# Unnamed comments on Swimming in Reasons

• Agreed—true propo­si­tions (or good ac­tions) typ­i­cally have more pro rea­sons than con rea­sons, and false propo­si­tions (or bad ac­tions) typ­i­cally have more con rea­sons. So when you start out with a high prob­a­bil­ity you should ex­pect to mostly dis­cover more pro rea­sons (al­though in a frac­tion of cases where it is ac­tu­ally false you will in­stead find a large num­ber of con rea­sons). When you as­sign a 50% chance to a propo­si­tion, you should still ex­pect the new in­for­ma­tion to cluster in the same di­rec­tion, you just don’t know which di­rec­tion that will be. So 10 straight pro rea­sons isn’t re­ally a 1/​1024 event since they’re aren’t in­de­pen­dent.

Skewed dis­tri­bu­tions also seem like the typ­i­cal case. If your prior prob­a­bil­ity is not 50% then one tail of the dis­tri­bu­tion for changes in prob­a­bil­ity will be longer than the other, so the dis­tri­bu­tion will be skewed (es­pe­cially since it must bal­ance out the long tail with prob­a­bil­ity mass on the other side to obey con­ser­va­tion of ex­pected ev­i­dence). The nat­u­ral units for strength of ev­i­dence are log-odds, rather than prob­a­bil­ity, which gives you an­other way to think about why the changes in prob­a­bil­ity are skewed when you don’t start at 50% (the same in­crease in log-odds gives a smaller prob­a­bil­ity in­crease at high prob­a­bil­ities).

Part of the trick in avoid­ing con­fir­ma­tion bias (and the like) is figur­ing out which rea­sons should be in­de­pen­dent. I’ll sec­ond JGWeiss­man’s recom­men­da­tion of Eliezer’s post for its dis­cus­sion of that is­sue.

• I agree and have added a clar­ifi­ca­tion at the bot­tom of the post.

• I’d recom­mend rewrit­ing the post to make it clearer. You go back and forth be­tween be­liefs and ac­tions in the post, with­out always mak­ing it clear whether you’re de­scribing similar­i­ties or differ­ences be­tween them, and you give ex­am­ples for your ex­cep­tions but not for your gen­eral prin­ci­ple (un­til the clar­ifi­ca­tion).

• I added some more caveats in the text.