-
Main
- Posts
- Comments
-
Discussion
- Posts
- Comments
Less Wrong is a community blog devoted to refining the art of human rationality. Please visit our About page for more information.
What are the optimal biases to overcome?
If you're interested in learning rationality, where should you start? Remember, instrumental rationality is about making decisions that get you what you want -- surely there are some lessons that will help you more than others.
You might start with the most famous ones, which tend to be the ones popularized by Kahneman and Tversky. But K&T were academics. They weren't trying to help people be more rational, they were trying to prove to other academics that people were irrational. The result is that they focused not on the most important biases, but the ones that were easiest to prove.
Take their famous anchoring experiment, in which they showed the spin of a roulette wheel affected people's estimates about African countries. The idea wasn't that roulette wheels causing biased estimates was a huge social problem; it was that no academic could possibly argue that this behavior was somehow rational. They thereby scored a decisive blow for psychology against economists claiming we're just rational maximizers.
Most academic work on irrationality has followed in K&T's footsteps. And, in turn, much of the stuff done by LW and CFAR has followed in the footsteps of this academic work. So it's not hard to believe that LW types are good at avoiding these biases and thus do well on the psychology tests for them. (Indeed, many of the questions on these tests for rationality come straight from K&T experiments!)
But if you look at the average person and ask why they aren't getting what they want, very rarely do you conclude their biggest problem is that they're suffering from anchoring, framing effects, the planning fallacy, commitment bias, or any of the other stuff in the sequences. Usually their biggest problems are far more quotidian and commonsensical.
Take Eliezer. Surely he wanted SIAI to be a well-functioning organization. And he's admitted that lukeprog has done more to achieve that goal of his than he has. Why is lukeprog so much better at getting what Eliezer wants than Eliezer is? It's surely not because lukeprog is so much better at avoiding Sequence-style cognitive biases! lukeprog readily admits that he's constantly learning new rationality techniques from Eliezer.
No, it's because lukeprog did what seems like common sense: he bought a copy of Nonprofits for Dummies and did what it recommends. As lukeprog himself says, it wasn't lack of intelligence or resources or akrasia that kept Eliezer from doing these things, "it was a gap in general rationality."
So if you're interested in closing the gap, it seems like the skills to prioritize aren't things like commitment effect and the sunk cost fallacy, but stuff like "figure out what your goals really are", "look at your situation objectively and list the biggest problems", "when you're trying something new and risky, read the For Dummies book about it first", etc. For lack of better terminology, let's call the K&T stuff "cognitive biases" and this stuff "practical biases" (even though it's all obviously both practical and cognitive and biases is kind of a negative way of looking at it).
What are the best things you've found on tackling these "practical biases"? Post your suggestions in the comments.
Article Navigation
Comments (64)
I very often conclude that people are suffering from the planning fallacy.
Not falling prey to the planning fallacy is the most obvious and quantifiable result from applying rationality techniques in my day to day life.
How often is that the reason that they aren't making progress toward their goals?
Very often. Any project that goes "over budget" - that's the planning fallacy. On a smaller scale, any meeting which goes too long or has too many scheduled presentations (90% of the meetings I've attended) - that's the planning fallacy. The people who plan meetings or budget projects are aiming for the meetings to end on time and the projects to be completed within their budgets, but they're not meeting those goals.
So... if there is a 10% chance that there will be a 25% cost overrun, and a 90% chance that the unexpected expenses will fall within the contingency budget, should the budget be 125% projected cost, or 102.5% projected cost? If there are 6 items on the agenda, and a 95% chance that each of them will take 5 minutes but a 5% chance that they will take 15 minutes, how long should the meeting be scheduled for?
Keep in mind that meetings will expand to fill the allocated time, even if they are completed before then, and projects will tend to use their entire budget if possible.
Granted, some people budget without a 'contingency' line item, but budgeting for the expected serious cost increase doesn't significantly reduce the odds of going over budget, because the frequency of a serious overrun is so low that the expected cost is much smaller than the actual cost should one occur.
Expecting all projects to complete on time and within budget, now that IS a planning fallacy.
Which is entirely the wrong way to go about the problem. If this project is critical, and it's failure will sink the company, you really, really want to be in a position to handle the 25% cost overrun. If you have ten other identically-sized, identically-important project, then the 102.5% estimate is probably going to give you enough of a contingency to handle any one of them going over budget (but what is your plan if two go over budget?)
Thinking in terms of statistics, without any actual details attached, is one of the BIG failure modes I see from rationalists - and one that laypeople seem to avoid just fine, because to them the important thing is that Project X will make or break the company.
I'd suggest that this is a solvable problem - I've worked in multiple offices where meetings routinely ended early. Having everyone stand helps a lot. So does making them a quick and daily occurrence (it becomes routine to show up on time). So does having a meeting leader who keeps things on-topic, understands when an issue needs to be "taken offline" or researched and brought up the next day, etc..
So, to refine Decius' formula from above, you'd want to add in a variable which represents expected marginal utility of costs.
I don't think the problem here is thinking in terms of statistics; I think that the problem is attempting to use a simple model for a complicated decision.
[edited for grammar]
Both geeks and laypeople seem to use overly simply models, but (in my experience) they simplify in DIFFERENT ways: Geeks/"rationalists" seem to over-emphasize numbers, and laypeople seem to under-emphasize them. Geeks focus on hard data, while laypeople focus on intuition and common sense.
"Intuition and common sense" sound more like styles of thought process, not models. The models in question might be called "folklore" and "ordinary language" — when thinking "intuitively" with "common sense", we expect the world to fit neatly into the categories of ordinary language, and for events to work out in the way that we would find plausible as a story.
If you have a project which will bankrupt the company if it fails, then it does not have a budget. It has costs. If you have multiple such projects, such that if any one of them fails, the company goes bankrupt, then they all have costs instead of budget.
Note that I'm assigning such a large negative value to bankruptcy such that it is trivially worse to be bankrupt with a large amount of debt as it is to be bankrupt with a smaller amount of debt- if the sunk costs fallacy applies, then there is a fate significantly worse than cancelling the project due to cost overruns; funding the project more and having it fail.
Tricks to avoid long meetings are different than figuring out how long a meeting will last.
Thus it instead being in response to the idea that meetings will expand to fill their schedule - if you don't solve that, then scheduling is that much less reliable.
Yes it does; even if the budget is "100% of the company resources", that's still a constraint. Given that the odds of success probably drop drastically if you stop providing payroll, paying rent, etc., then it's constrained further. It might also be the case that spending (say) 10% of your resources elsewhere will double your profits on success, but you have a corresponding 10% chance of failure because of it.
90% chance of a major success vs 10% chance of bankruptcy is not necessarily a trivial decision.