Saturday, October 08, 2016

[exgbexzz] Adversarial truth discovery

An adversarial framework seems like it should work well for discovering truth: each side challenges the other's version of the truth, causing lies and manipulation to be revealed.

We see this model in science, in the justice system, in political campaigns in democracies, in journalism.  However, these examples cynically or pessimistically demonstrate that reality is almost diametrically the opposite of theory.  It sometimes works in science (but often not: "cargo-cult science"), and both the justice and political systems are about who can bullshit better.  Journalism, motivated by advertising revenue, publishes the entertaining version of the story, not necessarily the true one.  In all these fields, effort seems better spent trying to take advantage of fallibility in human reasoning than to actually discover truth.

We imagine an alternate universe in which adversarial truth discovery does work well in all these fields.  It would have an interesting side effect: transcripts of jury trials and propaganda from a political campaign would be extremely good educational resources afterward.  Suppose unemployment were an important political issue between two parties.  As the sides debate and try to convince voters, both sides need voters to understand how the labor market functions and how domestic and macroeconomics work so that voters can evaluate propsed policy differences between the two parties.  Therefore, both sides will produce very good, easy to understand, educational material on these topics.  If the educational material is bad, confusing, the voter might come to the wrong conclusion and vote for the opponent, so there is tremendous incentive for the educational material to be good, and for the parties to attack opponent's bad educational material.  Compare this to the incentives (or disincentives) to create good educational material in the traditional education system.

Perhaps someday computers can help.

No comments :