Why Mathematicians Won't Help Cops
Vložit
- čas přidán 6. 09. 2022
- Algorithms are so amazing at predicting what you want that ads will be perfectly tailored to your interests and displayed to you just minutes after a complex system has basically mapped your mind. If that’s so easy, why can’t police predict crimes? A few companies claim they can -- but the truth is a lot more complex.
A decade ago, predictive policing algorithms were hailed as one of the most important inventions of our era. In just a few short years, thousands of mathematicians had publicly refused to work on predictive policing projects. What started a century ago in the concentric zone model of mapping urban areas and evolved into social disorganization theory has culminated in sophisticated algorithms that claim to pinpoint the place and type of crime well in advance.
The problem, though, is that algorithmic outcomes are only as good as the data going into them. Between flawed data collection/reporting, yet another black box algorithm, and a total inability to measure results effectively, the promise of predictive policing has come under scrutiny -- and the real question is how long we keep experimenting to get it right, and who we run those experiments on.
** ADDITIONAL READING **
“Precise Event-level Prediction of Urban Crime Reveals Signature of Enforcement Bias”: www.researchgate.net/publicat...
Letter, “Boycott Collaboration with Police”: www.math-boycotts-police.net/
PredPol’s Predictive Policing Algorithm: www.predpol.com/technology/
“Predictive Policing Software Is More Accurate at Predicting Policing Than Predicting Crime,” Ezekiel Edwards: www.aclu.org/news/criminal-la...
Chicago’s Strategic Subject List Dataset: data.cityofchicago.org/Public...
“What Can FBI Data Say About Crime in 2021? It’s Too Unreliable to Tell,” Weihua Li, The Marhsall Project: www.themarshallproject.org/20...
"Abraham Wald's work on aircraft survivability". Mangel and Samaniego. June 198, Journal of the American Statistical Association. 79 (386): 259-267: people.ucsc.edu/~msmangel/Wal...
“LAPD data programs need better oversight to protect public, Inspector General concludes”: www.latimes.com/local/lanow/l...
“UCLA Study Proves Predictive Policing Successful in Reducing Crime Over Several Months of Deployment with The LAPD”: www.predpol.com/ucla-predicti...
** LINKS **
Vsauce2:
TikTok: / vsaucetwo
Twitter: / vsaucetwo
Facebook: / vsaucetwo
Talk Vsauce2 in The Create Unknown Discord: / discord
Vsauce2 on Reddit: / vsauce2
Hosted and Produced by Kevin Lieber
Instagram: / kevlieber
Twitter: / kevinlieber
Podcast: / thecreateunknown
Research and Writing by Matthew Tabor
/ tabortcu
Editing by John Swan
/ @johnswanyt
Huge Thanks To Paula Lieber
www.etsy.com/shop/Craftality
Vsauce's Curiosity Box: www.curiositybox.com/
#education #vsauce #crime - Věda a technologie
So we actually invented Minority Report, but the people who invented it said "no" and refused to keep developing it. That's actually weirdly reassuring.
The good news of sorts is that since the police departments are allowed to refuse to hire cops that are too smart, the likelihood of them managing it on their own is rather weak. I think this was probably going to wind up being used as justification to continue targeting people in poorer neighborhoods that lack access to properly funded legal defense funds.
Yeah, but on the downside we totally did build the racist police robots in the first place, and because they target poor people in urban areas, more felony arrests result, meaning fewer blue voters, meaning more racist politicians, meaning more invested into the racist robots we already built.
The only way to fix this process is to start divesting from police entirely
@@isaacgeorge7773 yeah
That’s what happened…
@@isaacgeorge7773 sooo...there's blue politicians and racist ones? That doesn't sound like a bigot at all...Definitely not...
@@maskettaman1488 Crime is based on proximity and opportunity not race. Choose your words wisely, its not that hard.
Sounds like a great example of a “self-fulfilling prophecy”. Crime will be found wherever law enforcement wants to look for it.
partially. that might fit for possession of drugs but not for crimes like murder, the crimes most people actually care about. there is a specific group of people who commit most violent crime BY FAR. If you look at other demographic groups you will not figure out that a lot of murders happened there too which you just didn't notice because you were not checking.
@@Dorfjunge I’ll preface this comment with the fact that I’m speaking specifically to the dynamics of criminal law enforcement policies of the United States. Murder doesn’t normally fit the patterns they cite of predictive repeat offenses, and to the extent this sort of modeling can be applied to capital crimes, there is a clear, well-documented correlation between the destruction of communities and destabilizing forces of systemic racism and historical exclusionary actions that generate the types of areas that are subjected to higher than average levels of violence and those correlations are directly tied, in part, with over-policing. It’s too complex to just ignore the biases that infect our demographic makeup, even when referencing more serious criminal behaviors.
@@Dorfjunge Dorfjunge, der deutsche Benutzername der ohne (vermutlich) jemals in den betroffenen Gegenden gelebt zu haben von der halben Welt entfernt mit seiner rassitischen Linse das Problem akkurakt erkennt. Wunderbar
In english: A German username "accurately" diagnosing the issue through his racist lense from half the world away, wonderful
A paraphrased update of Russian Secret Police Chief Beria “show me the location and I’ll show you the crime.”
@@stevekru6518
This. The painful reality is that crime-as it is practiced in the real world outside of the dried ink of legislation and theory, is an interpreted definition that relies on the individuals enforcing the laws and within every level of the criminal justice process to make anything from a wild guess to a considered judgement call before someone is investigated, accused, arrested, indicted, tried, and/or convicted. That is the dataset. It’s impossibly subjective, and tends to directly reflect implicit biases of each one of those individuals-from the citizens reporting, witnessing, testifying, or sitting on a jury to the police officers to the district attorney and judge. There is no objective dataset for crimes simply because there is no objective definition or practice with which to produce one. I say that as an observable fact, and an unfortunate but necessary reality-not as a criticism or to level any value judgement towards it. I have personal opinions about the ways in which that subjective power is frequently wielded, but that’s an entirely separate issue.
I love how they "publish the algorithm on their website" and it's just an equation with no terms defined. It's clear their intended audience there does not include anyone who knows what they're talking about.
if it's patented as they claim then it should be possible to look up the patent to get more information on the algorithm. I'm surprised that didn't get addressed in this clip.
Also, an equation is not an algorithm, right?!
@@AubreyBarnard in a way, it is. An algorithm is simply a set of instructions. When you think about what an equation is, especially where one side is a single term, it's a set of instructions on how to arrive at the quantity described on a given side.
As for the undefined terms, that's not really an issue either. You can look at the orders of terms and the relationships between them to determine what kind of curve the results produce. That can be helpful in its own right.
The equation has OwO at the end, I doubt it is even real.
@@taodivinity1556 that is funny, but i think it was theta omega sigma. theta is commonly used as a variable and the omega is sometimes used as a coefficient or constant. dunno about the sigma though.
One of my favorite things about crime is that almost everyone is doing something that’s *technically* illegal during their day-to-day life. Therefore the likelihood of crime is almost guaranteed to be 100% everywhere. Put a cop there who wants to enforce a specific law and they will find someone committing it.
@@ody5199 No. The studies are.
@@ody5199 do you know where the term jaywalking came from? It used to be that roads were for everybody, bikes, pedestrians, horse-drawn carriages. When automobiles were invented only the rich could really afford them. These people would go to a pub, drink heavily and then drive their cars very. Many times people died, because of course they did, drinking and driving is dangerous. Lots of people called for banning these "metal monstrosities". And so the budding automobile industry worked to make a law that said "streets are for cars". They gave this law the name "jaywalking" because a "Jay" was a slur used against people native to rural mountainous areas. It was basically like every other racial slur you can think of.
And they got it coded into law. They got a slur to be the name of a law.
This is another type of enforcement bias. Because it is enforceable you believe that it's wrong. When, jaywalking was just a disinformation campaign by a bunch of rich people who wanted to sell cars.
There's a great episode of "Adam Ruins Everything" that has sources for all this information.
This shouldn't even have been a crime in the first place.
@@ody5199 to be fair, nobody in the US knows where the term jaywalking came from either.
@@ody5199 nope, it’s just how these models are made, **they** compare jaywalking to murder using them.
@@ody5199 considering both can get you shot if you’re the wrong color, I think it’s apt
The whole plane analogy reminds me of how the US moved to steel helmets in WW2 and saw a drastic increase in head injuries compared to WW1, what this really meant was that the helmets were working, leaving more people injured instead of dead
First, you are really confusing that story.
Second, the story was almost certainly not true.
The story is that the British identified an increase in head wounds when they started issuing steel helmets to soldiers in WWI and initially generals wanted to withdraw the helmets because of the increased wounds until someone pointed out the reduced deaths. The US doesn't make sense as they only joined the war after steel helmets were the norm and troops were provided them by the French and British before serving at the front.
From what I looked up before, the earliest known reference to this story is using it as a hypothetical example of such effects, not saying that it was real and there isn't any evidence shown that it really happened (and they meticulously recorded most of this type of stuff).
And while on the face of it it seems logical that there would be a significant increase in head wounds, that isn't necessarily the case. Yes, steel helmets would reduce some fatal head injuries to head wounds, but they would also reduce some head wounds to nothing worth noting. I don't know whether there would be a net increase or decrease in head wounds, but it isn't necessary that it would be and if it is an increase it may not be that large.
I believe it was the British and not the Americans who first used steel helmets in WW1 and almost decided to stop using them due to the increase in head injuries.
@@Namminamm Technically the French were the first to adopt a steel helmet. They first had a stopgap head protection deployed in early 1915 with the Adrian helmet entering combat in July 1915 while the Brodie was issued in September 1915. The Germans wouldn't issue their helmets until early 1916.
But all three were developing helmets simultaneously as the need was pretty obvious and the concept of a helmet was well known.
And the development time kind of shows. The French Adrian helmet had serious issues, both in design and the materials used.
The British Brodie helmet was the easiest to mass produce and arguably the best for trench warfare.
But there is a reason that most future helmets resemble the Stahlhelm more than the other two. It was clearly the best overall helmet design and far more ergonomic than the Brodie when not standing in a trench.
@@88porpoiseThe French infamously wore cloth caps and brightly colored pants at the beginning of WWI, leading to massive initial casualties because they were completely unprepared for German artillery.
@@BobtheRedead The only major army that didn't primarily wear cloth hats in 1914 were the Germans, who wore leather helmets which weren't any more useful than the cloth caps in WWI. The French were the first to do something about it.
"You know the precogs from The Minority Report? What if we named our predictive policing system in reference to that?"
Someone missed the whole point of cyberpunk dystopian fiction, because they were trying to make it happen
More likely they see those movies as inspiration for how much dystopian they can be before the public catches on
how about psycho-pass level of "prediction" instead? worked out fine!
Sci-Fi Author: In my book I invented the Torment Nexus as a cautionary tale
Tech Company: At long last, we have created the Torment Nexus from classic sci-fi novel Don't Create The Torment Nexus
There's a Chinese tech company called Skynet...
same w/ peter thiel and palantir
ReplyAll (RIP) has a great piece on COMPSTAT, the predictive policing system used by NYC and many other NA cities. It's called the Crime Machine, and I highly recommend it. Last I heard, like a dozen cops were suing the NYPD for their use of COMPSTAT, as they basically had soft quotas to issue a certain number of summonses (minor tickets) each day in specific areas, which then made those areas look more high-risk, which meant they were flagged as areas to return to, etc.
These models not predicting crime, but predicting policing, is a wonderfully succinct way of putting it.
No. No it won't. The police can hang around my neighborhood all day long and they won't catch any яápes because they don't magically make crime happen. This rhetoric clearly does not hold true for predictive policing of VIOLENT crime.
The cart driving the horse. . .
The tail wagging the dog.
I'm confident the mathematicians who do work on such systems knew about such feedback-loop and filter the input-data appropriately. Why wouldn't they know about this often repeated concept?
What gets me is that this isn't a flaw in the modeling, it's a flaw in the NYPD. This software isn't forcing them to have quotas -- they had quotas before the software and they'll have them afterwards.
Thus, the answer isn't to throw out the software, it's to throw out the quotas and those imposing them.
@@bvoyelr I don't get why anyone should a quota for things such as this. What if all you met were the most pleasant people on earth all day and you get reprimanded for it?
I do like the part where they identify that crime often happens due to a combination of environmental and sociological factors instead of completely rational actions by individuals and then rather than trying to address those factors (through changing policies, economics, social work and other non-police support) they just circle it on a map and say "that is where the crime will be."
Who said they don't? This video was in relation to policing, and those involved in the policing aspect wouldn't necessarily know or care how else these systems are being used by other wings of government.
@@bvoyelr What Cops need is to be Defunded and Debundeled. But HOW can we ever fix the System if most people refuse to even learn how to grasp the phrase Defund the police??!
Without the thought-provoking 'Defining Defund the Police'-Video of "Some More News", how can we help?
Yes, it's been known for decades that there are factors that can be addressed that would lead to a reduction in crime. Much of it involves making it harder for criminals to lurk and increase the likelihood of neighbors to be willing to narc on criminals they see. Compton ultimately cleaned up a bunch when the community stopped tolerating the criminal behavior.
Cutting back bushes, improving lighting, removing blighted houses that can't be sold and keeping the height of buildings below a threshold would all reduce crime more than what this proposal would do and without needing to deploy police to potentially harass locals that may not have anything to do with crime in the first place.
@@bvoyelr reality did lmao. They’d rather have a generation in chains than a functioning society
@@SmallSpoonBrigade why height of buildings tho?
quick question: if aura knows if your credit cards, security numbers, etc. are being tracked, doesn't that mean that they have access to that information? Wouldn't that just make the problem worse, because now there is another potential source which could leak your data?
that's what i was thinking, how can they know if my ssn has been found if they don't know what it is?
No. Even if it increases your potential leak sources by +1 if it then decreases the likelihood of those sources will be leaked by 90% it would still be much better. Let us say you have 10 sources and the likelihood any of those is going to be leaked in a year is 10%. That is 1 - 0.9^10 = 0.65 = 65% chance that there is a leak in one year. Now, let us say that Aura reduces the chance of a leak happening from one source down to 1%. Now you have 11 sources as it includes Aura. That is 1 - 0.99^11 = 0.10 = 10% chance that there will be a leak in a year. So down from 65% chance of a leak to only 10% even though you increased your potential sources by +1.
@@bobthegoat7090 Thank God someone used logic
@@antagonisticalex401 Yeah, some janky logic. Aura does not reduce change that some random server will be hacked. It's just imposible.
It only can notify you when a leak is happened and only when the Leak is in public (and yes, so called dark web is public).
As orginal post noted, Aura:s servers will be very sweet target with all that personal information. There is no statements how they store information.
@@MeltedMask while true, if you're notified the moment someone starts trading your info on a semi-public forum you can change your password(s) immediately and thus cut off others access, preventing the loss of more stuff identifying you.
Even _if_ the premise of predictive crime prevention didn't lead to increased arrests and thus an artificial & inaccurate amplification of its own data, it could not be a sustainably accurate system: it relies on big data that "it" (ok, the police, whatever) then tries to suppress! In other words, if it were 100% accurate and led to a true and just prevention of all crimes, well, there goes the entire dataset upon which its model relies. ¯\_(ツ)_/¯ Sure, you could continue to use historical data, but it's a near-certainty that that would become more and more inaccurate as time went on and criminals adapted to the new patterns.
Dynamic causal models -- models that can handle feedback loops and adapt to changing distributions of data over time -- are the types of models that could hope to (in principle) address the data problem you mention. They would address it by modeling interventions (policing, in this case) and how they affect the system. But, of course, those interventions need to be properly modeled (takes experts, time, and effort), the models need to be trained on good data (apparently doesn't exist in the USA), they need to be interpreted correctly by experts who understand them, etc. It seems that the potential for crime in a particular area would have to be modeled as an unmeasureable (latent) variable, but latent variables make querying and training harder, so I suspect many "predictive policing" systems avoid using latent variables, and therefore such systems can't really model criminal potential but only policing (as the video mentions).
Crime disappearing under the correct and just use of an accurate model would not ruin future data. One would continue to collect data on various factors that affect crime, like socioeconomic indicators, prevention efforts, etc., and one would still collect data on various effects of crime, like where police noticed no crime, where people reported no crime, etc. Since criminal potential is unmeasureable anyway, it wouldn't be collected as part of the data. But the upstream and downstream effects of crime would be measured, and they would "sandwich" the criminal potential, allowing people to monitor that potential and adjust if anything changes. For more on this sort of causal modeling, take a look at the work of Suchi Saria.
@@AubreyBarnard It sounds beautiful, but in practice it is utopian thinking. As long as we don't have some objective and perfect data gathering biases will always be included in the data set, as long as biases are included the results will be more biased, as the results are more biased the data set becomes more biased.
The core of this issue is that policing is an imperfect institution. Any algorithm/system that wants to use results from police enforcement will need to either use a different source for gathering data, or, either use more effective correction methods in the institution of policing or integrate the model with a correction facility focused on finding mistakes within the interaction between the algorithm/system and policing in order to prevent escalating degradation in its integrity.
Not the case at all.
One simple algorithm like this is dumb though. The ai algorithms google has are where things get more scary. They predict/know what you’re going to do before you even know yourself.
Ironically, these same algos can predict crime too… we just pretend it doesn’t exist yet. It’s like when you let your own troops walk into an ambush instead of letting the enemy know you know all their moves…
@@hungrymusicwolf It might work *_if_* (big if) we had an algorithm that would automatically "unbias" its data over time. I have no idea if such is even feasible, but that's necessary if you start from biased data (which is obviously all that's available). That's what, more generically, we invented science for - improve, over time, on what us fallible humans can come up with. And we've found that science isn't equally good at every subject and that people will keep gaming the system. Which doesn't fill me with hope for such an algorithm until artificial intelligence becomes at least as intelligent as cops. Which is a long time in the future, it seems.
@@KaiHenningsen It's not feasible because AI is made by people and people are biased, usually unconsciously. You can't unbiased something unless you can agree what unbiased looks like, which is fundamentally ruled out by the Human condition.
The phrase, "show me a man, I'll show you a crime" stands out.
Coming from an Economics background, I always find it weird that a lot of models trying to predict human behavior outside of my field or Decision Theory don't take the agent's available information into account. We've had the concept of an information set from Game Theory for almost a century and people still don't include anything remotely similar into their models. The problem with this is that the agent reacts to the existence of the model and adapt their behavior according to it. Predpol assumes people won't respond to changes in the amount of policing, since their behavior is only dependent on environmental factors. It's eerily similar to macroeconomic models prior to the 50s; they didn't take into account that people changed their consumption/investment patterns depending on what the Central Bank did, "dampening" its effect, so the predicted outcome of a policy was always overestimated.
This was the basis of the fictional science called “psychohistory” in Asimov’s Foundation novels. It was very unreliable when applied to individuals and small groups, but very accurate when the trillions of people on millions of planets in the Galactic Empire were involved. But it had two weaknesses: it could not predict the effect of the existence of a mind reading mutant like the Mule, or of aliens or robots, since the rules for likely individual behavior were based on normal humans; and it would not be valid if its predictions, or the math on which they were based, became widely known, since people could change their behavior based on that knowledge.
Not even all economic models take these things into account.
For many people 'rational expectations' is a slur. And that's on all points of the political spectrum.
This just shows that the police would in fact really use a 'Pre-Crime' unit from Minority Report, if given the chance.
Well in theory it sounds like a wonderful concept. Some people just actually tried to make it work, but it won't.
No it doesn't. Patrolling an area to reduce crime is not what pre-crime was. Pre-crime was about punishing people for a crime they were about to commit. Patrolling a neighborhood is not a punishment.
@@didles123 The police is still trying to find crime before it occurs still. Which means, if they had access to an actual device to find people guilty before it starts, they would call it a success. Imagine fitting the profile of a would-be thief, and be in the wrong place at the wrong time, and gettin' arrested for it, simply because some algorithm told the cops to circle the area a bunch of times because some similar looking street the next town over had a previous statistic. And since cops are cops, they be sketchy and trigger happy and pull you in for questioning, because they're fearing the worse.
@@didles123 Yeah, it's just a fancy way to make a patrol route. I mean, they have to patrol *somewhere* anyway so they might as well patrol areas where there might be crime.
That's because it's easier than having to do their jobs. The main reason why crimes are solved is because criminals are dumb. Smart criminals wind up in the corner office committing more acceptable crimes.
It's kind of the problem with sociology broadly: using data to draw generalized conclusions about individualized behaviors. Addressing the larger problems can have limited benefits, but doing only that - taking the sociology as gospel - causes plenty of problems of its own.
Treating data about large scale problems as info on the large scale problems is quite useful. There are large scale society wide issues we need to address and big data is a useful tool in tackling those problems. So id say the benefits are more than limited. But I fully agree trying to apply it to an individual is absurd.
You're both right ‐ that's where Anthropology & Psychology swoop in to save the day ‐ with a twist of neuroscience! ♥️
i feel like you forgot what sociology is.....
To be fair even those guys focus a bit on the top level of interections a bit too much. As someone studying all of these despite their main field being biology, I actually expect that the next direction for many of these fields will be increased focus on more sub layer type of interections. We actually already kind of see this existing with like the ecological type ideas of development and similar
@@huckthatdish
What Cops need is to be Defunded and Debundeled. But HOW can we ever fix the System if most people refuse to even learn how to grasp the phrase Defund the police??!
Without the thought-provoking 'Defining Defund the Police'-Video of "Some More News", how can we help?
Thank you Kevin - this video actually helped me to present analytics data which had "bad data" so that ... well, let's say: you helped me to explain a mathematical problem without using math language
As a mathematician I have to say that Predictive Policing, similar to AIs and other probabilistic algorithms, can only be the start. If it were to be used, it would require strict laws with clear oversight to ensure that this algorithm is used only as a hint of what could be worth looking into. In a state like the US, this would be impossible for many reasons, and even if mathematicians were working with specific states where it could work with laws already in place that are strict enough, they can't be sure that those laws are actually applied as they are written.
We're too woke now. We're not even at a point anymore where we can assume someone's gender based on physical and biological attributes lol
@@anonymeister123 No one anywhere said anything about gender until you did.
@@softgender The video already implied things about criminal race and gender. Just blind if you didnt get it imo
@@anonymeister123 Is this some weird roundabout way to suggest that racial profiling is actually a positive thing?
@@Parad8n no it's a roundabout way of saying society no longer allows certain valid and relevant datapoints, because of emotions
Love how the recent videos have been about crime in relation to math. Keep em coming!!
@ToxiqSna-iQ skill issue
@ToxiqSna-iQ L bozo
It might be a sign I’m too big a math nerd when as soon as I hear “World war II airplanes” I already know exactly where the video is headed 😂
Because that fact is more recited than the bible.
Ah yes, survivorship bias
nah i swear everyone heard about it because its ... repeated all the time lol
You don't need to be a math nerd 😂, the story is recited a lot which Vsauce admits.
Not even a math nerd here, just a WW2 airplane nerd who could guess why they might get brought up in a math video.
To answer the question at the end of the video, my first thought was "No, I would not want to be part of an experiment like that." but then another, more intrusive thought popped in my mind. "Even if I was part of it, I would not want to know it." and I have two reasons for that. One, it makes me paranoid in a sense, and two, if I knew it, that may skew the data, and ruin the experiment. Huh. Wouldn't want it to be in vain I guess.
Working in data and modelling we deal with this alllll the time.... It would be great to always have great data, but most of the time we have crap data... Should we really be modelling based on the crap data? Difficult question. In my short time working there I've seen hundreds of mistakes based on bad data, including mistakes I've made.
There is a county in Florida that uses this system. The results are pretty much as described. People who have never commited a crime being harassed in their homes all times of day and night. Then filing charges against them that get dropped. Code inforcement violations suposudely postted on peoples doors that missing court date will cause them to be arrested.
For what it's worth, the quote about not finding PredPol effective was from the Palo Alto PD, which is about 400 miles north of the LAPD's jurisdiction. The larger article is regarding the LAPD but not that quote.
Thanks for these videos, Kevin. Thanks for giving people a reason to think before they act. Not nearly enough of that happening anymore
Excellent explainer on why more data doesn't necessarily equal better data. This is one of those cases where it is difficult if not impossible to prove your model is working. How you qualify and quantify violent crime is going to have a major impact on what actions get that number down. You could have no crime if you locked up everyone or had 100% surveillance, but that's awfully Orwellian. Do we want that trade-off?
Singapore has little to no crime.
Part of achieving that is strict punishment for the crime. So it is possible.
One of the issues with data is that you need the right data and the ability to process it quickly enough to act on. So much of the data being collected by analytics firms in general is worthless, they're just hoping that with enough of it, they can look like they know what they're doing often enough to get paid.
Here's the thing, this is actively being used by many law enforcement agencies. A lifelong friend of mine works for a large wealthy US county PD who's sole job is a specialization in predictive policing for repeat offences, open cases & potential crimes. Their techniques are suprisingly accurate. I don't know if they use the same methods but data driven & algorithmic crime prevention has continued to evolve & have advanced when being combined with open source social media data points
My issues are both the potential issues described here & the violations on personal privacy. Preventing self fullfilling prophecies should be paramount.
I cannot tell you how helpful this video was to writing my law review article on Predictive Policing. Really helped me get my start.
"I thought the punishment came after the crime" - Captain America
4:16 "Real quick, if you want to help Vsauce2 end yourself, sign up for a free trial..." Not quite there yet, but give it some time.
this sorta remines me of the decision of a bunch of realty firms to remove neighborhood crime statistics from their platforms, because they noticed that they became a feedback loop of depressing property values in those areas.
Thanks for all these great videos! Always on the back on my mind, especially with the philosophical like questions at the end
great video, and tbh we encounter these dilemas of “being manipulated for a greater good” everyday. It literally reminds me to my country couple days ago that we were passing through a constitutional change, and that had a lot of bias from everywhere, approving or rejecting the new proposed constitution. and everyone was aware of this process to take accountability on their own countries
By the way, congrats in reject that horrible far left constitution in Chile, you dodge a big bullet
@@asysjr lol what
@@asysjr this tells a lot about yourself.
@@asysjr é claro q é um Br alucinando em cocaína, grande surpresa ser um vira-lata dos EUA 🧐
@@joaovmlsilva3509
E quem diabos está colocando EUA nesse assunto?
o atual presidente do chile mal chegou ao poder e já planejou uma constitução desastrosa para a economia do país, que segue os mesmos passos da Argentina. Levou puxão de orelha do excelentíssimo senhor ditador da Venezuela por não ter conseguido aprovar tal excelente constituição, e o Chile já está com previsão de recessão em sua economiza pela primeira em mais de 10 anos, graças aos planos dele anunciados para a economia causando evasão de investidores.
Se aprovam aquela lástima de constituição, a dívida pública desse pais vai estourar pra niveis nunca antes visto, e em poucas decadas, o Chile vai estar pagando uma fortuna em impostos para rolamento de juros de divida, tal como ocorre no BR, onde uns 40% da arrecardação vai para pagar rolamento de juros de divida publica criada por excesso de gastos de governo nos moldes exatos desse tipo de constituição que queriam para lá.
Uma dica: se o seu politico ou partido tem apoio do governo da Venezuela, da Argentina ou Cuba, boa coisa daí não vai sair.
Really enjoying this series man, nice work
An AI should not harm a human, by action or inaction, is probably the best way to answer that final question
Then you have the classic Trolley Problem in its many forms.
I think you made a great point about predicting policing, as opposed to crime. One would only be predicting where most of the arrests/known crimes occur. This doesn't take into account the likely large amount of crimes that go undetected in areas which are probably more likely to have a crime happen due to the lack of policing done there. Very interested
So if the police already over-hassle or more aggressively charge people in neighborhood/demographic A (so they get more arrests/reports/charges per actual crime regardless of the actual crime level), and under-police neighborhood/demographic B (meaning much of the actual crime going on isn't reflected in the data), they'd end up with recommendations to divert resources away FROM B and throw them at A from a predictive system. Any biases in policing or reporting that made their way into the data (even old data, if old data is used) would be magnified the more predictive policing was used (feeding back into the patterns of new data coming in), but they'd be given the false appearance of 'impartiality', even to those carrying it out! ("It's not because I'm discriminating against you, it's because the Crime Index is high here!").
Exactly
@@05Matz mathematician can easily filter the data, eg: separate the report made by civilian vs the report made by patrolling deputy, therefore removing bias created by patrolling strategy. Why would we kept thinking these mathematician were incompetent when creating their algorithm?
@@xponen There is also a bias in civilian reporting:
"the cops won't turn up so why ring them"
"the cops don't like so why call them"
"I don't want to waste police time"
"I'm busy, I will look the other way"
The data is inherently biased in ways that would be hard to predict - I suppose we could do a controlled crime spree and see which of the known crime events ended on the database - I can't see any ethical problems with that.
Love the shorts you've been doing, but videos like this one are why I subscribe.
Best channel in the past decade love your stuff thank you for continuing to make profoundly engaging content!
the problem of math and "predictive" AI to help cops : it is useful to find tendency, but is not able to do prediction.
The AI may predict that "there is a 0.7% probability that there will be a crime in a neighborhood on a certain day". Even if it is 40 time more than the average probability, it is not something very usefull.
The AI will never be precise, and will not be able to say things like "there is 80% chance that there will be a crime here at this hour". (or it would not be a probabilistic predictive ia, but a spy AI, that found that a criminal will act)
( "find tendency" can be very useful, but it need to be well understand, and not over-interpreted.)
Artificial Intelligence is abbreviated as AI, not IA. Once could be a typo, but, since you typed it wrong 4 times, I'm guessing you just didn't know.
@@SgtSupaman (i know, i just used by mistake the french version :
IA is : Intelligence Artificiel
i corrected.)
@@paulamblard3836 the downside of bilinguality
@@U20E0 just the downside of french tbh lmao
@@Goldy01
( this is the second time i post this, the first got deleted because CZcams )
The russian acronym combines the best of both: ИИ ( “II” )
I'm making the mother of all algorithm omelets here Jack, can't fret over every egg.
- Predpol, probably
Love your transitions. So smooth
I got so happy to see a new video from this channel! :)
CUrrent Justice system: Innocent until PROVEN guilty
Predictive policing: Everyone is guilty before being proved innocent
I know which one the Private For-Profit Prison System is gonna back. Hint: It's also the one that's the WORST for society and confidence in the justice system.
Don't tell me you have private in the US.
Even if it was 100% accurate it would be awful. Anyone watched psycho pass? It is literally about this
@@tf5pZ9H5vcAdBp Except that decades of trying this has shown that the perverse incentives makes the entire system far worse. COMPSTAT in NYC and other cities makes cops patrol where cops have patrolled before, because due to soft quotas, arrests and summonses happen in the areas where the cops are, so of course those are going to be the data points entered into the system, and it's where police are going to continue being sent, even if crime moves elsewhere.
Do some research. There are cops suing the NYPD over this right now because they realized they were just visiting these areas where crime wasn't happening to write summonses for people doing nothing wrong, while being told not to enter certain crimes into the database because it might harm property values in an area.
Current justice system: Innocent until evidence hints at possible guilt and no evidence for innocence is found.
@@tf5pZ9H5vcAdBp your right your just a rich elite sock puppet
I miss the days when I first discovered Vsauce2 and I could watch all his videos for the first time. Now I've watched them all, what can I do but wait?
Vsauce videos can be watched multiple times and you still discover something new in them.
Try HAI - Half As Interesting. If you like physics, try Presh or Sabine, for math, Mind Your Decisions
Interesting video! We studied this in one of my graduate classes, and, to oversimplify, it fits along the same lines as 'broken windows' policing. Which is to say, it doesn't really matter how accurate or inaccurate the underlying concept is, the application always leads to worse outcomes in practice.
Dang I thought he was gonna bring out the old 13 50. Or maybe some of those FBI crime statistics.
I was hoping so too
Holy crap, I had to rewind the video during your sponsorship spot... I heard "If you wanna help vsauce2 end yourself" when you said "AND yourself". You had me worried for half a second there Kev.
We literally have a full movie about why "predictive policing" is a terrible idea yet people keep trying to make it happen
Love your new vid Kevin!!! But at first glance it really sounds crazy you know:)
I love how he just jumps right in to WORLD WAR 2 AIRPLANES. LOL.
11:24 it depends heavily on the type of crime, generally speaking you get MUCH better predictions with things like murder than you would something like petty theft
Precisely, because most of murder data is collected whereas for more petty crimes the set may be incomplete, glad someone pointed that out.
Put 75% of police in certain neighborhoods and you will magically discover that 75% of crimes happen in those neighborhoods.
Well no, because crime isn’t evenly distributed. The same Cops can have different arrest rates in different location
I don't think that is true. Maybe for traffic violations, but not shootings, murders, and other similar crimes. People in a suburb aren't suddenly going to be getting arrested for those types of crimes because more cops are there lol.
@@Gandhi_Physique They would for drug use, still an overwhelming source of arrests.
@@nairsheasterling9457 That is like the traffic violation, expert mode. That is likely true though. I mostly meant violent crime wouldn't suddenly be more prevalent just because cops are in the suburb. Won't get too much into drug stuff, but yeah a certain drug not being legal is some bs.
That's one of the big dangers. Since we already have so many police in certain areas, there's going to be that much more data on those areas. Minor crimes might not be noticed at all in better off areas and what about when a body is dumped? How would they know where the crime actually occurred until after it was solved and probably used for further analytics?
“PredPol”
Welp… that’s a sinister name.
Wait a second wasn't this already a vid, maybe I'm thinking of something else, just vaguely remember a math/stats & policing vid somewhere
Predictive policing sounds like thought crimes and 1982.
Don’t you mean 1984….?
Total Recall
@@gorillaguerillaDK: No no, he's got a point, it's only a short time until it becomes a complete copy of 1984 😂
@@fetchstixRHD
🤣
1500 mathematicians really isn't that big of a sample size. I mean just hiring someone for a sales position has thousands of people either looking at it and not getting involved because they choose not to for whatever reason.
mathematicians aren't sales people.
Hell yeah new video !
Used to watch like this so long ago!
Super interesting and important video on how predictions fail due to human error!
Publishing a study that reaches the "wrong conclusion" can kill a researcher's career. Regardless of the scientific validity of the study.
Bingo. This is it. We can't have nice things because the usual suspects are the problem.
Lmao, publically available data shows that the most likely people to commit mass shootings are white, right-wing men. Keep bootlicking till the day posting on your edgy discord gets you swatted.
Always those damn racist statistics!
I used to love the series "Numbers" which was about the cooperation of mathematicians with the FBI. I remember that at some point, the main character has the same idea, wanting to predict crimes before they happen. But I guess it's the wrong approach? I would find it insanely cool to use the help of mathematics to help solve crimes, but there's probably a line to draw at trying to predict them.
I REFUSE to believe that they couldn't think of 300 better inventions than the baguette vending machine, let alone 50.
Everyone loves new Vsauce, it truly does make all our days better
It seems like there's a lot of work going into stopping crime and very little work being done to figure out why the crimes are happening in the first place
👆🎁👆 Thank you for watching and commenting …..you have been selected among the shortlisted winners for the ongoing Ps5/Pc/iPhone 13 giveaway message the name above👆🎁🎁🎁
Because some people are parasites and predators.
reminds me of
" minority Report " the movie
Loved it.
I appreciate your channel so much
Love the direction you are taking this channel towards
Give any random anon a demographic profile of an area and he'll give you a pretty good sense of the crime rate.
Hey dude you're not allowed to notice that certain kinds of people commit more crime.
@@marshallscot You're allowed to notice it. You're allowed to say it out loud; you're just not allowed to claim victimhood when people don't want to associate with someone that takes that data and decides to make an ambiguously racist statement. And nobody's really stopping you from doing that either, it'd just be rather unreasonable of you to expect people to want to hang around someone that has an ambiguously/blatantly worldview.
6:05 I mean, of course its gonna lead to reduced robberies if its telling you to patrol where robberies happen more often... You can come that conclusion without an algorithm that is 100% just going to be used as an excuse to overpolice certain areas and extend their power
But shouldn't the police patrol the more crime ridden areas more often?
Is it “overpolicing” if there is statistically more robberies happening there? Seems only logical to put more police where more crimes happen
@@ponny2948 tbh its better to prevent than to cure. So instead of policing, tackling the root causes
@@Tweogan Overpolocing exists. Look how little police people patrol in Europe
@@ponny2948 it's only logical if you don't understand how crime works, no offense. One, police naturally will find more crimes in places where they are more present. If they don't patrol rich neighborhoods, they won't see the underage girls going in and out of their houses, for example. Also, police presence does not prevent crime. Never has. In many neighborhoods, frequent patrolling only serves to put residents (law-abiding or not) on edge and makes them distrustful of police. It also often is an excuse for police to harass ppl who may not be committing crimes or those who commit minor crimes.
Amazing Kevin, thank you
So, you're telling me... if you know which neighborhoods have high crime, you can predict with high accuracy that there's going to be crime there? This is revolutionary. Groundbreaking. Wow, mind officially blown.
Can you imagine what it would be like if police had to act ethically?
I can: body cams.
@@patrickrunyan4410 Oops I guess I forgot to turn it on or oops I guess it didn't charge properly. Obviously BS excuses but ones that have certainly been used. Body cams certainly help but they can't stop every type type of abuse (especially if the judge for a given case is already predisposed to siding with the police).
@@grn1 Nirvana fallacy.
@@patrickrunyan4410 doesn't apply because you asserted that body cams would make the police have to act ethically not just that they would be an improvement
They will eliminate themselves.
As always, great video
👆🎁👆 Thank you for watching and commenting …..you have been selected among the shortlisted winners for the ongoing Ps5/Pc/iPhone 13 giveaway message the name above..👆🎁🎁🎁
Wise back lighting choice.
people don't realise most of the CRIMES happen SPONTANEOUSLY. especially, CRIMES like Road rages, Petty theft, Physical Violence.
7:49, the quote you are highlighting here isn't from the LAPD, it is from the Palo Alto police, who only used it for three years, just according to what the screenshot says around it
7:45 “But they were also totally unable to evaluate its effectiveness.”
What you highlighted is a comment from a spokesperson for an entirely different police department. Palo Alto is near San Jose, hundreds of miles away.
Either way California is a crapfest that needs to be blocked off from the rest of the country and give them free reign to destroy themselves which they have been doing for past 20 years at least in a slower way.
Loving this arc, acab
I love these videos so much. Micheal better watch out. These vids are on the rise lol
Who's Michael?
@@roberthunter5059 original Vsauce guy, the one with the unibrow and loves to say “nonono”
Where can I find this baguette vending machine?....
Asking for a friend.
using math to bully people sure is a twist
It's kinda like that episode of Avatar where Sokka tries to explain self-fulfilling prophecies to the guy who told him "[Aunt Wu] said I'd meet my true love wearing red slippers."
*Sokka* [slyly]: "Uh-huh. And how many times have you worn those shoes since you got that fortune?"
*Villager:* "Every day."
*Sokka* [angrily]: "THEN _OF COURSE_ IT'S GONNA COME TRUE!"
*Villager* [excitedly]: "Really?! You think so?! I'm so excited!"
👆🎁👆 Thank you for watching and commenting …..you have been selected among the shortlisted winners for the ongoing Ps5/Pc/iPhone 13 giveaway message the name above..👆🎁🎁🎁
The survivorship bias also counts towards complete criminal statistics.
Even if an agency gives absolute complete data on their own findings, that isnt actually proof that it is complete.
To make a reasonable predictive system like this, you can't work with exclusively positive data points, or exclusively negative ones.
But police can not track crimes that almost happened. Or crimes that did happen, that were never reported.
Basically any such system tracks where crimes did happen, and assumes that if there is no crime reported, that no crime happened.
This means, police can never give a complete enough data set for a system like this to actually create any level of useful predictions.
And then there is the magnification bias. Police will inevitably find more crimes where they tend to look.
Creating an algorithm that uses the same data will also reinforce such biases.
This is especially true for crimes that almost everyone does. Like drug possession and consumption. In any group of people you will have a large portion, if not a majority of people who consume and possess drugs on the regular. But if you only check in specific areas, then you create essentially a feedback loop, where you first have a spike in reported crime in an area, which leads to a bias to controll that area, which in turn reinforces the findings of increased criminal activity.
So yea, there is just such an absolute lack of useful data, that a real predictive system would not be possible.
The 50-50-90 rule: anytime you have a 50-50 chance of getting something right, there's a 90% probability you'll get it wrong.
Andy Rooney
🤔😐
The opposite is true as well. If it is not 100%, it is 50%
Since it is still perfectly possible to fail
The best way to prevent crime before it happens is to ensure people have financial security, access to good employment, mental and physical health resources, and access to good education. If we can diminish the core causes of criminal behavior, we can actually prevent many crimes. Of course, that requires fixing system problems instead of keeping our system exactly as-is and looking for bandaids...
what a great video, as always...
3:51 How are we determining whether someone is in a gang? I'm assuming they don't just come out and tell you (and of course most people will lie about this) and it's not like they have a list of every member, much of their connections aren't written down, they're familial, or organic and decentralized. You could also interpret from it that gangs have an outsized negative influence on people that aren't in gangs given that for every 1 gang member, six or seven non-gang members are involved in their crimes.
The police guess based on 'does he look like a gang member?' and manipulate the computer model to say yes.
They very often tell you they're in a gang. They wear gang clothes and flash gang signs.
And there might even be “dormant” members who while not necessary directly related to gang activity (ever is or intend to), are susceptible to quickly involve themselves/or force to be involved in gang activity when certain conditions are satisfied. This dormancy can be symptomatic within weeks or individuals may experience up-to years of non-activity.
To this end, everyone of us, including grade school children, can be considered a dormant gang member.
@@daniellin1726 That's nonsense. The majority of gang members basically boast about it to everyone in sight.
@@daniellin1726 so if we have no way of proving they are in a gang does it then make any sense to trust and use an ai claiming to predict it when we can not verify the data?
maybe having it give its prediction, and compare it with data from normal policing ( not using the systems data ) and analyze its accuracy could be a way to improve it?
but that might introduce more problems as it could reinforce current policing instead of bettering it. maybe math is not the right way to go with policing, at least never beyond adding an extra patrol through a neighbourhood
Or you could fund some actual research - get data on actual rather than reported crime and use that to feed the models rather than using an obvious feedback loop to generate garbage...
Wow! Very interesting!
If your training data is police records, you're just making an ai that mimics the average cop.
I mentioned feedback loops on the last crime prediction video, and I'm glad to here them discussed in this one!
[I'm just gonna delete this paragraph, since it's not very relevant to my comment, and its poor wording lends itself to misinterpretation]
(1) I think it's quite a stretch to say that GPT-3 has an accurate model of the world, and (2) researchers like Suchi Saria are working on understanding feedback loops between models and human actions.
@@AubreyBarnard I'm just using GTP-3 as an example of how fast the world model component of artificial intelligence is progressing. Compared to a human, it isn't that great, but you can still get it to generate paragraphs on a wide range of subjects. It's reasonable to assume that future models will continue to improve, possibly to the point where they can use information about the world to achieve their goals more efficiently.
No. No it won't. The police can hang around my neighborhood all day long and they won't catch any яápes because they don't magically make crime happen. This rhetoric clearly does not hold true for predictive policing of VIOLENT crime.
@@mickeymickey9914 The example I gave is just that: an example. I highly doubt that the specific scenario I mentioned would actually happen. I think real-world examples would probably be a lot more subtle (and, well, realistic).
Also, I can see how my wording might imply that I think police deployment will just "make crime happen," but that's not what I mean. Sure, if the police were investigating YOU, they might not find anything, but if they were investigating people who already had a history of crime, or who are friends with criminals, they probably would. To be fair, this is what the system is SUPPOSE to do: find crime, but how can you tell the difference between normal functioning, and strategic deployment for malicious/misaligned purposes?
We'll just have to hope and pray that AI never becomes intelligent enough to slip under our radar when it comes to these types of things.
As someone who did good grades in college algebra, it is a cursed skill, that comes at the cost of sleeping peaceful hours with nightmares of how to solve the problems.
Turns out training pattern addict monkey brain complex math is like giving a drug addict a comically sized syringe
... this is the weirdest "flex" I've even seen in a CZcams comment section
Pasco County, Florida... They are putting people on a list starting in highschool.
Love things like this where we study the studies 👌🤙
Solidarity with the mathematicians!! ✊🏼
👆🎁👆 Thank you for watching and commenting …..you have been selected among the shortlisted winners for the ongoing Ps5/Pc/iPhone 13 giveaway message the name above👆🎁🎁🎁
Maybe they just don’t snitch 🤷♂️
I literally laughed out loud @ 7:17 thank you 😂
Now I wanna see the baguette vending machine lmao
Anyone seen Psycho-pass?
So, in simple terms:
You have three neighbourhoods with similar populations and real crime statistics, but one has had more police presence, who have recorded more of those crimes, while more have been missed in the other areas.
You tell this to the algorithm, and it says crimes are likely soon in that area. To double the number of cops on the beat in that neighbourhood, you halve the number in the other two.
The handful of cops left in the other two neighbourhoods have been told that crime is unlikely there. The large number in the third neighbourhood have been told to be extra vigilant.
In each neighbourhood, there is a minor crime: a teenager shoplifts a bottle of alcohol. In the first two neighbourhoods, the shopkeepers know that there’s never much police response to a crime like that, and they grumble and write it off without reporting it, leaving it unregistered with the police. In the overpoliced area, a cop sees it happen, arrests and charges the kid, and inputs the data into the algorithm.
The algorithm concludes that crime is indeed more likely in that area and tells the organisation to send even more cops.
And that’s just to begin with. No one wants to move to an area that has crime so high there’s a cop on every corner. Businesses don’t want to open there, in case it’s not safe. A kid in that area ends up with a criminal record that follows them throughout their life, impacting the economic chances of people in the over policed area, which probably does raise crime there relative to the other two. They’re now worse off with more crime directly due to excessive policing.
Haha. Businesses aren't closing because of statistics they're closing because of reality.
@@mickeymickey9914 Right - but if you use statistics to make decisions in the real world, then the quality of those stats is actually going to have a real effect. We live in the era of big data.
@@Youssii Not really, because at the end of the day, they're patrolling. You have to commit a violent crime still. They're not making you commit crimes. Wh_te suburbia doesn't have as many m_rders as Detroit, there's no point in policing them equally.
2:35 You're amazing, Kevin.
This is good, this is very good! The video I mean, not the projects. Geez, that's troubling stuff.
👆🎁👆 Thank you for watching and commenting …..you have been selected among the shortlisted winners for the ongoing Ps5/Pc/iPhone 13 giveaway message the name above👆🎁🎁🎁
"So why did you pull me over Officer?"
"Says here you’re highly likely to commit gang criminality, so I have to detain you!"
"But Officer, I’m on my way to work"
"Won’t do, I have to detain you"
"But I’m a doctor"
"Still, I have to detain you"
"Ben, I’m telling mum and dad that you’re harassing me again"
"But Sis, it's the computer that says it"
Oh, this brings me back to playing 'cops and doctors' with my kid sister. Weird memories.
The new algorithm is just a camera that reads your skin color
That never happened
@@mickeymickey9914
So you don't understand the concept of a joke - okay, now we know,!