4 ms·
> Hard to describe, but it is not unlike someone coming up to you on the street and saying, "Excuse me, should I drive down that street, that one, or take the f
by derefr 13y ago
> Hard to describe, but it is not unlike someone coming up to you on the street and saying, "Excuse me, should I drive down that street, that one, or take the freeway?" The answer is meaningless unless you know the destination.
Actually, that is exactly what I was asking, and it seems like a perfectly sensible question to me. (Though it's not a question you'll encounter every day.[1])
I think you're ignoring what, exactly, a minimax algorithm entails: it's not a goal-directed least-cost search, like, say, A⁎. Instead, it's more about picking and weighting a set of potential goals, given the costs involved with pursuing them.
So the question becomes, in effect, not "what road should I take to get to X," but rather, "which road should I start heading down, assuming I'll narrow down the set of possible Xes I might want to pick later, that will cause me the least problem for any given X I might have chosen?"
Oddly enough, this is a question driving-directions programs have a heuristic for. If you chart a course, on e.g. Google Maps, between suburban addresses in two different cities, the directions will almost always start with what amounts to "first, get out of the suburbs, so you can reach the highway."
Given this, without knowing where you're going, "get out of the suburbs and onto the highway" is a high-value guess in how to get there.[2] You might be wrong, of course: the place you might like to go might end up being in the same neighborhood you were already in. But, across the set of all possible destinations, you'll regret the strategy of heading to the highway less often than you will the strategy of "sit in your driveway waiting to see where you're supposed to be going."
---
[1] It's kind of an inhuman way of thinking, really. Humans tend to come with both "axiomatic" terminal goals, and deontological notions of how they have to get where they're going. Minimax solutions only become relevant when, for example, you've decided that what you'd like to do with your life is "to pursue, in a consequentialist manner, the maximum amount of money-earned-per-unit-time, so I can then engage in effective altruism." ;)
(More generally, though, every real-world question people deal with where they want a "good answer that leaves my options open" is actually a bit of goal-directed search, and a bit of minimax. You do want to invest effort getting to X; but you don't want to have invested too much, if it turns out X is closed, or your friend wants to meet at Y instead.)
[2] Actually, I just thought of a real-world scenario for this. A lot of jobs (e.g. truckers, ambulance drivers, taxis) involve a combination of a driver and a dispatcher, where the driver has to be ready to get "anywhere" the dispatcher might tell them to go. If dispatches come in steadily, then the driver must simply travel directly from the end of whatever route the last dispatch took them on to the beginning of the next. But if the driver has enough time between the end of the previous dispatch, and the start of the next, then where should they drive toward, optimally, to wait?
- ChuckMcM 13y agoNice! You are correct of course, I did mis-understand your question. I suspect the minimax answer is Java. My reasoning for that is that enterprise shops both consume and turn over a lot of Java programmers, this is in part to consulting gigs with Salesforce, SAP, Oracle, or IBM. Thanks for the clear response and your excellent example. It is interesting to see what it optimizes for. I don't doubt you could stay employed "forever" at $45,000/year Java shop code monkey jobs for example. My personal conjecture is that Tech is not a 'connected graph' in the sense that not all nodes are reachable from any node, and not all paths can be traveled in both directions. The example I am most familiar with is 'CEO taint' which is folks who get to a certain level, and then become CEOs of their own startup, where they reach a certain level of qualified success. Then, in a quest for finding their next engagement they are un-hirable because 'former CEO' isn't something you higher in to be a coder, and maybe not manager. Their only accessible nodes are 'new startup' or 'appointment as CEO to startup' or 'consultant'. If you can't get funding for a startup, and you don't have enough street cred as a 'consultant', its a really sticky place to be. The other common one I've seen is people who 'grew up' in a company, started right out of college and put in 10, 15, or 20 years, and then left as UberDistinguishedPrinciple Engineer/Scientist but were unable to find anyone who would hire them at a level and salary they could easily accept. But I'm wondering how much of that transition resistance is simply pride/wages related.