencies is 1.5. I can then test this theory by picking other times and combinations of currencies and selecting those for which the exchange rate was 1.5 and declare victory. You may be tempted to say this is a ludicrous way of proceeding, but how could Friedman do that? He has said the realism of assumptions doesn't matter. Now, Friedman gives the following standards for the language in which a hypothesis is cast:

Are the categories clearly and precisely defined? Are they exhaustive? Do we know where to file each individual item, or is there considerable ambiguity? Is the system of headings and subheadings so designed that we can quickly find an item we want, or must we hunt from place to place? Are the items we shall want to consider jointly filed together? Does the filing system avoid elaborate cross-references?

Does this save Friedman's position? No. I can come up with a rationale for why my rule fits each of these standards. The items I want to consider jointly, those in which the exchange rate of two currencies is 1.5, are filed together. I could go on, but I won't. In reality, Friedman has some standards of explanation by which he judges theories. Other people have different standards by which the consider his theories unrealistic, e.g. - the criticisms of Friedman's position by Austrian economists. Since such standards are controversial they should be explained and discussed. Friedman's epistemology tries to shortcut such discussions and as a result it doesn't make sense.

Praxeology and Austrian epistemology

There is a branch of the Austrian school of economics that discusses economics as a branch of a more general theory of human action called praxeology as explained in the introduction and first chapter of Mises' book Human Action (HA). Human action is purposeful behaviour that applies means to attain some ends. Some behaviour is non-purposeful such as falling to the ground during an epileptic fit. A person can also say that he will do something like go to the opera when in reality he intends to stay home and listen to kpop. What matters for praxeology if what a person actually does, not what he claims he will do.

Mises held that historical phenomena involve multiple causal influences and so those can't be used to verify or test theories about human action. Rather praxeology is a priori: "necessary and ineluctable intellectual conditions of thinking" (HA, Chapter II, Section 2, p.93). He continues to say (p.96):

The fact that man does not have the creative power to imagine categories at variance with the fundamental logical relations and with the principles of causality and teleology enjoins upon us what may be called *methodological apriorism*.

Everybody in his daily behavior again and again bears witness to the immutability and universality of the categories of thought and action. He who addresses his fellow men, who wants to inform and convince them, who asks questions and answers other people’s questions, can proceed in this way only because he can appeal to something common to all men—namely, the logical structure of human reason. The idea that A could at the same time be non-A or that to prefer A to B could at the same time be to prefer B to A is simply inconceivable and absurd to a human mind. We are not in the position to comprehend any kind of prelogical or metalogical thinking. We cannot think of a world without causality and teleology.

Mises writes that praxeological knowledge is as certain as knowledge of maths (HA, pp. 101-102):

The real thing which is the subject matter of praxeology, human action, stems from the same source as human reasoning. Action and reason are congeneric and homogeneous; they may even be called two different aspects of the same thing. That reason has the power to make clear through pure ratiocination the essential features of action is a consequence of the fact that action is an offshoot of reason. The theorems attained by correct praxeological reasoning are not only perfectly certain and incontestable, like the correct mathematical theorems. They refer, moreover with the full rigidity of their apodictic certainty and incontestability to the reality of action as it appears in life and history. Praxeology conveys exact and precise knowledge of real things.

The word "correct" in "correct praxeological reasoning" and "correct mathematical theorems" is doing a lot of heavy lifting and it doesn't actually bear the load.

Mathematical results are sometimes disputed. It is famously true that when Andrew Wiles claimed to have a proof of Fermat's last theorem, his first attempt at a proof was refuted. He later came up with a new proof that hasn't been refuted. Wiles thought his first proof was correct and he was wrong. Mathematicians think his second proof is correct but they might all be wrong. What we actually know is that there no existing unanswered criticisms of the proof despite mathematicians looking for it. The proof describes and elaborates an argument about various branches of mathematics being connected in a way that is relevant to Fermat's last theorem. A mathematical proof is an argument of this kind that has no unanswered criticisms that mathematicians know of. For more explanation of this issue see Proofs and Refutations by Imre Lakatos.

What about praxeology? There are Austrian economists in the praxeological tradition who make conflicting claims about praxeology. For example, there is a dispute about interest theory between Robert Murphy and Jeff Herbener about whether the pure time preference theory of interest is correct and consistent with Austrian economics and praxeology and as far as I can tell this is an unresolved issue. Trying to understand this controversy in the light of claims that Austrian economics is just working out the consequences of indubitable premises doesn't make much sense. In reality, Austrian economists are looking for problems that arise from their existing theories and events in the real world and trying to come up with solutions to those problems. Sometimes this will look like working out the consequences of previous ideas but often it won't and there is no particular reason to think it will in any given case. When previous ideas don't solve the problem some new idea will be needed, which may or may not be consistent with previous ideas.

Economic knowledge, like other kinds of knowledge, is created by guesswork controlled by criticism. That criticism doesn't take the form of experimental testing or observation. It takes the form of thinking about what is happening in reality according to different economic theories and whether those claims make sense. On the level of the claims being made, Austrian economics makes sense to a much greater extent than its competition. Austrian economics solves problems not solved by other traditions and there are no known criticisms of its claims about how people make economic decisions.

For example, most economic theories treat capital, producer's goods, as a homogenous lump where you can just increase the amount and things will get better. As a result, many economists advocate stimulating the economy with government spending when there is a recession. But this doesn't make sense. If businesses are going under that's because people don't want to pay for what they are producing. Why would you divert toward paying for goods people don't want? Resources may remain unused for some time during a recession but that's because those resources were being used wrongly and it takes time to figure out how to employ them better. If people want fewer cars and more sandwiches, you can't just use a car production line to make more sandwiches. The government shovelling fiat money into the economy won't fix that problem.

Epistemology is about what you should do if you want to create knowledge. Economics is about what you should do if you want to create wealth. Neither of them is experimentally testable for similar reasons: you can't tell what would have happened if you did something better than what you're doing now. Knowledge or goods would have been created that you can't foresee because if you could have foreseen them you would already have them. they are also closely related because the creation of knowledge is required to run a successful business: knowledge about what to make, how to explain to people why they should buy a product and how to deliver it to them. In addition, goods are required to produce new knowledge and the incentives of the institutions of knowledge creation, such as universities, affect what kinds of knowledge they produce, or fail to produce.

Austrian economists and critical rationalists could learn from one another and solve problems that concern both traditions. Many Austrian economists at this point will be thinking that Hoppe refuted Popper's ideas, of which more here.