Hacker News Reader

Comment by tomrod | Hacker News Reader

Utility is a fundamental concept in decision theory. Arguing it cannot exist is the opposite extreme to arguing that homo economicus, with superhuman evaluative strategies and no gaps in rationality, exists.

People can compare their current situations to relative improvement, and often those are transitive (though not always). So in most reasonable cases the mathematical axioms needed for utility to be defined exist in a reasonable way, allowing for comparison through a formalized utility function.

One can certainly mathurbate themselves with utility, and many do, but ultimately utility is a discussion to simplify communication (instead of primitives) about why people make preditible choices. It extends pretty quickly to revealed preferences.

Of course, asking where preferences come from in the first place is a third rail.

View on HN

>Utility is a fundamental concept in decision theory. Arguing it cannot exist is the opposite extreme to arguing that homo economicus, with superhuman evaluative strategies and no gaps in rationality, exists.

It's probably more like arguing that leprechauns don't exist.

The burden of proof [for its existance] is on those making up those "fundamental concepts in decision theory". I won't be taking it for granted just because they came up with it.

In any case, I'm not saying utility can't exist. I'm saying some universal utility can't exist, or if you wish: sorry, guys, you can't determine my utility function for me. I'll do it myself, thank you very much.

>So in most reasonable cases the mathematical axioms needed for utility to be defined exist in a reasonable way, allowing for comparison through a formalized utility function.

If we could have a "formalized utility function" for "most reasonable cases" we'd hardly have different morals, political parties, and so on...

It's mostly irrelevant (trivial) cases that have formalized utility functions. Everything else is political, that is up for debate based on interests, preferences, morals, and so on -- and especially based on idiosyncrasy.

Even maximing one's life/health is not some constant. Many prefer to smoke, drink, eat, knowing fully well it might have them, because their utility function favors enjoyment over life span. Others might sacrifice their life for some cause or another.

And yet, the theory of utility works well under extensions of bounded rationality, with full acknowledgement of time inconsistent preferences.

Utimately utility is a simplifying model of human behavior.

> we'd hardly have different morals, political parties, and so on.

This touches on where preferences come from, which utility theory is mainly silent on.

> you can't determine my utility function for me. I'll do it myself, thank you very much

You have now stated that utility exists.

At no place should any argument depend on some universal utility function that applies to everyone. The idea of a universal utility function is a useful simplification for some beginner classes, but quickly becomes useless for anything in the real world. In fact if there were a universal utility function(s?) economics wouldn't be hard to study, just a simple optimization problem that businesses would use.

All that we need is for everyone to assign utility in some way. It doesn't matter if your function omits critical factors, applies the wrong weighting, or otherwise is a decision you come to regret. (note that this hindsight might be wrong because you don't really know what your regrets would be had you made the other decision) All that matters is at some point you weight all the factors you consider important and make a decision based on them. You can come up with a complex formula to put numbers to it, or just go with a "gut feeling" (in many cases others are involved - perhaps a spouse). Regardless you have made a utility function for your situation.

> You have now stated that utility exists.

Only in the context of a single person. The argument presented in the article, as well as ideas like the "utility monster" are based on the idea that the utility scales of different persons are comparable.

This is not the same thing as a universal utility function, but almost as outlandish.

> utility scales of different persons are comparable.

but a decision is only made by one person, so only that person's utility function matters. A different person, using their own utility function, would come to a different conclusion and make a different choice.

So while there's no universal utility function, it doesn't matter as long as the decision maker's utility function exists (and it does, by tautological argument). In the article, the utility values of 80 and 40 for the boys are the outcome of the parent's utility function. The boys don't get a choice, and so their utility functions don't matter.

That's fair. The article uses phrasing like "the disabled boy has a total utility of 40", suggesting that the utility is an attribute of the boy, but I guess it would become wordy and repetitive to phrase it any other way.

Is the concept of utility then anything other than tautological? If I understand what you're saying, it's roughly that, "A person chooses things, and since I imagine that their choice process can be caricatured as a linearized ranking system, a utility measure must exist for them".

I'm not saying that's a necessarily false model. But it strikes me as such a crashingly unsubtle simplification that I'd want to see a ton of data demonstrating that's really how it works. As opposed to just being something that academics assume so they can write bold, confident papers with conclusions that they like.

Von-Neumann Morgenstern utility is mathematically precise; there is a real-numbered utility function such that maximizing utility is equivalent to choosing the correct lotteries according to an agent's preferences. So long as every decision one can have a preference about can be stated as a preference over expected outcomes (e.g. 50% chance of ice cream over 30% chance of cake, or related to the article: 90% child-one succeeds and 70% child-two succeeds v.s. 85% child-one succeeds and 74% child-two succeeds) then the utility function exists.

Humans do not have utility functions. We have a lot of circular or contradictory preferences and other ancient machinery in our brains, and especially we do not reason about probabilities and expected outcomes accurately enough. We might be able to grow into having a utility function while still being happy about our preferences and without changing our humanity for the worse.

> So while there's no universal utility function, it doesn't matter as long as the decision maker's utility function exists (and it does, by tautological argument). In the article, the utility values of 80 and 40 for the boys are the outcome of the parent's utility function.

Yes, but ... the argument in the article presupposes fixed differences in utility when ascribing choices to priority, equality, and pure utilitarian views. Is it not easier to just say that the parent values improving the situation of the disabled boy more, and thus the utility of this improvement to the disabled boy's situation is higher in one parent's view but not the other?

Nearly any parent will choose a massive benefit from son A's perspective at the cost of a tiny expense to son B (looks utilitarian!). Nearly any parent will prioritize a sibling who is less well off in some circumstances. Nearly any parent will give the two sons equal slices of cake when they value them equally. But is it not easier to ascribe different utilities to these different circumstances instead of different allocation functions?

> The argument presented in the article, as well as ideas like the "utility monster" are based on the idea that the utility scales of different persons are comparable.

The article makes no implicit or explicit statement about how one defines a/the utility function, but I see no reason to believe the author thinks it's a universal function.

As indicated in the sentence directly following your quote.

>You have now stated that utility exists.

I never said it doesn't. In fact I opened this thread by writing "The very idea that there is some measusable "utility" to compare in the two cases, independent from your moral values and sentiments, is inane".

That is, it's the independent from the person, measurable, utility (as per the example in TFA/theory) that I called BS.

>but quickly becomes useless for anything in the real world

My sentiments exactly.

>All that matters is at some point you weight all the factors you consider important and make a decision based on them. You can come up with a complex formula to put numbers to it, or just go with a "gut feeling" (in many cases others are involved - perhaps a spouse). Regardless you have made a utility function for your situation.

Sure. But none of this has much to do with the theory as presented in the post's examples...

> The burden of proof [for its existance] is on those making up those "fundamental concepts in decision theory". I won't be taking it for granted just because they came up with it.

What? It's trivial. People want things. Things that satisfy wants have utility. You can just look this up. It's pretty basic to modern economics and philosophy.[1]

1. https://en.wikipedia.org/wiki/Utility

You're just defining satisfaction of wants as utility. "Things that satisfy wants have utility" is a statement of a definition, not a synthetic claim. There are plenty of measures of utility other than hedonic ones (or volitional ones, or whatever exactly your definition is specifying).

Yes, because that is what's meant by the word in this context.

I'm aware, but my point is that you're not making a synthetic claim - you're not proving the (axiological) meaningfulness of a concept. You're just saying "I use this word 'utility' to describe the satisfaction of wants".

It doesn't really answer any of the questions that were posed, about how you can measure and compare the 'want-satisfying-ness' of different things. How do you measure the degree of want? How do you measure the degree to which a want is satisfied? How do you compare those across human beings?

If by 'trivial' in your original comment you meant 'trivial' in the technical sense[0], then I'd agree with that. "I define 'utility' as 'satisfaction of wants'" is a statement that neither predicates nor proves anything of the world.

[0] https://en.wikipedia.org/wiki/Triviality_(mathematics)

It does answer the questions by reformulating them in exactly the way you did, which immediately highlights the GPs confusion: They somehow missed the subjective and relative aspects of the concept as it is used. There is no objective and absolute measure of want-satisfying-ness (or whatever).

> It's probably more like arguing that leprechauns don't exist.

> In any case, I'm not saying utility can't exist. I'm saying some universal utility can't exist, or if you wish: sorry, guys, you can't determine my utility function for me. I'll do it myself, thank you very much.

This is what I'm responding to.

You call it trivial, but the second sentence says, "Its usage has evolved significantly over time"; both can't be true. And the criticism section makes some good points: https://en.wikipedia.org/wiki/Utility#Discussion_and_critici...

I also think there are a number of questionable assumptions behind it, and even your simple version of a supposedly trivial concept doesn't match the current official definition well.

So to me this isn't so much an obvious fact about the world as a synthetic cornerstone to a worldview. Sort of like Peano's construction of the integers, or the way theists talk about the things that are "pretty basic" to their religion. Those things feel trivial to their adherents, of course. But the rest of us can find sweeping dismissals like yours as very offputting.

"In any case, I'm not saying utility can't exist. I'm saying some universal utility can't exist, or if you wish: sorry, guys, you can't determine my utility function for me. I'll do it myself, thank you very much."

The article does not posit a "universal utility function", nor does it require determining anyone's individual utility function. It merely requires that utility functions exist and that a comparison of the resulting utility is meaningful.

"If we could have a "formalized utility function" for "most reasonable cases" we'd hardly have different morals, political parties, and so on..."

Only if you have a universal utility function. Which you are the only one proposing.

Right, but if the essence of the question is by what function you measure utility, then the question as posed by the article is a moot point. utility, priority, equality - they're just slightly different cost functions for the utility. And it's not even the case that they're well-defined and clearly separated; some level of interpretation is going to be required regardless.

For example, people routinely act as if money has a non-linear utility; we'll insure ourselves against stuff partially because being destitute is worse than the mere loss of money might suggest; i.e. each additional dollar is worth less.

But exactly how you define those non-linear relationships, especially once you include stuff like happiness, health, and intend to aggregate over multiple individuals is clearly tricky, and it's not reasonable to expect any one simplified model to work well in all situations in reality. It's not even reasonable for that to be knowable or computable.

So it's both perfectly reasonable to consider it ludicrous to label one such scenario as having "40" and "80" utility without having had the critical discussion of what that utility is measuring, while also conceding that the concept of utility is reasonable and... sometimes... enlightening.

Thanks, this was a really insightful comment (as someone who spent years of my life getting a graduate philosophy degree, before doing something more 'useful'). I think the concept of utility is clearly, uh, useful, and the reason that it's aversive to people is that they tend to bundle it up with a lot of the (sociologically, not logically) related views, which tend to be more problematic.

Hedonic utilitarianism in particular turns a lot of people off, and partly for good reason. I'm deeply ambivalent about it, and I think the surrounding debates, and the assumed primacy of moral intuition in applied cases, are far harder and more open questions than most people reckon. But I can still see how examples like utility monsters, or gang rape being morally superior to garden-variety rape because there are more people to enjoy it, might make people feel like it's really on the wrong path.

Those examples are hilariously egregious, yeah! It's slightly taboo in polite conversation to see increased utility there, yep. Thanks for the kind words, too.

I disagree. OP is correct. Utility is a rhetorical construct. Sometimes it works well for describing morality, decisions, etc. Sometimes it's crammed in.

Using it to describe the two sons decisions is cramming it in.

> Utility is a rhetorical construct

It's a mathematical construct to show ranking of points in a topological space. With two simple axioms, comparability and transitivity, it is fairly well defined mathematically, though it typically enters the extrapolation zone at extremes.

But if the definition is barely more fleshed out than "some cost function", then the difference between utility, priority and equality as discussed in the article collapse; they're all the same thing. Cost functions, aka: utility.

Cost functions are typically the dual for utility maximization under cost constraints.

You are correct in that they are usually mathematically equivalent.

Yeah, I was struggling the find the best terminology for that, but given the ambiguity of the term "utility" in this context I though it better to avoid that ;-).

I'm certain nobody argues with you on that point. The question is: what does the fact that numbers can be ordered entail for ethics?

I’ve always heard utility in the context of a utility function. Basically:

f(u) = Wx*x + … + Wz*z, where x and z are variables that are impacted by decisions and constraints. Each variable is weighted for importance by the person / group using the utility function.

So for a home buyer needing to get to a city, the utility of the house improves as the location to the city gets better, subject to the constraint that it’s not in the river. A home buyer utility function might also weight cost, neighbors, amenities, square footage, local pollution, safety, and any other meaningful variable for the buyer.

Turning this into a quantitative formula can be cramming it in and quite hand-wavy, but ultimately it’s up to the person optimizing for their own utility to put in the variables and weights. These will be shifted by the person’s moral code (e.g. A Jewish person may highly value living in the city’s Eruv).

On a political note, big government supporters believe the federal government can define a utility function for the country that is best for the greater good. People who believe in smaller federal government and governing at the local level believe the utility functions should be defined at the individual level if possible - subject to the constraints one does not infringe on others’ rights. There are benefits to both sides (some things we can’t achieve if everyone acts independently, some things create externalities, some things have too many edge cases and unintended consequences).

I think the extreme of a shared utility function is communism, with an idea of central planning.

If the math of utility is interesting to you, check out Hal Varian's microeconomics book (he is/was chief economist at Google) or the intro grad text for microeconometrics Mas-Collel, Winston, and Green.

Utility theory's primitives are defined before the actual function.

Social choice theory is covered in MWG -- arrow's impossibility theorem is absolutely fascinating!

The field of mechanism design relies heavily on utility theory -- it's effectively the inverse of game theory, or, how to structure systems and incentives to get desired outcomes.

Yes, utility as a concept in decision theory is great, but it’s not the same as the concept by the same name from utilitarianism. In my understanding, coldtea is doubting not decision theoretic utility functions, but utilities as they are used here.

Most importantly a decision theoretic utility function is only defined up to any positive definite transformation. Inter-agent comparisons like "Imagine that the gifted boy has a total utility of 80, and the disabled boy has a total utility of 40” don’t make any sense in terms of decision theoretic utility functions.

I don't think GP is arguing that utility doesn't exist. I believe the GP is arguing that the OP is making arguments as if utility weren't subjective.

If you can attach a number to decisions, you can just do the math. The thing is, attaching a number to make a non meta argument about decision making can be bollocks since the actual utility can be -9999999 for me or 9999999 for you. An utility function is a function of the decision making agent

See the "independent from your moral values and sentiments, is inane" bit

Utility can be intrepret in many ways. Look at the social sciences and how they see minorities, e.g. disabled people. From the social scientist's view, one "utility" of these people is that they stabilize societies because they trigger empathy, which would otherwise be largely missing in a society that only aims for optimization. I want to emphasize that I find it generally humilitating to talk about utility and humans in one and the same sentence.

think of it as a label for the process by which you decide to prioritize cleaning up different areas of your house -- and utility becomes a rational and humanizing thing.

using it to prioritize your relations forces you to grapple with subjective and irrational things such as personal prefs, aspirations etc. so.. also rational and humanizing.

this leaves using it to mess with others without them participating in weight-setting (democracy as a weight discovery mechanism?), that's where it gets messy.

i fail to see where any of the three facets above make it humiliating to see where 'by priority' the most relative improvement can be made. i mean, this is all just fine talk about something innate to nature, no?

If someone has a special preference for egalitarian outcomes, this should be included in their utility functions.

Telling someone their utility values for each of the choices is equivalent to telling them their preference. Asking for their preference afterwards is pointless, they have already been told their preference.

Reply on news.ycombinator.com