Numbering The Good

The book Theory of Games and Economic Behavior, by John Von Neumann and Oskar Morgenstern, contains a formal mathematical theory of value. In the first part of the book they discuss some objections to such a project, as well as explaining why they are hopeful about it:

1.2.2. It is not that there exists any fundamental reason why mathematics should not be used in economics. The arguments often heard that because of the human element, of the psychological factors etc., or because there is allegedly no measurement of important factors, mathematics will find no application, can all be dismissed as utterly mistaken. Almost all these objections have been made, or might have been made, many centuries ago in fields where mathematics is now the chief instrument of analysis. This “might have been” is meant in the following sense: Let us try to imagine ourselves in the period which preceded the mathematical or almost mathematical phase of the development in physics, that is the 16th century, or in chemistry and biology, that is the 18th century. Taking for granted the skeptical attitude of those who object to mathematical economics in principle, the outlook in the physical and biological sciences at these early periods can hardly have been better than that in economics, mutatis mutandis, at present.

As to the lack of measurement of the most important factors, the example of the theory of heat is most instructive; before the development of the mathematical theory the possibilities of quantitative measurements were less favorable there than they are now in economics. The precise measurements of the quantity and quality of heat (energy and temperature) were the outcome and not the antecedents of the mathematical theory. This ought to be contrasted with the fact that the quantitative and exact notions of prices, money and the rate of interest were already developed centuries ago.

A further group of objections against quantitative measurements in economics, centers around the lack of indefinite divisibility of economic quantities. This is supposedly incompatible with the use of the infinitesimal calculus and hence (!) of mathematics. It is hard to see how such objections can be maintained in view of the atomic theories in physics and chemistry, the theory of quanta in electrodynamics, etc., and the notorious and continued success of mathematical analysis within these disciplines.

This project requires the possibility of treating the value of things as a numerically measurable quantity. Calling this value “utility”, they discuss the difficulty of this idea:

3.1.2. Historically, utility was first conceived as quantitatively measurable, i.e. as a number. Valid objections can be and have been made against this view in its original, naive form. It is clear that every measurement, or rather every claim of measurability, must ultimately be based on some immediate sensation, which possibly cannot and certainly need not be analyzed any further. In the case of utility the immediate sensation of preference, of one object or aggregate of objects as against another, provides this basis. But this permits us only to say when for one person one utility is greater than another. It is not in itself a basis for numerical comparison of utilities for one person nor of any comparison between different persons. Since there is no intuitively significant way to add two utilities for the same person, the assumption that utilities are of non-numerical character even seems plausible. The modern method of indifference curve analysis is a mathematical procedure to describe this situation.

They note however that the original situation was no different with the idea of quantitatively measuring heat:

3.2.1. All this is strongly reminiscent of the conditions existent at the beginning of the theory of heat: that too was based on the intuitively clear concept of one body feeling warmer than another, yet there was no immediate way to express significantly by how much, or how many times, or in what sense.

Beginning the derivation of their particular theory, they say:

3.3.2. Let us for the moment accept the picture of an individual whose system of preferences is all-embracing and complete, i.e. who, for any two objects or rather for any two imagined events, possesses a clear intuition of preference.

More precisely we expect him, for any two alternative events which are put before him as possibilities, to be able to tell which of the two he prefers.

It is a very natural extension of this picture to permit such an individual to compare not only events, but even combinations of events with stated probabilities.

By a combination of two events we mean this: Let the two events be denoted by B and C and use, for the sake of simplicity, the probability 50%-50%. Then the “combination” is the prospect of seeing B occur with a probability of 50% and (if B does not occur) C with the (remaining) probability of 50%. We stress that the two alternatives are mutually exclusive, so that no possibility of complementarity and the like exists. Also, that an absolute certainty of the occurrence of either B or C exists.

To restate our position. We expect the individual under consideration to possess a clear intuition whether he prefers the event A to the 50-50 combination of B or C, or conversely. It is clear that if he prefers A to B and also to C, then he will prefer it to the above combination as well; similarly, if he prefers B as well as C to A, then he will prefer the combination too. But if he should prefer A to, say B, but at the same time C to A, then any assertion about his preference of A against the combination contains fundamentally new information. Specifically: If he now prefers A to the 50-50 combination of B and C, this provides a plausible base for the numerical estimate that his preference of A over B is in excess of his preference of C over A.

If this standpoint is accepted, then there is a criterion with which to compare the preference of C over A with the preference of A over B. It is well known that thereby utilities, or rather differences of utilities, become numerically measurable. That the possibility of comparison between A, B, and C only to this extent is already sufficient for a numerical measurement of “distances” was first observed in economics by Pareto. Exactly the same argument has been made, however, by Euclid for the position of points on a line in fact it is the very basis of his classical derivation of numerical distances.

It is important to note that the the things being assigned values are described as events. They should not be considered to be actions or choices, or at any rate, only insofar as actions or choices are themselves events that happen in the world. This is important because a person might very well think, “It would be better if A happened than if B happened. But making A happen is vicious, while making B happen is virtuous, so I will make B happen.” He prefers A as an outcome, but the actions which cause these events do not line up, in their moral value, with the external value of the outcomes. Of course, just as the person says that A happening is a better outcome than B happening, he can say that “choosing to make B happen” is a better outcome than “choosing to make A happen.” So in this sense there is nothing to exclude actions from being included in this system of value. But they can only be included insofar as actions themselves are events that happen in the world.

Von Neumann and Morgenstern continue:

The introduction of numerical measures can be achieved even more directly if use is made of all possible probabilities. Indeed: Consider three events, C, A, B, for which the order of the individual’s preferences is the one stated. Let a be a real number between 0 and 1, such that A is exactly equally desirable with the combined event consisting of a chance of probability 1 – a for B and the remaining chance of probability a for C. Then we suggest the use of a as a numerical estimate for the ratio of the preference of A over B to that of C over B.

So for example, suppose that C is an orange (or as an event, eating an orange). is eating a plum, and is eating an apple. The person prefers the orange to the plum, and the plum to the apple. The person prefers a combination of a 20% chance of an apple and an 80% chance of an orange to a plum, while he prefers a plum to a combination of a 40% chance of an apple and a 60% chance of an orange. Since this indicates that his preference changes sides at some point, we suppose that this happens at a 30% chance of an apple and a 70% chance of an orange. All the combinations giving more than a 70% chance of the orange, he prefers to the plum; and he prefers the plum to all the combinations giving less than a 70% chance of the orange. The authors are suggesting that if we assign numerical values to the plum, the apple, and the orange, we should do this in such a way that the difference between the values of the plum and the apple, divided by the difference between the values of the orange and the apple, should be 0.7.

The basic intuition here is that since the combinations of various probabilities of the orange and apple vary continuously from (100% orange, 0% apple) to (0% orange, 100% apple), the various combinations should go continuously through every possible value between the value of the orange and the value of the apple. Since we are passing through those values by changing a probability, they are suggesting mapping that probability directly onto a value. Thus if the value of the orange is 1 and the value of the apple is 0, we say that the value of the plum is 0.7, because the plum is basically equivalent in value to a combination of a 70% chance of the orange and a 30% chance of the apple.

Working this out formally in the later parts of the paper, they show that given that a person’s preferences satisfy certain fairly reasonable axioms, it will be possible to assign values to each of his preferences, and these values are necessarily uniquely determined up to the point of a linear transformation.

I will not describe the axioms themselves here, although they are described in the book, as well as perhaps more simply elsewhere.

Note that according to this system, if you want to know the value of a combination, e.g. (60% chance of A and 40% chance of B), the value will always be 0.6(value of A)+0.4(value of B). The authors comment on this result:

3.7.1. At this point it may be well to stop and to reconsider the situation. Have we not shown too much? We can derive from the postulates (3:A)-(3:C) the numerical character of utility in the sense of (3:2:a) and (3:1:a), (3:1:b) in 3.5.1.; and (3:1:b) states that the numerical values of utility combine (with probabilities) like mathematical expectations! And yet the concept of mathematical expectation has been often questioned, and its legitimateness is certainly dependent upon some hypothesis concerning the nature of an “expectation.” Have we not then begged the question? Do not our postulates introduce, in some oblique way, the hypotheses which bring in the mathematical expectation?

More specifically: May there not exist in an individual a (positive or negative) utility of the mere act of “taking a chance,” of gambling, which the use of the mathematical expectation obliterates?

The objection is this: according to this system of value, if something has a value v, and something else has the double value 2v, the person should consider getting the thing with value v to be completely equal with a deal where he has an exactly 50% chance of getting the thing with value 2v, and a 50% chance of getting nothing. That seems objectionable because many people would prefer a certainty of getting something, to a situation where there is a good chance of getting nothing, even if there is also a chance of getting something more valuable. So for example, if you were now offered the choice of $100,000 directly, or $200,000 if you flip a coin and get heads, and nothing if you get tails, you would probably not only prefer the $100,000, but prefer it to a very high degree.

Morgenstern and Von Neumann continue:

How did our axioms (3:A)-(3:C) get around this possibility?

As far as we can see, our postulates (3:A)-(3:C) do not attempt to avoid it. Even that one which gets closest to excluding a “utility of gambling” (3:C:b) (cf. its discussion in 3.6.2.), seems to be plausible and legitimate, unless a much more refined system of psychology is used than the one now available for the purposes of economics. The fact that a numerical utility, with a formula amounting to the use of mathematical expectations, can be built upon (3:A)-(3:C), seems to indicate this: We have practically defined numerical utility as being that thing for which the calculus of mathematical expectations is legitimate. Since (3:A)-(3:C) secure that the necessary construction can be carried out, concepts like a “specific utility of gambling” cannot be formulated free of contradiction on this level.

“We have practically defined numerical utility as being that thing for which the calculus of mathematical expectations is legitimate.” In other words, the reason for the strange result is that calling a value “double” very nearly simply means that a 50% chance of that value, and a 50% chance of nothing, is considered equal to the original value which was to be doubled.

Considering the case of the $100,000 and $200,000, perhaps it is not so strange after all, even if we think of value in the terms of Von Neumann and Morgenstern. You are benefited if you receive $100,000. But if you receive $100,000, and then another $100,000, how much benefit do you get from the second gift? Just as much? Not at all. The first gift will almost certainly make a much bigger change in your life than the second gift. So even by ordinary standards, getting $200,000 is not twice as valuable as getting $100,000, but less than twice as valuable.

There might be something such that it would have exactly twice the value of $100,000 for you in the Von Neumann-Morgenstern sense. If you care about money enough, perhaps $300,000, or $1,000,000. If so, then you would consider the deal where you flip a coin for this amount of money just as good (considered in advance) as directly receiving $100,000. If you don’t care enough about money for such a thing to be true, there will be something else that you do consider to have twice the value, or more, in this sense. For example, if you have a brother dying of cancer, you would probably prefer that he have a 50% chance of survival, to receiving the $100,000. This means that in the relevant sense, you consider the survival of your brother to have more than double the value of $100,000.

This system of value does not in fact prevent one from assigning a “specific utility of gambling,” even within the system, as long as the fact that I am gambling or not is considered as a distinct event which is an additional result. If the only value that matters is money, then it is indeed a contradiction to speak of a specific utility of gambling. But if I care both about money and about whether I am gambling or not, there is no contradiction.

Something else is implied by all of this, something which is frequently not noticed. Suppose you have a choice of two events in this way. One of them is something that you would want or would like, as small or big as you like. It could be having a nice day at the beach, or $100, or whatever you please. The other is a deal where you have a virtual certainty of getting nothing, and a very small probability of some extremely large reward. For example, it may be that your brother dying of cancer is also on the road to hell. The second event is to give your brother a chance of one in a googolplex of attaining eternal salvation.

Of course, the second event here is worthless. Nobody is going to do anything or give up anything for the sake of such a deal. What this implies is this: if a numerical value is assigned to something in the Von Neumann-Morgenstern manner, no matter what that thing is, that value must be low enough (in comparison to other values) that it won’t have any significant value after it is divided by a googolplex.

In other words, even eternal salvation does not have an infinite value, but a finite value (measured in this way), and low enough that it can be made worthless by enough division.

If we consider the value to express how much we care about something, then this actually makes intuitive sense, because we do not care infinitely about anything, not even about things which might be themselves infinite.

Pascal, in his wager, assumes a probability of 50% for God and for the truth of religious beliefs, and seems to assume a certainty of salvation, given that you accept those beliefs and that they happen to be true. He also seems to assume a certain loss of salvation, if you do not accept those beliefs and they happen to be true, and that nothing in particular will happen if the beliefs are not true.

These assumptions are not very reasonable, considered as actual probability assignments and actual expectations of what is going to happen. However, some set of assignments will be reasonable, and this will certainly affect the reasonableness of the wager. If the probability of success is too low, the wager will be unreasonable, just as above we noted that it would be unreasonable to accept the deal concerning your brother. On the other hand, if the probability of success is high enough, it may well be reasonable to take the deal.

Advertisements

3 thoughts on “Numbering The Good

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out / Change )

Twitter picture

You are commenting using your Twitter account. Log Out / Change )

Facebook photo

You are commenting using your Facebook account. Log Out / Change )

Google+ photo

You are commenting using your Google+ account. Log Out / Change )

Connecting to %s