Meetup : Bath, UK: Agreement, practical meetups, and report from last meetup

Dis­cus­sion ar­ti­cle for the meetup : Bath, UK: Agree­ment, prac­ti­cal mee­tups, and re­port from last meetup

WHEN: 02 Novem­ber 2014 02:00:00PM (+0000)

WHERE: 5-10 James St W, Avon, Bath BA1 2BX

Bath, UK will be hav­ing its sec­ond meetup this Sun­day 2nd Novem­ber at 14:00 in the King of Wes­sex, which is a Wether­spoons pub in the city. I shall wait at least ninety min­utes (i.e. un­til 15:30) for the first ar­rivals.

I’ll put a sheet fea­tur­ing a pa­per­clip and say­ing ‘Less Wrong’ on the table so you know you’ve found us. Make sure you ven­ture into the pub, since there’s no guaran­tee our table will be near the door.

In case you need to con­tact me (e.g. if the venue is un­ex­pect­edly busy and we have to move el­se­where and you can’t find us), my mo­bile num­ber is the product 3 x 3 x 23 x 97 x 375127, pre­ceded by a zero (so eleven digits to­tal).

We have a Face­book group.

At the start we’ll chat for a bit, then move onto an agree­ment ex­er­cise: Un­like last meetup, where we made pre­dic­tions for our own Pre­dic­tionBook ac­counts some­what in­de­pen­dently with­out nec­es­sar­ily shar­ing all our in­for­ma­tion, this time we shall try to reach con­sen­sus in our prob­a­bil­ities and then see how our con­sen­sus is cal­ibrated, by means of a sin­gle Pre­dic­tionBook ac­count for the meetup group.

After that, we shall dis­cuss ideas for fu­ture mee­tups and ac­tivi­ties. In par­tic­u­lar, we shall dis­cuss how we can move for­ward with prac­ti­cal mee­tups and in­stru­men­tal ra­tio­nal­ity, and how to bal­ance this with dis­cus­sion and ‘ab­stract’ or epistemic stuff.

====Pre­vi­ous meetup (2014-10-19 Sun­day)====

It went well. There was me, some­one who tagged along with me, some­one from Bris­tol, and some­one from Brad­ford-on-Avon. I think ev­ery­one had ar­rived by 14:30, and we prob­a­bly stayed un­til 17:30 or 18:00. (We stayed long enough that we all got some­thing to eat in the pub.)

We got to know each other a bit then did 15 pre­dic­tions. The pre­vi­ous night, I had pre­pared a list of prompts for things to make pre­dic­tions about, rang­ing from things where I thought we might have very high (or low) con­fi­dences, to things where I ex­pected that most of the at­ten­dees would be ba­si­cally in­differ­ent (e.g. whether an even or odd num­ber of el­e­ments have been ob­served, whether the den­sity of wa­ter is above or be­low 1kg per liter, etc.).

We skipped some of the am­bigu­ous prompts, and for a cou­ple we had to sort of figure out what we’d use to judge the pre­dic­tion mid­way through. I’d state the prompt, then where nec­es­sary we’d pin it down into some­thing we could judge ob­jec­tively enough. There might be some brief dis­cus­sion, but we weren’t try­ing to share all our in­for­ma­tion. I would type in (but not sub­mit or write my prob­a­bil­ity for) the fi­nal word­ing of the pre­dic­tion on Pre­dic­tionBook. At a suit­able point, when ev­ery­one un­der­stood what we were pre­dict­ing and how it would be judged, I would give 90 sec­onds for ev­ery­one to stop com­mu­ni­cat­ing and log their fi­nal prob­a­bil­ities. I’d then type in my prob­a­bil­ity and cre­ate the pre­dic­tion on Pre­dic­tion­book, then we’d go round stat­ing the prob­a­bil­ity we’d writ­ten down.

Some of the prompts were in­ten­tion­ally un­der­speci­fied. For ex­am­ple, the first pre­dic­tion was about Wladimir Klitschko’s mass. In that case we each in­de­pen­dently (to avoid prim­ing) wrote down a figure (af­ter ex­plain­ing who he is, of course). Then we took the usual mean of the figures to ob­tain a ‘wis­dom of crowds’ es­ti­mate and used that as the mass for the pre­dic­tion.

(If you’re wor­ried that av­er­ag­ing the guesses would lead ev­ery­one to put 50% prob­a­bil­ity on the propo­si­tion, then you can shift by some amount to en­courage more ex­treme con­fi­dences. But re­mem­ber that it’s still use­ful to test cal­ibra­tion at the 50% level!)

That was one of the cases where we had to de­cide part­way through how we were go­ing to judge the pre­dic­tion, since we re­al­ised his mass would fluc­tu­ate a lot de­pend­ing on e.g. whether he’d cut weight for a weigh-in. We agreed that if Google gave a unique figure and it seemed plau­si­ble, then we’d go with that. I’m not cer­tain, but I don’t think we ac­tu­ally shifted the av­er­age in this case, and the mean of our ini­tial guesses turned out to be ex­actly cor­rect (110kg).

In some cases, where the ini­tial es­ti­mates varied wildly, I sug­gested we use a ‘log­a­r­ith­mic av­er­age’, i.e. use the ex­po­nen­tial of the mean of the log­a­r­ithms of the es­ti­mates, i.e. exp(ar­ith­metic_mean(log(es­ti­mates)).

Then we’d check the pre­dic­tion and I’d mark it Right or Wrong ac­cord­ingly on Pre­dic­tion­book. When they got home, the oth­ers marked the pre­dic­tion Un­known, then put in the prob­a­bil­ity they’d made a note of, then re-mark the pre­dic­tion as Right or Wrong as be­fore.

I had my lap­top and used the pub’s Wi-Fi to cre­ate the pre­dic­tions on Pre­dic­tionBook with my es­ti­mate. The oth­ers made a note of their prob­a­bil­ities. After each pre­dic­tion, we checked the pre­dic­tion and I marked it Right or Wrong on Pre­dic­tionBook ac­cord­ingly. When they got home, each of the oth­ers who at­tended then marked the pre­dic­tion Un­known, then sub­mit­ted their prob­a­bil­ity from ear­lier, then re-marked the Pre­dic­tion as Right or Wrong.

Dis­cus­sion ar­ti­cle for the meetup : Bath, UK: Agree­ment, prac­ti­cal mee­tups, and re­port from last meetup