Towards no-math, graphical instructions for prediction markets

You are prob­a­bly fa­mil­iar with pre­dic­tion mar­kets. As of the last gen­eral up­date, Robin Han­son notes that while there is in­creas­ing re­search in­ter­est and a few gen­eral plat­forms are com­ing soon, the real goal of im­ple­ment­ing in­side or­ga­ni­za­tions re­mains elu­sive.

I strongly sus­pect that it will re­main so un­til com­pa­nies are founded which use them from the be­gin­ning. This is be­cause in­tro­duc­ing peo­ple to new tools is hard. In­tro­duc­ing them to new tools that don’t even have a di­rect bear­ing on do­ing their jobs is even harder. We know that triv­ial in­con­ve­niences are im­por­tant; it was the dom­i­nant theme of the crypto au­topsy. If it hap­pens in an en­vi­ron­ment like Less­wrong with crypto, I ex­pect it will be worse in vir­tu­ally all other con­texts.

If pre­dic­tion mar­kets are go­ing to be widely adopted, they need to be widely ap­pli­ca­ble, which ba­si­cally means that any given schlub needs to be able to make a bet in a mar­ket with­out any math/​eco­nomics/​com­put­ing back­ground. Hap­pily, we have ex­tremely ro­bust ev­i­dence that any given schlub is perfectly ca­pa­ble of mak­ing a bet, cour­tesy of cas­inos.

One way to help with this prob­lem is to make it as easy and in­tu­itive as pos­si­ble to learn. To wit: in­ter­ac­tive vi­su­al­iza­tions. The biggest in­spira­tions are Up and Down the Lad­der of Ab­strac­tion by Brett Vic­tor, and Me­tac­u­lus. The first is about ex­plor­ing sys­tem be­hav­ior with­out math, and the train­ing for the sec­ond does a pretty good job of let­ting you make pre­dic­tions based on the shape of the prob­a­bil­ity curve (though this doesn’t work as well for un­der­stand­ing the pay­offs). Another rele­vant tool is the Sankey Di­a­gram, which makes the rel­a­tive sizes of flows ap­par­ent at a glance. I don’t know of any­thing which makes such a thing in­ter­ac­tive though—which is a shame, be­cause I bet it would make con­trols com­pre­hen­si­ble to peo­ple who don’t know any differ­en­tial equa­tions or lin­ear alge­bra. I am cur­rently im­pressed by See­ing The­ory. I ran across a few other web­sites which are de­signed to com­mu­ni­cate math, and even one about risk, but for some rea­son all these sav­ages are us­ing Flash.

How­ever, all of these is mostly for teach­ing math, and that is not ac­tu­ally the goal. Fun­da­men­tally the idea is in­for­ma­tion flows into the mar­ket, and only pay­ments need to flow back out. So what we need is a way for peo­ple to spec­ify what they think the out­come will be, with what con­fi­dence, with­out num­bers.

Naively I am lean­ing in this di­rec­tion:

  • Num­bers are pre­sented as areas

  • Func­tions are pre­sented as shapes

  • Oper­a­tions show up as trans­for­ma­tions of the ar­eas/​shapes (for ex­am­ple, show­ing how a given bet af­fects the over­all mar­ket, or maybe com­bin­ing sub-bets).

  • We want peo­ple to be able to trans­late their be­liefs this way pretty eas­ily. Stuff like “the more sure you are, the nar­rower the shape should be,” where the shape is the dis­tri­bu­tion curve, is in­tu­itive.

I feel like the in­struc­tions could be col­lapsed into a few drag-and-drop shapes, and a few heuris­tic guidelines for how to ma­nipu­late them to rep­re­sent a be­lief ac­cu­rately. Train­ing peo­ple how to do this should eas­ily fit into a sin­gle-ses­sion course, not more than a day if we in­clude mo­ti­va­tion, the pay­offs, and a bunch of trial runs.