I have some notion that an argument tree could be translated or incorporated into a Bayes net model. There’s an intuition (which we share) that, given links with a particular imperfect strength, arguments consisting of a few long chains are weaker than arguments that are “bushy” (offering many independent reasons for the conclusion). A Bayes net model would quantify that intuition.
In a perfect world, bushiness would indeed imply high reliability. Unfortunately in our world the different branches of the bush can have hidden dependencies, either accidental or maliciously inserted—they could even all be subtly different rewordings of one same argument—and the technique won’t catch that. So ultimately I don’t think we have invented a substitute for common sense just yet.
With logical links as weak as those in your example, most arguments longer than 10 steps will reach incorrect conclusions anyway.
Agreed.
I have some notion that an argument tree could be translated or incorporated into a Bayes net model. There’s an intuition (which we share) that, given links with a particular imperfect strength, arguments consisting of a few long chains are weaker than arguments that are “bushy” (offering many independent reasons for the conclusion). A Bayes net model would quantify that intuition.
In a perfect world, bushiness would indeed imply high reliability. Unfortunately in our world the different branches of the bush can have hidden dependencies, either accidental or maliciously inserted—they could even all be subtly different rewordings of one same argument—and the technique won’t catch that. So ultimately I don’t think we have invented a substitute for common sense just yet.