What are the most common and important trade-offs that decision makers face?
This is one part shameless self-promotion and one (hopefully larger) part seeking advice and comments. I’m wondering: what do you guys think are the most common and/or important trade-offs that decision makers (animals, humans, theoretical AIs) face across different domains?
Of course you could say “harm of doing something vs benefit of doing it”, but that isn’t particularly interesting. That’s the definition of a trade-off. I’m hoping to carve out a general space below that, but still well above any particular decision.
Here’s what I have so far:
1) Efficiency vs Unpredictability
2) Speed vs Accuracy
3) Exploration vs Exploitation
4) Precision vs Simplicity
5) Surely Some vs Maybe More
6) Some Now vs More Later
7) Flexibility vs Commitment
8) Sensitivity vs Specificity
9) Protection vs Freedom
10) Loyalty vs Universality
11) Saving vs Savoring
Am I missing anything? I.e., can you think of any other common, important trade-offs that can’t be accounted by the above?
Also, since so many of you guys are computer programmers, a particular question: is there any way that the space vs memory trade-off can be generalized or explained in terms of a non-computer domain?
Relevance to rationality: at least in theory, understanding how decisions based on these trade-offs tend to play out will help you, when faced with a similar decision, to make the kind of decision that helps you to achieve your goals.
Here’s an intro to the project, which is cross-posted on my blog:
About five years ago I became obsessed with the idea that nobody had collected an authoritative list of all the trade-offs that cuts across broad domains, encompassing all of the sciences. So, I started to collect such a list, and eventually started blogging about it on my old site, some of which you can find in the archives.
Originally I had 25 trade-offs, then I realized that they could be combined until I had only 20, which were published in the first iteration of the list. As I noted above, at this point I wanted to describe all possible trade-offs, from the space vs memory trade-off in computer science, to the trade-offs underlying the periodic table, to deciding what type of tuna fish you should buy at the grocery store.
Eventually, I decided that this would not only be a) practically impossible for me, unless life extension research becomes way more promising, b) not particularly interesting or useful, because most of the trade-offs that come up over and over again occur because of the context-dependent structure of the world that we live in. In particular, most trade-offs are interesting mostly because of how our current situations have been selected for by evolutionary processes.
Upon deciding this, I trimmed the trade-offs list from 20 down to 11, and that is the number of trade-offs that you can find in the essay today. This new goal of indexing the common trade-offs that decision makers face is, I think, still ambitious, and still almost certainly more than I will be able to accomplish in my lifetime. But this way the interim results, at least, are more likely to be interesting.
Ultimately, I think that using these sort of frameworks can be a helpful way for people to learn from the decisions that others have made when they are making their own decisions. It certainly has been for me. I’m actively seeking feedback, for which you can either email me, leave me anonymous feedback here, or, of course, comment below.
A few quick thoughts.
You should look into TRIZ, which seems to be related to your idea, albeit it is mostly related to engineering (however, proponents of TRIZ try to apply it to other domains). Basically TRIZ is a collection of common engineering trade-offs that are arranged into a contradiction matrix and a set of advice how to come up with workarounds, called 40 Inventive Principles (How to use it? You are supposed to formulate your problem as a trade-off, find a relevant cell in the contradiction matrix, and look the numbers up in the list of 40 principles). However, taking the outside view, you should note that TRIZ is still relatively unknown outside the former Soviet Union. If it was as useful as it is sometimes claimed to be, wouldn’t it be very popular by now? Or was it cultural barriers that prevented it from spreading? Frankly, I don’t know.
In statistics, there is a well known Bias–variance tradeoff, which affects a wide range of situations. The trade-off between the goodness of fit of the model and its complexity (4th item in your list) is somewhat related (though not identical) to it.
A general pattern. Some trade-offs are due to Berkson’s paradox, because sometimes the situation we have is “prefiltered” by the past, and we can observe negative corellations between variables even though they are not intrinsically negatively correlated. A special case: if we can do both X and Y, but that requires resources and we have a fixed amount of resources, then the decision how much X and how much Y should be done becomes a trade-off.
Closely related example is the trade-off between quantity and quality, e.g. r/K selection theory.
A different kind of “bias-variance” tradeoff occurs in policy-making. Take college applications. One school might admit students based only on the SAT score. Another admits students based on scores, activities, essays, etc. The first school might reject a lot of exceptional people who just happen to be bad at test-taking. The second school tries to make sure they accept those kinds of exceptional people, but in the process of doing so, they will admit more unexceptional people with bad test scores who somehow manage to impress the admissions committee. The first school is “biased” against exceptional students with bad test grades—the second school has more “variance” because by attempting to capture the students that the first school who wrongly reject, they admit more low quality students as well. You might interpret this particular example as “sensitivity vs specificity.”
Another example would be a policy for splitting tips at a restaurant. One policy would be to have all the staff split the tips equally. Another policy would be to have no splitting of tips. Splitting tips incurs bias, not splitting incurs variance. An intermediary policy would be to have each staff member keep half of their own tips, and to contribute the other half to be redistributed.
Thanks, these are great examples. How did you come up with them?
I think TRIZ has multiple reasons it’s not adopted as much as the benefits might imply due to multiple reasons, its Russian origin only one of some. The reason is also: heavily paywalled and most low-hanging fruits in technological innovation have be taken, thus the value of picking the remaining must be balanced against all the other efforts of a technological business.
I hadn’t heard of this and it’s interesting. Thanks. In what context did you find it?
I have it subsumed under Precision vs Simplicity; I’ll make this explicit in the next iteration.
Thanks, this is an interesting point, and one that I have thought about, see here.
I have these subsumed under Surely Some vs Maybe More; I’ll make this explicit in the next iteration.
The most standard business tradeoff is Cheap vs Fast vs Good, which typically you’re only supposed to be able to get two of.
Yeah I find these three pronged trade-offs fairly interesting. I think it’s wrong to say “choose two”; for example, you could always choose to be somewhere in the middle if you consider the space to be a triangle.
Do you know of the word for a three pronged trade-off?
Trilemma maybe?
But in a trilemma you can only get one, not two, right?
Type I vs. Type II errors?
I think that’s Sensitivity vs Specificity
They are slightly different, but in practical terms they describe the same error; sensitivity and specificity are properties of a test while Type I and II errors are properties of a system, but both errors are basically saying, “Our test is not perfectly accurate so if we want to catch more people with a disease we need to misdiagnose more people”
To illustrate the distinction, consider a test which is 90% sensitive and 90% specific in a population of 100 where a disease has a 50% prevelance. This means 50 people have the disease, of which the test will identify 45 as having the problem (90% sensitive). 50 people are free of the disease, of which the test will correctly identify 45 (90% specific). So if diagnosed your probability of the diagnosis being a Type I error is 5⁄50 = 10% (if given the all clear the same logic applies for a Type II error). You derive this from the number of people in the population who were told they have a disease who were incorrectly diagnosed divided by the total population who were told they have a disease (rightly or wrongly)
But if the disease prevelance changes due to demographic pressue to 10% then 10 people have the disease of whom 9 are diagnosed, and 90 people are disease-free of whom 81 are given the all-clear. This means the probabilities of the different ‘Type’ errors change dramatically; now 9⁄18 = 50% for a Type I error and 1⁄82 ~ 1.2% for a Type II error. But the sensitivity and specificity of the test are completely unchanged.
How about:
Specialization of Labor vs. Transaction/Communication costs: a trade off between having a task split between multiple people/organizations vs. done by a single person. Generalism vs.Specialization might be a more succinct way to put it.
Also, another pair that has a close connection is 3 and 7. Exploration is flexible strategy, since it leaves open resources to exploit better opportunities that turn up, while exploitation gains in commitment.
I was thinking build vs buy or I source vs outsource being much like some of the first point.
I believe Signalling vs Actually getting your job done is extremely pervasive and not accounted for by your list.
I think that Signaling vs Actually doing things would map to Saving vs Savoring in the typical case that you want status in the short run via signaling (“savoring”) and shifting the world more towards your values if you choose to signal less (“saving”).
No, I think these are quite different things. Signaling vs Doing is less about short-term vs. long-term and more about satisfying external/appearance requirements vs satisfying internal/actual requirements.
Those two really need a better name. Profit-Taking vs Investment?
One marshmallow vs. Two marshmallows :-)
Economists would call it consumption vs. investment.
The trade-off is basically about delayed gratification.
Why is unpredictability a good thing? Is this something where you’re playing a game with another agent and you don’t want him to predict what you’ll do?
I don’t think those are opposite. They’ll occur in tradeoffs, but only in the sense that any two things will. There’s also speed vs cost, or speed vs documentation, etc.
I believe this is normally stated as “Risk vs Reward”.
Yes, exactly. Kind of narrow, but I just find it really cool. And, lots of things decision makers have to do can be formulated as games with other agents.
I agree that they might not be “intrinsic”, but if they’re very commonly inversely correlated in situations we care about (which I think they are), then considering it as a trade-off can still be useful. See here for more.
Thanks. I’ll make this explicit next time. I like to have an upside connoted by both words of the trade-offs, which is why I didn’t like risk vs reward.
Come to think of it, “Risk vs Reward” is oddly-worded. It’s both vs neither. If you had to choose between the two, you’d pick reward every time.
That’s because there ain’t no such thing :-) there is no “vs” there.
People usually talk about risk/reward ratios. The trade-off is between a larger reward with more uncertainty and a lower reward with less uncertainty. Your preferences in this trade-off are often described as risk aversion.
Constant, predictable gains vs. Black Swans
A lot of gains that aren’t constant but are variable have still nothing to do with Black Swans.
I think this is Surely Some vs Maybe More—right? If so, helpful to recall the black swan meme and map it to this, thanks.
It seems to me that 6 and 11 are pretty close; what’s the difference between them that you wanted to highlight?
Good find and I agree that they are pretty similar. As I write in the text,
Do you agree?
I interpreted “saving” as “saving for yourself,” like putting money into a bank account so you’ll have it later rather than spending it now. Universality over the long run seems to me more like ‘donating’ or ‘investing’ or… in Triumphs of Experience, this gets called “guardianship” vs. “hoarding,” but I think that’s still framing both in long-term terms, and it’s not obvious that “guardianship” means “developing a community using your resources” (the intended meaning) rather than “protecting things.”
Interesting, thanks for explaining how you interpreted it. I interpreted it as “saving the world.” I’ll try to see what others think as well.