BBC's prof. Marcus du Sautoy explains how a group of people know more than one individual. Amazing stuff! The explanation is not hard to understand, but still it is hard to believe.
I would be fascinated to see how differently the collective accuracy of guesses changes when the people witness every other person's guess, and when each person guesses separately and secretly. To see if social factors play any impact on guesses.
This is why we should always stick together and let nothing divide us and let nothing come between ourselves......we simply are smarter when we're together
People think the Asian girl who gave 50,000 was way off. When in fact she had estimated how wrong the 159 people were and gave a number that would lead to the correct answer. This is the power, the power of Asian.
I want to do an experiment. I want to see what percentage of people came here from Vsauce. Since there's no way of really figuring it out. I'm going to try and see what Wisdom Of the Crowds has to say about this whole thing. I'll write the percentage I think came from Vsauce at the bottom, before you check my answer, have an answer in your head to make sure mine doesn't influence yours. - - - - - - - - - - - - - - - - - - - - - - I think it's around 70% of all the viewers, that came from Vsauce.
There are many more examples of this phenomenon in James Surowiecki's book 'The Wisdom of the Crowd'. Definitely worth checking out if this video interested you.
There is an absolute minimum, but no absolute maximum. Taking the average cannot provide a reliable answer because no one can guess -75000 to cancel out ridiculously high guesses. In such a situation, anyone that guesses over twice the real answer is doing more than one person's worth of damage to the mean.
Guys this is really mind blowing. I've studied statistics at school when i was young but i've never seen this with the curiosity and awareness that i've today at almost 40. Do you realize how this simple experiment opens up to interesting discussions about reality, consciousness, collective consciousness, and many other existential mysteries? This is a very underrated topic that should be taken more into consideration for its importance, imho.
I am blown away you ask this.. I have to ask why you ask that! Indeed, I was watching VSAUCE about 15 minutes ago, but it didn't directly bring me here. I don't even remember which video I was watching, but it reminded me of a video I saw a long time ago about jellybeans and Google data and I couldn't find it.
Engineers use this all the time. If you're trying to measure something that is extremely difficult to measure (noisy data for instance), considering only one single measurement would be very dangerous, since it might be extremely imprecise. However, if you measure 1000 times and take the average, the errors cancel out and you have a pretty accurate measurement :)
The right way to do this scientifically would be to have the guy running the experiment have no idea how many beans were in the jar, in fact not even see the jar himself only asking others to look (after the end, of course, they can look). Also, the contest needs to be announced ahead of time otherwise you run the risk of only contests which produce the shocking result being revealed publicly. Reminds me of XKCD #882, "Significant" ("Found a link between green jelly beans and acne, p>.05").
this is absolutely amazing. Markets do this after all.. I wonder why no-one actually exploit this to make decisions. We talk so much of AI but maybe crowds are actually the cleverest thing could ever exist.
THX Vsauce for letting me see this amazing video... It is so complicated but at the same time so easy. At first I thought quantum mechanics was at work when they were talking about the "code" but after wards I realised it's just crazy math!
The concept of Wisdom of the crowds IS repeatable and it is a rather thoroughly grounded phenomenon. Do the experiment for yourself if you want (as long as you get an appropriate sample size). Or you can just look up the effect if you want. Of course it's not going to always be reliable but the effect is well documented.
With the one guess of 30,000, if you subtract that from the total 722,383.5, and put a more "educated" or similar guess to some of the others, with 3,000 and average it accounting for the 160 people. The final average comes out to be 4346.2. Which is a few hundred different to the actual number of beans compared to the 4 bean difference with the 30,000 guess. So although it is still accurate, that 30,000 guess was actually pretty lucky.
I think leaving out the 80,000 or 50,000 would make sense because you would have to have some kind of filter. For example in crowd sourcing you would try to avoid people with significant mental disability. And without counting normal people would know 80,000 is was way off.
Ten years on and it's still the best RU-vid vid on this fascinating subject. Thanks Marcus. Here's a thought experiment that occurred to me. Get a couple of dozen ordinary people to each shoot an arrow or gun or whatever as accurately as they can at a small spot of light projected on a barn door or similar. Afterwards switch the light off, and deduce where the target was just from the grouping of punctures. I'm pretty sure you'd be spot on just by finding their centre. Would this mean the crowd is collectively a better shot than the individual?
I'm not sure I understand your question, but what you might be referring to is what engineers call noise with non-zero mean. That means that when averaging all measurements, the noise cancels out but with a biais. For instance if humans had a tendency to underestimate things, the average number in the experience in the video would be much lower. But apparently, at least when it comes to beans, humans seem to be pretty unbiaised estimators and thus the "noise" they produce has a mean of zero :)
I used this principle today to guess the weight of a pumpkin at a work event. The person to guess the weight correctly or guess closest to the actual weight won a $10 gift card. Pretty cool. There were only 17 guesses when i averaged and put my guess in. Even crazier is my guess was right on the money.
I think he meant the .5 bean one, because it was added up to something and .5 bean, which is half a bean. only after it was divided by 160 that it was caused by fraction. The .5 was from someone guessing half a bean.
@oglommi i reccomend watching all 3 parts (1 hours each) from The Code. Can find it on BBC website. It's all pretty amazing! and i was wondering the last few days, why don't we guess alot of things in this way? Like howmany planets with life you think there are in the universe? Ask 1 million people, take the average, and we will know it :)
Which shows that it's pretty likely there's some bias in there. If the average is sensitive to one answer, if you get very close to the right answer, you're lucky or you're cheating, even if subtly and unconsciously (like second-guessing the original 80,000 guess that she gave). If he hadn't biased the girl's original response of 80,000 down to 50,000, the answer would've been ~4702, which is off significantly. This subtle guiding of people's answers when you know the right answer is cheating.
It kind of proves that out brains are capable of making accurate guesses, only with large errors on top of the accurate guesses which go equally either way.
Is this group thought or a group influenced. If the woman at 2:00 had stuck with her original 80,000 instead of reducing it to 50,000 then the math would have worked out differently. When she submitted to doubt she reassessed her guess to 50,000 (its even more interesting why she did? It is a trivial question, maybe it is human nature to second guess your self.) So the math 722.383.5 / 160 = 4,514.896875 (eureka!) However if she had stuck with her original 80,000 guess it would have been 752,383.5 / 160 = 4,702.396875 (not so eureka) So influence must play a part in this, the interesting thing is why did she choose to reduce the number instead of increasing.. for example if she said 100,000 then it would have been 772,383.5 / 160 = 4,826.396875 (not eureka!) ---- Is this group though or individual influence over the group that creates this mathematical picture?
Well, even if she said, 100,000, averaging it to 4,826 isn't that bad for such an awful guess. I would think you'd need a larger group guessing to offset stupid guesses.
What if someone guesses a googol as a joke? Even if there are a million guesses, that one guess will push up the average 10000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000 beans, which is pretty far form the real answer. You could take the logarithms of the guesses, but then someone might guess a googolplex.
im actually kind of surprised there is no response trying to duplicate this and se if it came true for them too, i myself want to try it now after seeing this
No. We call that a normal distribution. Had he analyzed all of the guesses, he would have probably found out that guesses close to the true amount were the most frequent.
@timb6 I've seen all of them. The planets with liffe thingy would'nt work because everybody would just say random numbers. Wich is not the same as this video.
It's interesting how close the average was to the "real" answer ("real" because the experimenter could have made a small counting error either way also). However, I would also like to see what the standard deviation (StDev) was. If the average was less than a StDev or so from the answer, we could say they were "right on." How wide was that range? I mean, an Avg of 4515 with a StDev of 20 is much more certain than an Avg of 4515 with a StDev of 200. And yes, I came here from VSauce. :)
So, would the answer become more accurate with a larger number of guesses? Or worse with smaller group guessing? I go with more accuracy with more people. Thoughts?
I think that the number of people does make a difference. If we only take two guesses, for example, 400 and 50,000 then the average is 25200 (I know those guesses are outliers but it makes the point). If we graph the accuracy with the number of participants, I'd expect an s curve, but that's just a guess as well.
If you remove the 50k guess and divide by 159 it's 4228 to the nearest bean, still only 6 % away and if you remove the highest guess you should also remove the lowest.
Actually, it would've been 4229 - assuming her guess is removed entirely, the total is then divided by 159 rather than 160. But the point still stands. :P
"it's incredibly difficult for anyone to guess how many jellybeans there are.." 2:19 Guessing is actually very easy, getting it accurate is the hard part. Thanks.
Seems interesting.. however, the girl that was hesitant in her number choice coincidentally allowed the experiment to be a success. If she had chosen one of the other numbers that she guessed first, the final number would be off from 4510.