The value of information is the extent to which additional information improves an agent’s decision. This depends on how likely that information is to change the agent’s decision and how much of an improvement that change would be. Value of information will usually be higher the less resilient the agent’s relevant credences currently are, because that will tend to mean new evidence is more likely to change the agent’s credences.
Further reading
Askell, Amanda (2017) The moral value of information, Effective Altruism, June 4.
Muehlhauser, Luke (2013) Review of Douglas Hubbard, How to Measure Anything, LessWrong, August 7.
Related entries
alternatives to expected value theory | credence | explore-exploit tradeoff | forecasting | model uncertainty
Maybe in future, this entry should also discuss option value? Or perhaps we should have another entry for that?
It’s a separate concept!
Yeah, definitely. I was just thinking that it might be sufficiently connected to this concept that a (say) 5 paragraph version of this entry should include a paragraph on that concept or the connections. Like how I’d say that the page on philanthropic diversification should mention diminishing returns (and indeed it does).
But I’m not particularly confident about that; I just shot off the above comment without really thinking about it when I was quickly looking for VoI-related posts to give this tag to, and I saw one I remember liking that was focused on option value but also mentioned VoI a few times.
Re-reading the relevant section of that post now, it does sound like that author would probably approve of option value being mentioned in a 5 paragraph version of this entry. But I haven’t really thought about it.
(I sometimes use these Discussion pages in a brainstormy / note-to-self-y way as I flit about various entries.)
I was already planning to have a separate entry on option value, which I will create shortly. I agree the connection between the two should be briefly discussed in at least one of the two articles.
I think that this article could be usefully expanded using content from the stuff in the Further reading section (including the LessWrong posts with the equivalent tag).