Applying decision theory when you have insufficient domain understanding leads to the “optimizer’s curse”: it’s a mathematical fact that you will typically choose options that have less-good outcomes than you expect. https://pubsonline.informs.org/doi/abs/10.1287/mnsc.1050.0451 …
-
Show this thread
-
Formally, probability theory only applies in extremely restricted conditions that are rarely or never found in the real world. It is, of course, often useful when it does not quite formally apply. Some rationalists insist on applying it universally, which doesn’t go well.pic.twitter.com/yXyMkgUaSg
1 reply 4 retweets 12 likesShow this thread -
Discussion of
@Chris_PK_Smith’s article here includes both useful elaboration/clarification and a display of rationalist stubbornness, digging in and insisting that decision theory MUST always work. Because otherwise, it wouldn’t always work, as it MUST!https://forum.effectivealtruism.org/posts/Wghi6hpu5gGBZHvtj/link-the-optimizer-s-curse-and-wrong-way-reductions …1 reply 0 retweets 4 likesShow this thread -
When rationality doesn’t work well, it’s time to get meta-rational. That is: step outside the rational system to examine how and when and why it works, and on that basis to revise its use in a situation, select an alternative, or combine it with other systems to do better.
1 reply 0 retweets 10 likesShow this thread -
Chris suggests several meta-rational approaches to revising or replacing the use of decision theory in selecting what altruistic projects to fund. These apply more broadly to decision making in other fields.
1 reply 1 retweet 3 likesShow this thread -
Venture capitalists address extreme uncertainty by using several quantitative and qualitative systems simultaneously, e.g.: - market size estimate - comparable-company ratios - discounted cash flow calculation - qualitative management team assessment - industry fashion trends
1 reply 0 retweets 7 likesShow this thread -
Using multiple incommensurable rational systems is a typically meta-rational move. When they disagree, it’s often not feasible to figure out which is right, or to reconcile them. When most investment analysis methods point in the same direction, you may choose to act.pic.twitter.com/rA0ESS4YXY
1 reply 0 retweets 8 likesShow this thread -
Chris gives a concrete example in the Effective Altruism domain, namely malaria net distribution, which looks good when evaluated in many different ways. (Whereas some other projects he discusses got funded on the basis of a single dubious expected-value model.)pic.twitter.com/AFFQgnqf6r
1 reply 0 retweets 4 likesShow this thread -
When you know your probability and/or value estimates are highly uncertain, you can gather information to get better estimates. Also, your model may be wrong overall, not just its numerical parameters. Deciding how much effort to put into better understanding is meta-rational.pic.twitter.com/9jDgzi2uqN
2 replies 1 retweet 3 likesShow this thread -
David Chapman Retweeted David Chapman
Decision theory assumes you are constantly forced to make specified decisions. In reality, you can walk away from most situations, and that’s often sensible. Don’t make bets with Dutch bookies. Carry a musket, and if you meet a Bayesian Thug, shoot them.https://twitter.com/Meaningness/status/1001172948894466048 …
David Chapman added,
2 replies 8 retweets 17 likesShow this thread
The world is full of unknown unknowns, which decision theory can’t handle at all. Increasing a system’s general robustness to failure is often possible, and… might be the most effective form of altruism.pic.twitter.com/RiEKnEssp0
-
-
Replying to @Meaningness @Chris_PK_Smith
Carry a musket! You sound a lot like
@nntaleb this week
I'm missing one half of the barbell though: increasing a system's aliveness, or ability to gain strength from encounters with unknown unknowns (antifragility).1 reply 0 retweets 0 likesThanks. Twitter will use this to make your timeline better. UndoUndo
-
Loading seems to be taking a while.
Twitter may be over capacity or experiencing a momentary hiccup. Try again or visit Twitter Status for more information.