Jessica writes:
I get so tired of people dumping on decision theory because real world decisions are complex. If decision theory is so deeply flawed, I’d love to know what alternative methods the critics advise for trying to evaluate and improve decision making in some real world setting. Should we give up on modeling completely because some cause problems for our assumptions? What happened to the epistemic value of attempting to formalize goals so as to better understand what components we think are at play? Do we really want to go back to talking about man as a creature of instinct and habit and leave it at that?
I agree, and this reminds me of a discussion from twenty years ago (!) about the transition from viewing people as “rational animals” to viewing people as “irrational computers.”
Here’s Thomas Jefferson from 1823:
We believed . . . that man was a rational animal, endowed by nature with rights, and with an innate sense of justice; and that he could be restrained from wrong and protected in right, by moderate powers, confided to persons of his own choice, and held to their duties by dependence on his own will.
He’s coming from a liberal (in the U.S. politics) perspective, with the idea being that rationality is a way to move forward from outmoded feudal arrangements. Not that this was so easy–Jefferson owned slaves!–, but nobody said that rationality was easy, just that it’s a way forward.
This association, in which the left was associated with utopian rationality and the right was associated with sensible acceptance of irrationality, continued for another century. Consider, for example, the contrast between the rationalist and socialist George Barnard Shaw and the Catholic conservative G. K. Chesterton. This association of rationality with the left continued through the New Deal period in the U.S. and the idea of the Soviet Union as being scientifically socialist. The second world war pitted Soviet central planning and “Fordist” American organization against the blood-and-soil Axis powers.
Sometime during the mid-cold-war period there was a shift, at least in the U.S. and its allies, where science and technology was associated with the military-industrial complex and gained a conservative tinge, while the left embraced an anti-technology, back-to-the-land vision. “Humanism” moved from a conservative, roll-back-the-tide, Chestertonian position to a liberal, fight-the-Man position.
Nowadays things are a mess: conservatives support military and police hardware, coal, nuclear power, bitcoin, data centers, and gas guzzlers more generally, but conservatives also oppose vaccines and scientific more research more generally, and Biblical creationism hasn’t gone away either. And, with conservatives in charge of the country and much of public discourse, liberals are often defining themselves based on what they oppose.
I’m with Jessica in that I see no conflict between humanism and rationality. Rationality is an ideal or a way of being, not an algorithm. Yes, we’re animals, and rationality is one of our very useful tricks. I wouldn’t want to abandon rationality or define ourselves against it, any more than I’d want to abandon running or singing or any of the other things that we can do so well, when we do them well.
Confusing post to me: you seem to equate rationality and decision theory, but these don’t seem the same to me. I see decision theory as encompassing differing views of how decisions are, or should be, made. One of these is rationality, although that term is far from uniquely defined. Several theories of “irrational” human behavior are forms of decision theory in my view. So, I don’t see equating the two.
I also have issues with the initial quoted paragraph from Jessica (perhaps the full context addresses some of this, I don’t know). I feel like I can dump on plenty of versions of decision theory without declaring the whole field worthless. Decisions and humans are complex, and all theories must simplify this reality by parsing the important parts of reality and simplifying the less important parts. Since this involves judgements, there will always be disputes about any given theory. It isn’t clear to me who rejects the idea of decision theory in its entirety. Andrew’s comments are focused on an unspecified version of “rationality” and Jessica’s seem focused on an unspecified rejection of any model of decision making. I’m sure there are examples of what each is objecting to, but it is not obvious to me how widespread these are. I don’t think these apply to all critiques of particular forms of decision theory or rationality – but they sound like they are. I think it would be clearer if the terms were more precisely defined. What particular decision theory or what particular definition of rationality is being defended?
This occurred to me as well. The only way I can think of rational vs irrational decisions at all would be, for example, those that are driven by some logical calculus, and those that can be characterized in say, emotion (I think about those in finance, some of whom use model-based and fundamental analysis of markets to make investments, vs those who buy-in some thing due to ‘vibe’, ‘fomo’, etc etc).
At the base level question though. Who is scatting on decision theory? In fact sequential decision analysis as a policy level way of thinking about RL in ML (as well as for humans) seems to have gained quite a bit of traction vs ‘decision theory’ often previously employed in industry that can only be described in relatively vague terms and even more nebulous methods.
Reminds me of this story:
“Some years ago I was trying to decide whether or not to move to Harvard from Stanford. I had bored my friends silly with endless discussions. Finally, one of them said, “You’re one of our leading decision theorists. Maybe you should make a list of costs and benefits and try to roughly calculate your expected utility.” Without thinking, I blurted out, “Come on, Sandy, this is serious.”” — Persi Diaconis, “The Problem of Thinking Too Much”, pp. 36–37
My actual opinion is something closer to this:
“This is a fairly distinct topic from actual human decision-making, since people do not seem to conform very well to any of the theoretical ideals. This sometimes leads to much wailing and gnashing of teeth over our irrationality; if anything, however, it leads me to doubt that these theories are good formalizations of rationality.”
https://bactra.org/notebooks/decision-theory.html
As for alternatives: I don’t need any to conclude the same as Shalizi, that they are not good formalizations of how humans behave. We simply don’t have a good theory at this moment. Maybe you could look at Herbert Simon’s work or Gigerenzer’s for promising directions.
This is an analogous situation to the Tukey quote “The combination of some data and an aching desire for an answer does not ensure that a reasonable answer can be extracted from a given body of data.”
Anon:
Serious decision analysis exists; we give some examples in chapter 9 of BDA3. But, yeah, stupid theoretical decision analysis is useless. The point of the above quote is not that decision analysis is not helpful for real decisions; the point is that the person making the decision was using the stupid stuff.
Or, to put it even more simply, GIGO.
Every time decision analysis comes up, I feel I should plug “Analysis of Local Decisions Using Hierarchical Modeling, Applied to Home Radon Measurement and Remediation” (Lin, Gelman, Price, Krantz). https://sites.stat.columbia.edu/gelman/research/published/lin.pdf
Among papers I’ve co-authored, this is one of my five favorites. I like to think it makes Andrew’s top 40, down there in the twenties or thirties somewhere.
The specific decision we considered is very simple, and even so the formal decision analysis gets pretty complicated. But ours is a real-world example and we really did come up with a useful analysis, in the sense that, if I were in a position to make policy decisions for the government, I would hew pretty closely to what we lay out in the paper. At the same time, you can see how a more complicated real-world problem would be too complicated for this kind of treatment.
For decades, many people (including me) have suggested that even if you aren’t going to do a full decision analysis it can be very helpful to act as if you’re going to for as long as you can: figure out what model structure you would use, what data you would need, which parameters you know precisely and which you don’t, and so on. Take it as far as you can before it becomes untenable. This can help highlight what the really important issues are, and which parameters it would really help to nail down better, even if you have to give up on the idea of collapsing a dozen different parameters into one utility function that you truly believe in.
Phil’s perspective is well-aligned with my own: rather than throwing our hands up and declaring statistical decision theory useless because people are irrational, we can often get a much better read on what we’re trying to do and what we’d need to do it well if we try to formalize the problem we think we’re solving.
What I’m complaining about in that passage is a response I see a lot (e.g. from ML reviewers) which is knee jerk skepticism that anything can be accomplished by comparing human behavior to rational ideals. It usually comes with zero suggestion of how to proceed instead, probably because it’s from people who have never tried to design an informative experiment or method involving people. It holds back entire research areas from learning much because they remain mired in basic measurement issues.
@Jessica – I think what I’m trying to understand is, ‘who is slinging the mud’ at statistical decision theory methods? This schism isn’t something of which I’m entirely aware, but will be looking for info on this now.
Kind of a footnote wrt Andrew’s sidebar on Jessica’s OQ: the fact that we can even loosely classify ‘rational’ vs ‘irrational’ decisions at all tends, in my mind as well hopefully others, that we’re at the very least dealing with complex systems (with NLDS types of behavior over time). In such a case, abstractions like ‘rational’ v ‘irrational’ become relatively meaningless given a specific decision is an event, or sequence of them, rather than an adjective.
If, rather, we consider a decision more fundamentally a ‘discrete action that is a stochastic response to one or more variable inputs – *whatever and however structured they may be* – then what even is the sense of going down the path of imposing binary abstractions like ‘rational’ or ‘irrational’ on decisions (this is a decision too – kinda meta amirite?)?. Just too open for bias and misinterpretation
_(Quick ex: in Weberian terms, ‘rational’ doesn’t imply ‘ethical’. A ‘rational’ person may indeed cheat/free-ride, if that calculus is likely to generate some favorable outcome for this type of rational actor. Felt the need to point this out)._
Social behaviorists (including economists) and psychologists I’m guessing each have their own internalized semantics for defining and categorizing decisions and their meanings.
That’s not to say there aren’t common threads between definitions/categories. But I think certain rigorous distinctions must be agreed upon when we’re in the business of analyzing this stuff.
I really should read up on this. I feel out of touch with current academic drama.
Fischel:
We’ve discussed this distinction between selfishness and rationality. It’s an important point that has confused a lot of people.
Jessica:
Here is what you should do instead: use Gerd Gigerenzer’s heuristics to make models of how people actually make decisions. Then use those.
See https://direct.mit.edu/books/monograph/5001/Classification-in-the-WildThe-Science-and-Art-of
Phil’s paper does seem pretty good, but why is it considered decision theory? It does not mention utility functions at all. Maybe it is what decision theory should be, reading off your specific threshold value from the graph.
Anon:
You write, “Phil’s paper does seem pretty good, but why is it considered decision theory? It does not mention utility functions at all.”
Follow the link! “Decisions” is right in the title. It’s an entire paper about decision making under uncertainty. The utility functions are in section 4 of the paper, where we explicitly talk about putting dollars and risks on the same scale, and where we work out expected loss.
Anon,
Yes to what Andrew said.
But also, you can only read off your decision from a graph once you’ve accepted a bunch of assumptions. Some of the key assumptions are reasonably precise (like, how much variation is there in indoor radon concentration in a given county) and some of them are very poorly known (like, what is the risk from inhaling radon decay products, as a function of the concentration, and the person’s age and smoking status). And some of the assumptions are personal, like, what is your personal risk tolerance (expressed as the amount you would be willing to spend to reduce your risk of premature death by 0.1%). In the paper we provide examples, including exemplary plots, based on default values for those parameters, but the paper provides the mathematical machinery for figuring out the optimal decision under a different set of values.
Quoting Thomas Jefferson is a losing game. His actual behavior, as opposed to the soaring, uplifting words he created, was terrible. From
https://www.bing.com/search?q=how+many+slave+did+jefferson+have%3F&cvid=e2134b1561fb4091b1f9888c5090afd3&gs_lcrp=EgRlZGdlKgYIABBFGDkyBggAEEUYOTIGCAEQABhAMgYIAhAAGEAyBggDEAAYQDIGCAQQABhAMgYIBRAAGEAyBggGEAAYQDIGCAcQABhAMgYICBAAGEDSAQkxMDA1MGowajSoAgiwAgE&FORM=ANAB01&PC=U531
————————————————–
“Over his lifetime, Jefferson enslaved over 610 people, with roughly 400 at Monticello and about 200 on his other properties in Virginia.
At any given time, around 130 people were enslaved at Monticello, reflecting the rotation and management of his labor force”
—————————————————-
Yes, Jefferson was good with English words and by all accounts, his French, Italian and Latin were pretty good as well. Unfortunately, reverence for Jefferson is misplaced.
A little late, but I realized I wrote something longer on this blog about the many varieties of anti-decision theory gripe! https://statmodeling.stat.columbia.edu/2025/10/17/separating-the-whack-from-the-chaff-in-critiques-of-decision-theory/
Yes, I recall that discussion well. And I remain confused about supposed rejections of decision theory. I’m sure there are people who reject the entire idea of decision theory, just as there are people who reject science, people who believe what politician X says regardless of what they see with their own eyes, etc. But I think you are talking about academics – trained in some discipline – who reject the entire idea of decision theory. I’m not aware of these views. What I believe is that there are people that object to particular formulations of decision theory (such as rejecting the notion of continuous utility functions) and people (like me) that reject the automation of decision theory in place of humans aided by decision theory. But rejecting decision theory in any form because the real world is complex – who does that? And, what do they suggest in place of decision theory? Are they really saying that all modeling is bad?
I think there’s a subset of technical types who see people as somehow too mysterious to model at all. And conversely a subset of humanists who find any attempt to formulate standards for decision making personally offensive. In general the notion of using theory as a tool for learning—not as a verdict on what needs to be true— is very hard for people to grasp sometimes. Despite how widespread the whole “all models are wrong, some are useful” is.
Also there are professors who, because they teach to students who know less than they do, mistakenly think they’re experts in anything related to anything they teach. So you’ll get an arrogant professor dissing decision theory because he never actually learned how applied decision analysis can work; he just knows it from textbooks and maybe some scattered research articles.
I’ve seen lots of this. Remember, I used to work at a place where my colleagues dissed Bayesian statistics all the time. They were hugely confident while coming from a position of near-complete ignorance.
Christian Robert and I wrote an article about the general phenomenon of ignorant anti-Bayesians: “Not Only Defended But Also Applied”: The Perceived Absurdity of Bayesian Inference along with a rejoinder to discussion, The Anti-Bayesian Moment and Its Passing.
And, as you know, we continue to have to deal with such people.
I don’t mind disagreement or criticism–not at all! Ignorant and overconfident criticism, though, that’s another story.
Jessica
I am unconvinced regarding the “technical types” you refer to – you may well be right, but can you provide specific references to the types you speak of? In the other post you linked to, you similarly refer to a couple of media articles that talk about issues with political polls and AI believers, but neither seem to me to be “dumping on decision theory” like you suggest. If you are complaining about the general public rejecting human behavior as too complex to analyze, then I have no disagreement. There are plenty of human believers in ridiculous things (ghosts? election fraud? UFOs?), but hardly worth posting about. I took your post to be that educated people (for example, academics) reject decision theory – and not specific theories, but the whole idea of decision theory. I’m not aware of this as a general phenomenon, but if it is, I’d appreciate some examples.
Dale:
An example would be the quote shared in this earlier comment, where the speaker smugly dismisses a field that he does not understand.
Andrew
I’m not sure I was reading the quote you cited. If it is the one in the post you reacted to, it only says that human decision making does not seem (to the author) to resemble any of the purported decision making theories. I don’t take that as a rejection of all decision theory, but a rejection of the available theories that people have proposed. That is a tenable view, although I don’t share it (at least not in its entirety). But it doesn’t strike me as saying that no decision theory is possible. Maybe this is degrading into an issue of semantics, but I think it is a bit more than that. I’m not at all convinced that the reactions here are to a straw man (that serious people reject the idea that decision theory can be useful). The danger is that this can be used as an excuse to support someone’s favored decision theory (much like rejecting the null can be used as support for someone’s preferred alternative). I do believe that most decision theories contain some useful information, but that all fail when confronted with particular human behaviors. I don’t know of a single general theory of decision making that seems to describe all human behavior, but I also don’t see that as equivalent to rejecting the usefulness of decision theory.
Dale – the passage from me that’s quoted in the post was literally me sharing my response with Andrew to seeing a prominent machine learning researcher denounhcing the “vulgarity” of decision theory. Such people exist.
It would be helpful to see the post you are referring to.
I love that intro paragraph! I have been working on modeling decisions in cancer therapy for several decades and the resistance to any rational analysis is pretty strong. This is interesting because there is a lot written about the many forms of biases and limitations in physician decision making. Yes, it is true that we have not come up with a perfect model, but as you say, “do you have a better idea?”. Spending time thinking about your goals and values cannot be a bad thing when making such important decisions.