Think Before You Speak (And Signal It)


23


Wei_Dai

In deciding whether to pay attention to an idea, a big clue, if it were readily available, would be how many people have checked it over for correctness, and for how long. Most new ideas that human beings come up with are wrong, and if someone just thought of something five seconds ago and excitedly wants to tell you about it, probably the only benefit of listening is not offending the person.

But it seems quite rare for this important piece of metadata to be straightforwardly declared, perhaps because such declarations can't be trusted in general. Instead, we usually have to infer it from various other clues, like the speaker's personality (how long do they typically think before they speak?), formality of the language employed to express the idea, the presence of spelling and grammar mistakes, the venue where the idea is presented or published, etc.

Unfortunately, such inferences can be imprecise or error-prone. For example, the same speaker may sometimes think a lot before speaking, and other times think little before speaking. Using costly signals like formal language is also wasteful compared to everyone simply telling the truth (but can still be a second-best solution in low-trust groups). In a community like ours, where most of us are striving to build reputations for being (or at least trying to be) rational and cooperative, and therefore there is a level of trust higher than usual, it might be worth experimenting with a norm of declaring how long we've thought about each new idea when presenting it. This may be either in addition to or as an alternative to other ways of communicating how confident we are about our ideas.

To follow my own advice, I'll say that I've thought about this topic off and on for about two weeks, and then spent about three hours writing and reviewing this post. I first started thinking about it at the SIAI decision theory workshop, which was the first time I ever worked with a large group of people on a complex problem in real time. I noticed that the variance in the amount of time different people spend thinking through new ideas before they speak is quite high. I was surprised to discover, for example, that Gary Drescher has been working on decision theory for many years and has considered and discarded about a dozen possible solutions.

The trigger for actually writing this post is yesterday's Overcoming Bias post Twin Conspiracies, which Robin seemed to have spent much less time thinking through than usual, but which has no overt indications of this. (An obvious objection that he apparently failed to consider is, wouldn't corporations actively recruit twins to be co-CEOs if they are so productive? Several OB commenters also pointed this out.) A blogger may not want to spend days poring over every post, but why not make it easier for the reader to distinguish the serious, carefully thought out ideas from the throwaway ones?