The Bayesian Trap

  visningar 2,538,829

Veritasium

3 år sedan

Bayes' theorem explained with examples and implications for life.
Check out Audible: ve42.co/audible
Support Veritasium on Patreon: ve42.co/patreon
I didn't say it explicitly in the video, but in my view the Bayesian trap is interpreting events that happen repeatedly as events that happen inevitably. They may be inevitable OR they may simply be the outcome of a series of steps, which likely depend on our behaviour. Yet our expectation of a certain outcome often leads us to behave just as we always have which only ensures that outcome. To escape the Bayesian trap, we must be willing to experiment.
Special thanks to Patreon supporters:
Tony Fadell, Jeff Straathof, Donal Botkin, Zach Mueller, Ron Neal, Nathan Hansen, Saeed Alghamdi
Useful references:
The Signal and the Noise, Nate Silver
The Theory That Would Not Die: How Bayes’ Rule Cracked the Enigma Code, Hunted Down Russian Submarines, and Emerged Triumphant from Two Centuries of Controversy, by Sharon Bertsch McGrayne
Bayes' theorem or rule (there are many different versions of the same concept) has fascinated me for a long time due to its uses both in mathematics and statistics, and to solve real world problems. Bayesian inference has been used to crack the Enigma Code and to filter spam email. Bayes has also been used to locate the wreckage from plane crashes deep beneath the sea.
Music from epidemicsound.com "Flourishing Views 3"

Kommentarer
Zachary Wiebe
Zachary Wiebe 10 timmar sedan
I want full netherite armour.
Tejas Chavan
Tejas Chavan Dag sedan
Holy ship! I never thought about this while studying Bayes theorem 😵😵
Nathaniel Cabrera
Nathaniel Cabrera Dag sedan
Were you talking about COVID ? :0! Nice Video!
Rokas Godel
Rokas Godel 3 dagar sedan
5:55 What you said in the beginning is that the test identifies true as true in 99% of cases and identifies false as true in 1% of cases. That is not the reported accuracy of the test, sir, nor does it give us any insight into it, please do not confuse us. To tell us anything about the accuracy you need to convey in some form how the testing results measure up against the determined actuality of the situation. Unsolvable problems which is what you gave in the beginning ought to come with a warning... As for a conclusion (without getting into metaphysics, but sticking to the topic at hand) I feel a better one would be that, given how complex and multi-dimensional of a system life as a whole is, we frequently overestimate our ability to identify the determining factors of an outcome and thus become way too confident about it's overall probability while in reality we may only be aware of a very thin slice in the equation.
Roger Lundström
Roger Lundström 3 dagar sedan
Hmm.. As you described it, it is a self-fulfilling prophesy, but it's quite often that the problem is that people don't see the flaw in what they were doing when they were testing, so they do new tests that they intend to make it more or less certain whether the first test gave the result they thought they get.. but the same flawed stipulation, or prejudice is used as a base for the second test too, so even if it looks from the problem in many ways differently, you haven't fixed the underlying problem. So.. What I would say is "Asking the Question".. Sometimes the answses depend only on the current knowledge and potential for new ideas.. it doesn't have to be that you are LOOKING for a specific answer, but currently available data, ideas, technology, etc, makes you blind for something.
Bixbe
Bixbe 3 dagar sedan
I like that you did not provide an answer but set the stage for further consideration.
I leave mean comments
I leave mean comments 4 dagar sedan
"she suggests..." There's your problem right there... you have a female doctor.
VD Songs
VD Songs 6 dagar sedan
well I love veritasium and no disrespect while saying this. when he took p(H)=.001 he took in the factor of people who are not showing symptoms nor going for a test. I guess p(H)=probability of people that are truly diagnosed with this disease out of total tests done.(I know tests are 90% accurate but it is hard to determine exact number so we can start from this number ) what do you guys think about this? this will make the p(H|E) way closer to 90.
Vinayak Joshi
Vinayak Joshi 7 dagar sedan
You predicted covid 19 man!!!!
Sebastian Contreras
Sebastian Contreras 7 dagar sedan
Its that an UFO at 8:25 on the top corner? ? check it at speed 0.25
Quasar Supernova
Quasar Supernova 8 dagar sedan
Derek is the best. He is so articulate, erudite and handsome.
brightmatter -
brightmatter - 8 dagar sedan
Could this BE any more relevant in 2020?
George Tosounidis
George Tosounidis 10 dagar sedan
Grats, very good explanation of Bayes.
you2tooyou2too
you2tooyou2too 10 dagar sedan
re 2:10 "Incredibly"? Unbelievably? Perhaps he means unexpectedly, or even surprisingly. 1 in 11? I would not cross the street blindfolded with those odds! re 9:10 We should be grateful for 'black swans', even if they are Trumpian. (But not Nazi?) The nature of life is that new things happen spontaneously. How hard should we push for more change than 'nature intended'?
How simple it is
How simple it is 10 dagar sedan
The disease example is actually a question in class 12 maths ....
Gaurav
Gaurav 10 dagar sedan
Yes... probability chapter
gumikebbap
gumikebbap 11 dagar sedan
I feel like ancient greeks used to talk about math like that, going on walks, and taking a stop to draw in the dirt with a stick
Prince Kumar Mahto
Prince Kumar Mahto 11 dagar sedan
I'm proud i did not think that chances of getting disease is 99% ,
Ethan Hollander
Ethan Hollander 11 dagar sedan
Great video. Thanks. Can you tell us the filming location in the video notes? What a beautiful spot!
smokey04200420
smokey04200420 11 dagar sedan
Nah you gotta take into account your symptoms. You wouldn’t be testing 1000 people and only out of 1000 one will be infected. You’ll only be testing those with symptoms. More like 1 out of 50 who take the test will have the disease. If you test positive, then you likely have it.
Jason Perry
Jason Perry 11 dagar sedan
Chances are 100% that you should get a different doctor - one that doesn't require theorems!
SolREI Studios
SolREI Studios 11 dagar sedan
Bingo. You have no idea how important Bayesian theory is relative to Langer Epistemology Errors. When we encapsulate interpretive models Bayes techniques illuminate in an illustrative way. Where this all leads is nothing short of unification.
Рома Любченко
Рома Любченко 11 dagar sedan
Привет Veritasium. Смотрел видео в переводе на русский. Оно меня очень вдохновило и заставило задуматься. Так сказать - взорвало мозг. Спасибо за годный контент)
Tropin
Tropin 12 dagar sedan
you also have to factor in FALSE NEGATIVES (but that is if you test negative for the disease not if you test positive)
Steven Schmidt
Steven Schmidt 12 dagar sedan
Seems like this is a good video to review, with all the Covid testing happening everywhere.
Chris Mason
Chris Mason 12 dagar sedan
This feels so relevant to wants going on now.
Cinnamon Kittamon
Cinnamon Kittamon 12 dagar sedan
I'm at about 1:00 and decided to figure it out myself. Let's see if I got it right. So we have 3 different percents 1. 0.1% of people have it 2. The test correctly identifies 99% of positive cases 3. The test incorrectly identifies 1% of negative cases With such, since 0.1% of people have it, that means 0.099% of people both have it and actually get a positive result. Then, we have the 99.9% of people who don't have it. If the test incorrectly identifies them as positive 1% of the time, that means 0.999% of people get a false positive. Since we know the character in this has a positive result, we only need to focus on the positive result cases. 0.099% of the population gets a true positive and 0.999% of the population gets a false positive. This means a total of 1.098% of the population gets a positive result. If we then divide the true positives by total positives and multiply by 100 (since the result is a drcimal) we see that only roughly 9% of positive results actually have it. Final answer before watching: You have about a 9% chance of having the disease. Now let's see where I messed up lol. Will edit this comment after watching Edit: just a minute later I find out I got it right what. I guess you're just going to have to trust me that I actually did this before he did, but wow that was pretty cool to find out i did my math right
Vibhav Deo
Vibhav Deo 6 dagar sedan
don't you think it is inaccurate to take p(H) = 0.1% it should be more of number of positives out of number of tested...hence the p(H|E) will turn out to be much higher
kpw84u2
kpw84u2 14 dagar sedan
As for your last thoughts here... applies to addiction recovery thoughts too... that's something I've been thinking about this applies to... "trying the same thing over again while expecting different results".
c l
c l 14 dagar sedan
Where is this hike? I wanna go hike there
Luc Chapon
Luc Chapon 14 dagar sedan
Heh, I watched the very same topic a few years back here, in french. Also Mr Phi gives this disease a name, the smurfitis. Don’t play with you smurf! :) That was a memorable moment! svfrom.info/history/video/Y6mIpo-xqrFxe7o
Curtis Confuz
Curtis Confuz 16 dagar sedan
awesome making education good!
Purab Patel
Purab Patel 16 dagar sedan
This aged well
Christian resources
Christian resources 16 dagar sedan
True for P and E Not equal to 0 as would make it undefined.. And
Akshay Rawat
Akshay Rawat 16 dagar sedan
great video.. wished someone had explained like this in my probability class
Roy Long
Roy Long 17 dagar sedan
would recommend watching this video together svfrom.info/history/video/eL2Ad6e6rJiSrq4
Yoni Dellarocha
Yoni Dellarocha 17 dagar sedan
Who's here after Elon Musk's tweet? For those who don't know, he said that after having symptoms he took the same test 4 times, in the same day, with the same machine and the same nurse; 2 results came back positive and 2 came back negative.
n1kho1as
n1kho1as 18 dagar sedan
Can we know the time of the day by the image with your shadow?
Cooper Roberts
Cooper Roberts 18 dagar sedan
Damn bro thanks so much
Shreshtha Alambayan
Shreshtha Alambayan 18 dagar sedan
6:46 *me thinking about shadow.* 🙄😅😁 Imagine a light traveling and send it's wave which travel some distance to wall it strike. *suddenly after a complete traveling of a light wave I put my hand.* (whatever opaque thing) in between light source and the wall. *Will my shadow will not illuse for very very short time?* As remaining light wave behind hand when a light wave travel still there with speed c but takes time by our perspective _(acc to light's perspective there is no distance so speed and no time) so very very very small time no shadow or????_ plz tell
Kamran Majid
Kamran Majid 19 dagar sedan
Hi
Lucas Perez
Lucas Perez 19 dagar sedan
Veritasium guy at 0:18: "tested positive for..." Literally everyone in 2020: "Cuhronuhvaaarus?!"
Tomasz Płókarz
Tomasz Płókarz 20 dagar sedan
So today Elon Musk said on Twitter that he had 4 PCR antibodies tests and he got 2 positive and 2 negative results. 2 weeks after FDA's warning about false positives in these tests.
Black Rider
Black Rider 20 dagar sedan
thanks your videos are amazing
BARTU ONAT ARIK
BARTU ONAT ARIK 21 dag sedan
great video greaaatt video with great ending. Perfect veritasium ending
Brett Evill
Brett Evill 21 dag sedan
I feel compelled to mention Cromwell's Rule: "I beseech you in the bowels of Christ to think it possible that you may be mistaken". Never set priors to zero or unity in any field that you actually plan to think about.
Laura Marin
Laura Marin 21 dag sedan
I'm in a highly competitive field and deal with quite a bit of impostor syndrome, and I really needed to see this today, thank you!
Asif S
Asif S 21 dag sedan
So Derek, would you consider doing this calculation for the pcr data for Covid19? :) or would that be too hot a topic to touch?
John Shaff
John Shaff 21 dag sedan
In the top five best SVfrom videos I’ve ever seen. Pure poetry. Science with social impact. Really great 🙏🏻
Matthew Mitchell
Matthew Mitchell 21 dag sedan
Particularly relevant now with the false positives of SARS-CoV2.
Norton Gartino
Norton Gartino 22 dagar sedan
Thank you
wertyuiop
wertyuiop 22 dagar sedan
For God so loved the world, that he gave his only begotten Son, that whosoever believeth in him should not perish, but have everlasting life.
Khashayar Motarjemi
Khashayar Motarjemi 22 dagar sedan
You are a good person.
Romain Puech
Romain Puech 22 dagar sedan
😁
Valentina Jerenec
Valentina Jerenec 23 dagar sedan
Oh yes! Thank you for that!
Rishabh Verma
Rishabh Verma 23 dagar sedan
The equation looks like PewDiePie thingy
radicaledward84
radicaledward84 23 dagar sedan
99% of those who have the diesase and 1% of those who haven't. Wait a minute! ;)
Chun Chun
Chun Chun 24 dagar sedan
The explanation starting from 2:20 to 2:56 is very good and intuitive, but there is a wrong calculation at 2:51. If you do the math, the correct chance is 0.090163934. However, at 2:51, the calculation of 1/11 is equal to 0.09090909. This is because, even though there are a total of 11 people, "the chance of being a true positive case (as a person of the 1/11 people)" ~ is different from ~ "the chance of being a false positive case (as a person of the 10/11 people)". So you cannot simply say your chance is 1/11. You need to weigh each chance for the correct calculation.
seth mortimer
seth mortimer 24 dagar sedan
I’d like to change always working in low paid jobs
Dave P
Dave P 24 dagar sedan
Does anyone else see the UFO starting at 6:42??
J Denmark
J Denmark 24 dagar sedan
Thanks for an excellent explanation of this subject. Too often in statistics classes do instructors forgo introducing this and when brought up, don’t really understand the process and what it means for research. For most students it’s beyond intuitive understanding, even though they do use it sub consciously for making the same decisions in life because they don’t change the approach.
Sai Prabhav
Sai Prabhav 24 dagar sedan
i guess the testing is true for Corona
Anupam Bam
Anupam Bam 25 dagar sedan
Superb PoV!
David Caster
David Caster 26 dagar sedan
There is no rational basis for the assignment of probabilities to novel events about which you have no prior information. There are however and unlimited number of irrational bases for assigning such probabilities. This is the classic zero-frequency problem that appears in, for example, data compression algorithms and, apparently, spam filters.
peter De Pauli
peter De Pauli 26 dagar sedan
can you use a different cameramode/selfi mode next time, I got "seasick/motion sickness" watching this stablisation/warp thing effect. :)
Michael Roditis
Michael Roditis 26 dagar sedan
what is the probability of having positive results for covid-19 if you haven't got it ?
Matthew Mitchell
Matthew Mitchell 21 dag sedan
In the UK, the government doesn't even provide this information. How can you begin to trust any data?
Camcongab
Camcongab 27 dagar sedan
Baye seemed to have a strong philosophical influence on his work.
Praveen Kulkarni
Praveen Kulkarni 28 dagar sedan
just be uncertain about your priors :)
cocktailpost
cocktailpost 28 dagar sedan
This has huge implications in the political thinking of humans, that's why those above are so busy trying to get us to think some things are just good or bad without any reasoning
satyris410
satyris410 28 dagar sedan
To be fair, Laplace might not get a theorem, but he does have a transform and an equation named after him.
amireza sobhdel
amireza sobhdel 28 dagar sedan
this video would be a great teaching example for all teachers around the world
João V M L Silva
João V M L Silva 29 dagar sedan
Error matrix be like:
Ronen Martinez
Ronen Martinez 29 dagar sedan
2:55 There is one imposter among us.
Mattias Sigurdsson
Mattias Sigurdsson 29 dagar sedan
The example doesn´t count for the normal case when we are being tested due to symtoms or genetical disposition. Where there is a pre-selection of the persons being tested. For those people - a vast majority in a normal situation - the probability of having the disease, pre-testing, is a lot higher than 0,1 %, right? Hence the accuracy is a lot higher. But i surely cast a shadow of the idea of broad testing, if it´s not for to get a general, pandemic picture. For the individual it´s not very useful.
George Chirayil
George Chirayil Månad sedan
I don't know what it is, but something about seeing you walk along that overwhelmingly beautiful landscape with the moon's great presence in the sky floating over everything. I can't describe it. This looks like something from thousands of years in the future... but it wouldn't be planet earth that you're standing on- instead a new world we've inhabited
Erin Butler
Erin Butler Månad sedan
It's why philosophy is an essential study.
J G
J G Månad sedan
I love statistics just as much as I love boolean algebra.
Yang Cui
Yang Cui Månad sedan
Hi, I like your video. Could you let me post it to my aiqiyi.com account in China?
James Hall
James Hall Månad sedan
Blows my mind, ever time. I've watched this video 40, maybe 50 times by now and it always sends me into a self-reflective spiral. Thank you Veritasium, thank you.
uzijn
uzijn Månad sedan
Using Bayes Theorem, what's the probability of receiving a single positive COVID-19 result meaning you have COVID-19?
Rubaeth Hossain
Rubaeth Hossain Månad sedan
I just realized... Veritasium 420 XD
griffin6002
griffin6002 Månad sedan
The problem with Bayes, which I've studied extensively and use every day, is that it can't be applied equally well to all types of data. Essentially, Bayes is an average of an average. The thinking is that the average of the average will weed out all the outliers, but that doesn't work if the datasets that the original average were derived from were faulty. Bayes might work relatively well with diseases that have clear cut outcomes like terminal diseases, but the best example of Bayes NOT working is the 2016 Election. All of the polls that showed Hillary Clinton winning were wrong because their datasets were fundamentally flawed, thus taking an average of them yielded an incorrect result. Want to know one of the major flaws of COVID policy? Double counting. We were told that we were short tens of thousands of hospital beds and ventilators because we were going to have the regular flu season patients PLUS COVID patients. Do you see the double counting error? COVID patients are often the old frail patients who would have needed that bed or ventilator anyway because they are the ones more likely to get the flu. We were never short ventilators or hospital beds. COVID and flu patients, which are often one and the same, were double counted. Using Bayes for COVID potential morbidity is actually ridiculous because there is no clear cut outcome. Asymptomatic people never get tested because they don't even know they have a disease. People who die WITH COVID are often tabulated as dying FROM COVID. (millions of people die WITH warts every year, but they don't die FROM the warts). Again, averaging averages with Bayes is useless if the datasets are faulty to begin with. But... there are a lot of people making a lot of money pushing Bayes and other statistical snake oil on government officials and corporate executives who don't want to admit they have no clue about statistics.
Donald Fink
Donald Fink Månad sedan
This was great, thank you!!!
Bobblonia
Bobblonia Månad sedan
Welp, he has correctly predicted COVID, uh yeah that's kinda creepy
RamBoZamBo123
RamBoZamBo123 Månad sedan
Tell this one to all the Corona panicers.
BigMackdombles
BigMackdombles Månad sedan
7:30 hit home. I always thought bayesian models was more or less HOW people thought; tbf I learned about it for the first time in a language developement course. But your notion of it as problematic to someone's development or self appraisal is very insightful V.
Arpad Beszedes
Arpad Beszedes Månad sedan
A lot of comments are talking about flaws in the approach as it is explained. Here is another one: the test can make a mistake in two ways: false positive and false negatve. The beginning of the video talks only about false negatvies, then in the Bayes formula 99% (1%) was used in two different contexts, so it is assumed that this error rate is the same for both cases.
Pierre C
Pierre C Månad sedan
Doing the same thing over and over again, and expecting different results means you take Bayes Theorem into account, and are probably right!
Dan Rapp
Dan Rapp Månad sedan
Bayes' Theorem = universe-sized game of You Sank My Battleship!
Sundar Subramaniyan
Sundar Subramaniyan Månad sedan
6:53 Look up at the top left of the screen. It looks like a UFO. Now what’s the probability of this hypothesis being true given that it’s a sunny day?
Rahma Tasnim
Rahma Tasnim Månad sedan
not me watching this right after my statistics midterm
Feben H
Feben H Månad sedan
Interesting. I have no math knowledge but I got the right answer, because it just seemed like common sense, that it'd be about 9% chance.
No Body
No Body Månad sedan
*Meteor lands on a person, killing him. Bayes' Theorem: "There is only a 9% chance that he was actually hit by a meteor." Cold Hard Facts: "Hold my beer..."
Walter Zagieboylo
Walter Zagieboylo Månad sedan
Maybe it would be a good idea to make an experiment in reading the Book of Mormon. It has a promise that if you read it with sincere intent, then God will show the truth of it to you. If you have 100% certainty that there is no God, then never mind.
Québécois François Game
Québécois François Game Månad sedan
no result repetitives : datting apps
janofb
janofb Månad sedan
Corollary to Mandela's theory: No one's a murderer until their first murder.
Soham Dhamnaskar
Soham Dhamnaskar Månad sedan
i guess this also applies to covid tests
Vaibhav Deodhe
Vaibhav Deodhe Månad sedan
When I was watching this, I felt like I understood the bayesian trap, but now after video ended, I feel like a fool because I can't quite articulate the same. I'm starting to questioning my memory retaining abilities or is it natural because I'm not familiar with topic itself.
yola1029
yola1029 Månad sedan
Air pollution looks crazy far away in the background. That city's having a nice bath in a proper dark grey cloud.
Hello Kitty Lover Man!
Hello Kitty Lover Man! Månad sedan
"Thought to himself"? As opposed to... that he got telepathy (thinking to someone else) down pretty well?
Jimmy
Jimmy Månad sedan
What? 1 in 1000 is not 1%. The whole premise of the explanation is wrong.
Fubu Maruda
Fubu Maruda Månad sedan
8:55 It's Vaas's definition of insanity. Doing the same thing and expecting a different result.
Bayes theorem
15:46
3Blue1Brown
visningar 980tn
HAYATI
3:17
Haval - Topic
visningar 82tn
Dude Perfect Corn Maze | Nerf Battle
6:54
Dude Perfect
visningar 9mn
Is Most Published Research Wrong?
12:22
Veritasium
visningar 2,6mn
Bayesian Statistics with Hannah Fry
13:48
Stand-up Maths
visningar 211tn
World's Lightest Solid!
12:02
Veritasium
visningar 24mn
Half the universe was missing... until now
14:10
The Infinite Pattern That Never Repeats
21:12
Feynman's Lost Lecture (ft. 3Blue1Brown)
21:44
Could We Prevent An Impact?
20:06
Veritasium
visningar 852tn
Why do prime numbers make these spirals?
22:30
What I Wish I Knew When I Was Younger
08:51
Veritasium
visningar 2,1mn
How Bayes Theorem works
25:09
Brandon Rohrer
visningar 391tn
HAYATI
3:17
Haval - Topic
visningar 82tn
Dude Perfect Corn Maze | Nerf Battle
6:54
Dude Perfect
visningar 9mn
96 Hours Inside Afghanistan in 2020
33:56
Yes Theory
visningar 2,3mn
Newkid - Starkare - Så mycket bättre (TV4)
2:43
Så mycket bättre
visningar 286tn