Here's a brief summarly of the first meetup. It took place on cafe on top of Waterstone bookstore on Saturday 2009-04-04, starting at 14:00 and lasting until about 17:15. Six people showed up: Tomasz (me), Michael, Will, Julian, Shane, and Marc.

We started with a game of estimation warewolf - there was 1 decider, and 3 participants, one of the dishonest. 2 people showed up later, so they joined the game as openly honest. I don't think the decider position was really necessary, we could simply randomly assign people to honest and dishonest set. Or if the dishonest guy was necessary, we were confused enough without any active help. The subject was "maize production of Mexico". We did our estimate twice. First based on what Mexicans are likely to eat. Our estimates were:

  • 100 million Mexicans
  • 3000 kcal / day / Mexican
  • 30% of their calories from maize
  • extra 50% on top of the result for animal feed, biofuels and other uses
  • giving 44.3 mln tons of maize per year

Most of the time we discussed it first, and then took the median of our guesses as the estimate. The discussion was really the fun bit, I'll give a few examples later. There was a lot of joking about anchoring effect, but I don't think there was that much of it.

Then we hid the first estimate, and tried the same question again, in a different way, estimating:

  • Land area of Mexico was 3,125,000 square km. This estimate took surprisingly long time, with paths like "USA has 4 time zones, there are 15 degrees per time zone, north Mexican border is about half of south USA border, Mexico is shaped more or less like a triangle" etc., drawing pictures from memory and taking medians of them, guessing how many times different countries are bigger than each other, and guessing from territorial changes after Mexican-American wars. I guess everybody enjoyed it so much, because we all had so many bits of information vaguely related to what we were trying to guess, but no direct estimate.
  • Land used for growing plants is 1/2 of all land.
  • Land for growith maize is 1/3 of that.
  • 10 plants per square meter.
  • 500g per plant.
  • giving 2600 mln tons of maize per year

That was almost two orders of magnitude off. I was so sure that the second estimate failed any sanity check that I took a 10000:1 bet against it (conditional on our arithmetics being correct), taking 1p against 100 pound donation to the Singularity Institute. At this point we gave our point estimates, and checked it against reality. If my notes are correct, they were:

  • Tomasz - 100 mln tons
  • Shane - 150 mln tons
  • Michael - 250 mln tons
  • Will - 75 mln tons
  • Julian - 100 mln tons
  • Marc - 650 mln tons
  • Reality (as spoken through Wikipedia) - 22.5 mln tons

So the Singularity Institute isn't getting their money, our first estimate was very accurate considering our lack of clue about the subject matter, the second was widely off, and everybody except Marc gave more credence to the first. They might have been convinced by me taking that absurd bet via Aumann's Theorem.

Errors on individual estimates were:

  • population - 109 mln Mexicans (estimate 9% low)
  • area - 1972550 km (estimate 58% high)
  • maize consumption - 400 kg / year (estimate 22% low)
  • 13% of Mexican land is arable (estimate 280% high)
  • area used for maize in Mexico - 75500 square km (estimate 588% high)
  • yield per square km - 286 ton/square km (estimate 1650% high) - this number had more doubts than any other, and it seems we were correct in our doubt

After that, we just had chat about the Cabal on Wikipedia (which doesn't exist), tvtropes, early Internet sentimentalism, option pricing, financial crisis, quantum computers, prospects of AGI, and other random subjects. We also invented a new way of paying the bill "by the bailout" - everybody puts as much money on the plate, or takes as much from it as they want, as long as the final result sum was right.

We're most likely going to do the next meeting in about a month, in the same place. Here's the picture. Please correct me if I misremembered anything important.

New Comment
14 comments, sorted by Click to highlight new comments since:

Which of you was the Deceiver and what tactics did he use?

I think having someone advocate for the number to be in a particular direction (either high or low) is a more realistic kind of bad faith to learn to defend against.

[-]taw60

Julian was the Deceiver and from what he said he was basically trying to add noise to the process. I wonder how one can successfully deceive haveing as little clue about the real answer as everyone else.

My main problem was I had so little domain knowledge I couldn't invent any red herrings, let alone realistic ones. The only points where I could throw the result out of whack were those where number guesses were called for - and then I couldn't stray too far from the group too often without giving myself away (and managed to anyhow, Mike says he rumbled me when I estimated Mexicans ate 80% maize).

I got tricked into coming to the defense of what a deceiver ought to do. Heh! Identification and self-justification are powerful.

Mike says he rumbled me when I estimated Mexicans ate 80% maize

That made me highly suspicious, but the best evidence was...

I got tricked into coming to the defense of what a deceiver ought to do.

Hehe. No-one seemed to be trying to put the estimate in a particular direction so I tried to make the problem a solution - make fun of Red and spot the reactions. I was still only .85 sure as those were the only 2 good clues.

[-]taw10

I wasn't actively trying to find out who was Red. Julian said some weird things like 80% maize calories, but everybody said some weird things every now and then.

He hadn't a good reason and stuck to the 80% anyway in a suspiciously out of character way... it's hard to describe. He can take it as a compliment that I expected high standards :-)

[-]taw10

I expected everybody to have massive misconceptions about things, so while I believed his 80% maize calories estimate to be completely wrong, he seemed to me like someone who simply doesn't know much about food. For example notice how completely wrong we were about maize yield per square km, surprisingly all in the same direction - that's the kind of mistake that amateurs make often .

Also now that I think of it - our estimate was for weight of corn with cobs. If by any chance one cited by Wikipedia is for corn without cobs, then our first estimate might have been insanely accurate. I don't really have energy to investigate this, but if someone else wants, go ahead.

What can I say - it was just a gut feeling! :-) As I said on Saturday, I already knew Julian.

I don't think it was too overconfident for that gut feeling to count as the lesser of two pieces of evidence raising p(RedJulian) from .5 to .85.

[-]marc20

I definitely enjoyed the meet up.

In defence of my fairly poor estimate I was unconvinced by the assumption that all the maise in mexico was eaten by mexicans. It seemed that this was an uncontrolled assumption but i felt that i could put reasonable bounds on all the assumptions in the land area estimate (if you're asking, yes, the final answer did fall within my 90-10 bounds :) ).

Hopefully with a bit more notice we can get a few extra people next time but i think it was a great idea to get the ball rolling. Thanks to Tomasz for organising.

In defence of my fairly poor estimate I was unconvinced by the assumption that all the maise in mexico was eaten by mexicans.

Why does it seem so bad? Food products are the most common products countries defend with tariffs and similar policies, and if Mexico is not self-sufficient then even more reason for all consumption to be local.

Maybe I'm missing something here, but isn't the first estimate completely meaningless, since it ignores the possibility of maize import/export? I mean, I know it was closer than the second estimate, but seriously, you guys should have known better than to estimate that half of all of Mexico was used for growing crops.

[-]taw10

We discussed imports and exports, and couldn't decide if Mexico was a net importer or exporter, so we went on assumption that most likely imports and exports are small percentage of total maize production and consumption. I stand by this.

50% estimate was quite reasonable based on European land use patterns. I mean - what else would you use all the land lying around for if not food? I would go for lower number now, but it still doesn't strike me as absurd.

From the perspective of someone in the American west, most of it just isn't used for anything. I think 50% of occupied land being used for agriculture is a decent heuristic; it's just that less land is developed for any purpose in North America.

[-][anonymous]00

That link is for arable land, not total land.

You can see on Wikipedia that only around 12% of Mexico's land is arable.