AIs Wide Open: Making Bots Safer Than Completely Unsafe

all right how's everybody doing everybody making it there we go excellent all right here's your public service announcement the training ground and public ground are at the Platinum next door so please make sure you're going over there to visit that okay again we want to thank you to our sponsors our critical sponsors being critical stack in Vail male and then to our other sponsors Amazon blackberry the National Security Agency silence Microsoft Robin Hood secure code warrior and paranoids okay with that we're gonna get started here real quick you've got the fabulous Davi he's gonna walk through his presentation please be easy on him as you can see he's ready for tomatoes and everything else in his fabulous suit here what all

right Bobby it's all yours take it away thank you all right this is a big presentation I'm gonna cut it down significantly so we'll see how far we get in the 25 or 30 ish minutes that I'm being given thanks for coming today unfortunately were being video recorded so I can only speak about so much my preference is to go really deep into some of this stuff but because we're probably gonna see this later it's good for me I can see how badly I did but I feel in some sense is limiting so the talk today if you've seen the abstract is really about one of my favorite topics which is science fiction and it brings forth concept of Blade

Runner into sort of the modern context of robots and what we're doing about them this is a true story I get asked now not infrequently about how I'm going to help companies stop killer robots they don't necessarily call them killer robots but they say we have robots running amuck in their companies these are big big you know global companies and they're saying can you help us find them can you help us kill them and that's not actually unsurprising when you think about how many people get asked you know can you find stuff inventory and get rid of it but it's just now we've gotten to the place where people are running lots of robot robotic

process automation so I was asked to give this talk and I thought sure I'll give it but let me give you a quick background maybe as to why I was asked to give a talk about this sort of thing in 1994 or five to about 2005 I did a bunch of stuff that was sort of behind closed doors data center type stuff in 2005 I got a little more aggressive and I came out and said hey we really can solve this PCI problem if we do tokenization encryption and then over the years I did a bunch more stuff around that to try to fix this problem that we have with confidence reality released a book in 2012 about

cloud that I'd worked on you know for about 10 years I didn't a lot of cloud stuff one of my favorite things was I started a crypto solution in 2017 after multiple attempts to get it started from 2010 ish finally got some traction with a large database provider and just announced last month that I was able to put field level encryption on the client-side in this database provider which is a huge win for me anyway I'm proud to announce today that I'm gonna make a transition now into the interrupt team Tim berners-lee is trying to reinvent the web and I agreed to go and try to work on the security models for that now one of the reasons we're doing

we're doing this is because the threats to humanity are so great we have to rethink how we date a platform and how we give access especially when machines are getting involved because what if those people who are running those platforms turn into machines running those platforms and they actually out to get us so that's a little bit about me in fact I'm joining interrupt here's what I'm not as an old philosopher I like to say what I am and what I'm not I cannot believe people say things like this in the public media space I don't want to disclose 200 this is maybe you know but they're basically saying hey if you're a cool 17 year old guy and you

want to do some cool stuff then obviously you join white nationalism because that's where all the fun happens and people on the Left have no fun well that to me is just completely insane that a security researcher would be saying this in today's day and age and I am definitely not at all in agreement with that because what he's defining as fun is for white dudes and everyone else doesn't get to have fun and he thinks that's the definition of fun which is tragic and awful and so a lot of that will come out in this talk not only because I just disagree fundamentally as a human but also because my my master's degree really was in what's called

mission 101 which is the liberation of Ethiopia from fascists and I talk a lot about mine it's not a great I don't wanna tell anyone to read it but it's not a great master's thesis but I was young and stupid but I basically talked about how you can take a little bit of information which is what orde Wingate did and he took a few irregular Sudanese soldiers and he defeated a massively larger fascist army basically using propaganda and some guerrilla tactics sort of the birth of Special Forces warfare if you will so that sort of makes me I guess a different animal in this whole world of information security like for sometimes you might see me

doing things like this this is a bit of art that I created to represent what's going on this is very similar to what my master's thesis was about they actually had planes over Ethiopia dropping pamphlets and my master's thesis I have one of the pamphlets that I found in the basement of a British library but basically you have the Russians dropping all kinds of hints and suggestions and propaganda to try to change people's opinions and you see tweets like this that say hey what's going on well robotic automation really amplifies this and so that's sort of the thesis of the talk today but before I get there let's talk about how great a AI is going to be because it's not rain

entirely on this parade it's supposed to do lots of cool things for us help us drive cars translate language all the things you don't want to do you can expect robots to come and solve for us and this is not far off because silicon brains are here basically if you've been watching the neuromorphic efficiencies I don't know if this is familiar to you or not but things are hard and slow so if you put like a McAfee endpoint you know analysis engine black blacklist engine it can go really really slow trying to figure out all the attacks but if you move over into the new and more the neuromorphic side suddenly the efficiencies pop up because they have

better decision tree gating and so there's representation right now of these chips being built Intel's behind it to give us real silicon brains and one of the funny things I do is I sit in some ethics committees and salons and stuff and I go meet with people and the big big firms about what they're doing about these sorts of transitions in tech and one of the phrases that came out of one of those salons was the silicon brains really are technology and social processes to support value-based system innovation which is I think a great way to describe what we're trying to achieve makes no sense to me but that's humans trying to describe what's coming so

value-based decisions is like you know riderless bicycles and this is real I think Google made fun of this a few years ago but it's actually now happened neuromorphic chips are able to make enough fast decisions that they can balance a bicycle and you know I asked here well they turn on their creators that's kind of a joke because the bicycle has to turn and it's cognitive science if you will but the the point is I don't think a lot of people saw this as our science fiction future they were thinking of like flying cars not you know critical mass if you're familiar is a bunch of bicyclists trying to stop cars imagine if you have robotic

bicycles at your disposal to send hundreds of thousands of bicycles on their own into the streets to stop the cars from moving around can really have an impact or in real terms if you don't want to park your bicycle it just rides around by itself until you're ready to go and comes back for you so the spectrum o of expectations is more like this and we've seen in science fiction a lot of stuff like Douglas Adams he says 42 you just get meaningless information maybe an iRobot you sort of get benevolent robots you might get in Star Trek iRobot that's kind of incompetent can't laugh can't do certain things to where you get Red Dwarf

how many people heard of Red Dwarf this is a fantastic robot that's constantly making life miserable but just for fun for the occupants there's how of course everyone knows kills everybody and then there's Blade Runner everybody know Roy Roy is sort of the Arians a stereotype of like the perfect human who wants to live forever you might have read about somebody named Epstein in the news who's very similar to this character who thinks he can like you know preserve himself forever because he's a superior human and he'd say abuses everyone else for his own benefits and so there's sort of foreshadowings Blade Runner really was 1968 so it actually for me represents a time period in automation

soon after the Cuban Missile Crisis if you will where people were really worried about the future based on automation but who really is missing from the science fiction story has been coming up lately a lot in my mind any guesses who might have been left out here who might be the most important robot of all time anybody shout out nothing all right so first let me a little hint quick poll how many people say that unregulated AI would lead to human extinction not many people okay what percent says harm to society will come from the evolution of machine learning to rapid not many people not that pessimistic all right well the polls done by the

official pollsters say about 12% think they're all going to die and 34% say there's going to be a lot of harm and really who I think is being left out as Mary Shelley who in 1816 wrote Frankenstein in 1818 it got published and stuff but this is a teenager woman who's saying basically the future of technology is really scary because these robots are gonna kill us all and she really invented both sci-fi and these robots of doom I can't say robots of dystopia because that word didn't even exist until John Stuart Mill 50 years later coined it which now we talk about dystopia all the time but what she's asking is how do we regulate the

destructive nature of this technology that's coming because Frankenstein was basically the modern science she actually borrowed the idea from some research in Italy about frogs ever being electric tried to like reanimate them and so it's interesting to see how this bleeds into lots of topics like anti-vaxxers and stuff are afraid of Technology so how do we both be afraid of technology but rationally afraid of technology in order to improve it and I think it comes out of philosophy that we have this idea of trusts which is what I think security is based on and so when you get all this information and you're a robot are you really making educated decisions that are trustworthy and at in the

Frankenstein model they called it the criminal brain and there's actually a joke in the story by Frankenstein about that and let me make a finer point on that what's actually happening isn't necessarily that you have this overarching superpower because you're a robot there's a sort of superpower robot model Roy for example is one of those superhero super power type robots but what if you just took a bunch of like sort of mediocre technologies and combined them together so in this case let's say you take barbed wire and you take some repeating rifles and you take some piston engines and you put those three things together each one of them by themselves isn't like the most

amazing robotic innovation that automation helped a little bit but when you combine them any guess is what you get what does humanity experience as a result of little improvements in automation being put together you get concentration camps and so what I'm really looking for is the change in technology that when combined really has a huge domain shift and I think we're getting there with multiple types of technology that by themselves don't seem super scary like you guys didn't seem to think that humanity will be exterminated but what have multiple little things that aren't that scary get put together in the right hands the criminal brain if you will and then you have sort of the

the Challenger launch situation where people are like should we launch and they're like well we think there's risk here but we can't express it in terms of what to do so we might as well launch until we and then they figured out later that what they should have done is not launch and they should have reevaluated the o-rings and cold temperatures so there's basically two phases to these presentations I give and this book I've been working on for seven years now after my first book to try to answer the problems of securing large data platforms and making killer robots less likely the first is what I call the ICEA which is that critical phase of figuring

out what's going on science of evaluating the risks and then the second is coming up with really easy judgement calls so people can take a pill for example like if you had a Frankenstein monster that was angry could you give it a pill and make it happy or make it like regulated in a way right we have that mindset in medicine already so I actually borrowed this from the rinderpest epidemic this is how they solved it in in Africa was they basically used a Nicaea and an erm model and it shouldn't be unfamiliar because if you think about the ISO standard it's like what should we do and you think about the NIST standard it's like this is how you do it

so this is common a lot of scientific models so let's talk briefly because I'm not gonna have much time about moving through these phases basically robots are our shift in power and it's the same thing as Beowulf if you remember when it started was like you can take 30 machines and it can give you as much power as a supercomputer and so that's really promising so what do you do if you want to restrict power they're like giving people that much supercomputer power all of a sudden makes them much more powerful than if they were just by themselves and the same way concentration camp works you get a repeating rifle at some barbed wire one

person now keeps thousands of people at bay so I've actually built over time and I'll explain later today I have another talk and depth what these different boundaries are but if there's a criminal brain basically behind the technology Facebook if you will how do you build a boundary for them to sort of back off where you say you've overstepped the line and this is where you need to go back you have too much knowledge and I don't have enough privacy and where does that regulatory model work and so that gets complicated quickly and this is where it really gets interesting for me where you start gamifying all these boundaries I mean first you say hey I

want to create an easy and yeah I'm trying to get to erm where I have this easy pill for you to take but then people say yeah but what if I do this other thing and my game actually you know outsmarts your attempt to control me and this is very true this is what I do essentially all the time with machines now with robots is you know if they have a mug you break the mug and you say do you still see the mug because I've decomposed into these parts and that's a real test and one of those interesting tests comes from sports games if you will where people all the time are adversarial in NASCAR cheating

is extremely common in fact champions talk about how they won because they were the best cheaters very different than bicycling where Lance Armstrong got in a lot of trouble he really should have been a racecar driver it would've been fine but basically when you look at robots cheating you have awesome examples like there's a pancake robot that was measured on whether the pancake hit the floor so it just threw it as high as it possibly could so it would never come down and so it had this sort of infinite height that it would attempt to get in every pancake so you couldn't eat either sort of the downside driving robots were able to figure out that

they're only being measured on bumper impacts so they just never hit the bumpers they would slam into each other the side panels because it wasn't measured or one of the more interesting recent examples was in tic-tac-toe they allowed an infinite board because they thought that was cool for computers to play the infinite space but one of the robots figured out that the others didn't have enough memory and it would put a piece so far out that they would crash and then they'd forfeit the game and this is not unlike what the Russians were doing where they figured out the hole in the wall where they can actually undo the entire anti-doping laboratory and Russia I don't know how many people

know this actually bombards athletes with sort of propaganda information to get them to just forfeit the games very much like the tic-tac-toe episode so if you're an athlete going into a competition and someone can manipulate your ad feeds or in manipulate your information feeds they might just send you a lot of really depressing stuff that makes you sad and you don't want to compete so people start giving up and that's very similar to what we saw with Cambridge analytic aware they just figure out that they could target persuadable people who are trying to do something and they would try to get them to do something else and this actually for me comes from a very seminal report

that came out of Bismarck in La Salle's documentation I don't know if you know who the Salle's they may heard of LaSalle we heard of Karl Marx all right well if you've heard of Karl Marx you should have heard of LaSalle it's a tragedy you haven't so I'll leave it at that but basically LaSalle had some interesting ideas to Bismarck about what socialist future would look like and how persuasion was very much a part of the electoral and voting process and just a quick tangent because again I'm very limited on time the the idea that you can persuade people with certain signaling back to my master's thesis and the whole idea of changing outcomes is

sort of sensitive right now because right now you see in the messaging arena that people are talking about infested this word infested and this is a classic in military study this is a classic term that's used to dehumanize people by comparing them to nonhumans and we did this to the Japanese in particular in World War two we didn't do it to the Germans in World War two but we did it to the Japanese and that actually was a way that we were able to push them into concentration camps if you will and in Rwanda for example there was a lot of this that led to genocide and I just want to point out I bring that up

because I want to point out in the InfoSec community I've seen the same thing and this is very distressing to me that people who are supposed to be security experts who are supposed to be making a safer future for all of us and stopping robots from doing bad things are engaging in very dangerous insect-like communication there's forever the roach and they literally said I'm not making this up they said in the communication that Trevor the roaches family is caught up in the disaster in Puerto Rico so comparing brown-skinned people to roaches and then they when people criticize them they started to say are you hating on children and widows and it's and they're spreading in multiple mediums so they

weren't just doing this like in an isolated group they're actually trying to get it on TV they're trying to put on CNN they're trying to put it in the newspapers so we have a very dangerous future ahead when people inside the security community as I said in the beginning and I show again here are doing things that ultimately if they get control of these robots can do mass harm to humanity so bias like this combined with automation of Frankenstein's combined with the fact that you can game rules means we're in a deep dark state right now because how do you create a rule or control a firewall if you will when you have people trying to get around them

all the time and they're doing it for biased reasons so I would say based on that that AI is the civil rights battle of our day and I haven't made this up myself I see this a lot in a lot of forums particularly on the ethics committees and stuff but the dilemma that comes forward for robots is if you see somebody like a human are they an insect that you would squash or are they a human they have to respect the rights of and that's why when we get into harms it gets really interesting to talk about robots already killing people of course from my perspective if you're given the ability to see harms I'm gonna see harms

everywhere for me that's like saying a matter of ulnar ability tester well I found a bunch of vulnerabilities and they say why do you always see vulnerabilities I'm like that's what I'm trained to do that's why you hired me it's touchy because when I say I see racism everywhere people say do you have to keep talking about racism and for me that's like do you have to keep talking about CVEs yeah that's kind of what the whole point is when I talk about biases so robotic deaths are actually on the rise significantly and here I've captured a bunch of them where you can see that there's a couple 79-81 2007-2009 and then there's just a huge

spate of them from 2015 on they start happening more and more I want interesting things as the people who are manufacturing cars were killing people making the cars and now they want us to trust the robots that are driving on the roads now that they've been killing people who've been making the cars so anybody who says the victims are to blame any of these robotic cases you might be familiar with anyone all right good well here's just one example and like I said going a great depth later this afternoon in another talk but I know how many people know in 2007 there was a gun that was designed to shoot automatically it was a robotic gun and instead it turned

and killed all the people who were running it so 9 South African soldiers were murdered by the gun that were supposed to be firing on a computerized basis Tesla I gave a talk about right here in b-sides where I tried to make the case that the card chose to kill the person in the car and I would hold Tesla accountable for that because it decided that I proved in adversarial testing why that happens and a 90 percent failure rate I could throw different things at these cars and I can make them fail 2017 we saw a similar problem where the car was easily confused by signs on the road and then again in an adversarial test we

see 90% Plus failure rates for these cars here's another example in just another just I'm not talking just about cars I want to show you like when you get something like antivirus involved then it's easily bypassed as well you can throw images at it and it's very it's not hard at all to basically break the AI and make the robots do what you want this is one of the more tragic examples people haven't talked about much but the auto pilot in 2018 was actually following another car and then that other car took off and this car slammed into a barrier and just destroyed the front of the Tesla that is an active adversarial attempt that you

can repeat very easily you can get the Tesla's to follow you and make them crash into things and so I have another adversarial test which we haven't been able to share the real it'll make the Tesla do what we want and basically make it crash where we wanted and how we wanted 2018 uber example of course there's a pedestrian death this is so tragic because you know almost you know over seventy 2 / 7 T percent of crashes happen at night with pedestrians and yet they didn't design at all for that that is just a negligent engineering in my opinion and so another example of how you could test for this is in Ukraine there's artists who actually popped a

pedestrian out of the street of a manhole cover and if the car didn't stop for the balloon that popped out of the street you could see that it wasn't recognizing humans at night that's literally the test they created that was a long time ago I have some footage of it 2019 one of the more tragic cases that have come up recently I don't know how many people know this but the robot airplane was fighting with the pilots 26 times the pilots tried to take control back from the plane and the plane was like no I'm going to crash into the ground and they were like no I'm the pilot no I'm gonna crash into the ground they

lost and people died right hundreds of 300 plus people died from the Boeing crash so all this basically comes down to the summary that Microsoft put into its 10-q I don't know if you've seen this but their intelligence their AI is if you will may be very harmful to their performance because it may have an impact on human rights it may impact privacy it may impact employment it presents all sorts of ethical issues so basically beware the clip enstein what they're creating now is the Frankenstein the new Clippy and it may try to kill you not just spy on you so that's it I'm gonna run out of time obviously so I'm gonna take you a little bit through the

constructive phase because I think that what everyone's wondering is like oh my god if these things are so easily broken and I mean easily broken I mean like we sat in a Tesla stupid I admit and then we made it try to run off the road with us inside and it worked and so we thought well that's great we almost killed ourselves but to make a finer point you don't know if the Tesla's on the road today are actually connected to Tesla or they're connected to a rogue server and you don't know who's running that rogue server I'm not joking like that's legitimately you drive next to a Tesla you don't know if it's controlled by someone who has

adversarial intent and is basically putting them all on the road so the point I sometimes make is when I used to study like ICBMs flying in from North Korea into the safe San Francisco you would think about all sorts of countermeasures even people talk about day taunts and mutually assured destruction and things like that well what if they just hacked into the algorithm that controlled all the uber is or the Tesla's on the road and move them 15 feet over and ran everybody over on the sidewalks mass death mass cast Rafi the catastrophe shutting down all the streets in a major city all from an algorithm change or from pollution of the algorithms using the cars that are

just running around in and these are real possibilities given what we've seen already so a constructive phase let's talk about integrity we can do confidentiality that's what I spent the last you know 10 plus years last two years I built client-side field bubble encryption into a database so I think we can solve confidentiality availability solved I can keep databases up pretty much anywhere anytime forever but integrity hard to solve but we can do it and I think what we're gonna do is build moral frameworks instead of trying to use these games which are based on legal frameworks I feel like you're looking in the past moral frameworks are looking into the future security people are good

at that what is the intent of the regulation as opposed to how specific written so we can catch people trying to get out of the game into the big nets and then using analytic intended integrity engines which I think come from philosophy philosophical backgrounds in other words we get people to agree that history for example is a good way to decide what is truth and what is fiction we shouldn't get too wrapped up around assassins which is the Bladerunner model although it is attractive to say there's a hero in the room that's kind of what CISOs do they say I took out an XP box today I found on the network and I eliminated it that

is a model but problem with that model is Blade Runner is too late and you really want to say hey I've killed enough XP already I'm retired I'd like to do more interesting security stuff and we've actually seen in history that this is true by the time we have to go into Nazi Germany we're actually building huge integrity machines I don't know how many people know but there was a giant integrity machine that the United States built using black women soldiers and was so efficient it cleaned through all the backlog of messages in the European theater six months ahead of time anybody heard of this story they weren't even recognized until the 2000s for the work they did in World War

two they're recognized marginally but lately they've been getting some recognition so we have examples of mass integrity engines that have worked in hostile theaters of warfare and it wasn't based on a Blade Runner single person assassin taking out this whole person it's really based on moral compass and that's where we're basically trying to say that we have to change how we educate people and I believe Hume gave us one of the best examples of this in the 1740s but I'd actually say it even better was the mother of Mary Shelley Wollstonecraft created basically two treatises that said is the system of Education that we have to fix and so if women aren't treated as humans there's

your problem she wrote about this in the 1790s and I think if we apply the same logic now to integrity models that machines are going to be based upon what we get is machines that won't try to kill us because they recognize the value of life or they recognize the value of a human and so they'll start to change their behavior so in other words what robots are really doing is modeling out what we're telling them but we don't admit to ourselves what we've told them and so if you put robots in a situation where they follow what they think is true and they don't have the analytic reasoning of a historian for example they're not able to get out of the box

and this is just a perfect example of that somebody says you know here's an amazing picture of a Spitfire flipping a v1 and then here's another person who says much earlier here's an actual picture of a Spitfire tipping over a v1 which had to be done so it would had to be done because it was too dangerous to shoot at it neither of these are true and now a historian can tell you neither of these are true so if you ask me about deep fakes or fakes or other things I will tell you have you been to a history class have you worked with a historian do you know any historians because they will tell you how to see what's truth

and not just in terms of a rational doctrine but in a moral doctrine as Hume was explaining they will be able to tell you you know what should be true what's more likely to be true because sometimes perspectives are skewed right and so just because it's true in the sense of somebody saw it doesn't mean it's realistic so I believe we have solutions for this and there's quickly get to what I think is the real punchline because I'm at a time is things like jaywalking our proof that we have racist doctrines within the United States that criminalize certain humans jaywalking is a way to criminalize people who walk which typically are non-whites and so you can

either easily see how you can then basically put people in jail anytime you want because they're moving around the streets without a car it also creates a privileged class who can afford cars and so forth now there's a long backstory to this and it's been well documented but the point is if you put a robot in this situation it will literally decide to kill people in the street because they're not humans they're jaywalkers if you will and may even treat them as worse so if you re humanize pedestrians then robots are going to behave more normally but once they behave like we behave we just don't admit it to ourselves that jaywalking is inherently racist and designed as a way to transfer

power to whites then you know if we can't admit that then we really aren't going to be able to solve the problem with robots and no matter how many blade runners we have chasing around all these cars killing people on the streets will never catch up so thank you that's my presentation [Applause]

[Music] great question so what are the if you will the erm s that we can do and so I'll talk at length about this later today the two things are first you have to have that framework that accepts that you can have morality as your guide and people do this already they build ethical teams whether they have ethicists or just people who are concerned and they create the argument that they will decide what's moral and immoral that's like people who decide whether something is too insecure a risk register in ISO terms so where are we running to hot because it's immoral the second thing you do is you build an adversarial test scenario where you

actually can go and this is where historians come in they're trained at testing things in a way to see if it's false or true they push and pry at it and so some of the latest research shows for example if you have results that come out reliably and you say this obviously isn't biased you flip some of the decisions so you say what if I put black people in do I get the same result what if I put women in do I get the same result and once you start testing it's called auditable AI rml once you start testing for specific non positive outcomes or bad outcomes you start to see results so you have a group that can

address the issues when they're found using a moral framework not a regulatory one and then you have a series of tests that you start to prove that they're vulnerable and then you take that book and I've been on calls recently where people say can you give me that software and I was like well I guess someone's gonna have to start a company that does this automation of tests but right now you just get people who understand how to do analytic tests to prove something false true yeah any other questions I mean it's fun to break this stuff but it's also scary because you find racism everywhere and I put in those terms because if I find some financial fraud

people are like oh how much money will I lose but if I say racism and it's real racism then it's like it's a humanity issue right or sexism or bigotry it's a you know the CBS s score be at ten we don't really rank racism that way but that's what I'm talking about one more question reimagining of mother

what are everything well yeah I think sort of underlying this whole concept is that we are threats to ourselves and machines accelerate that and so it's already in the presentation I could spell it out more clearly a good example of this I mean to Amazon's benefit although I give them a rough time all all the time Bloomberg did a story where they showed that the American experience was to segregate cities into white black and that was part of the history of the country and then when they did prime Amazon just inherently delivered prime to whites not blacks and Amazon's response to that was because they weren't delivering prime to any black neighborhoods was to treat all

zip codes as equal that's a solution that's literally like a all zip codes are equal and they could've said all humans are equal but effectively they found that there was a flaw and they addressed the flaw by treating everyone as equals so I think those are good solutions that Amazon's they do horrible things on the street though they're Amazon trucks and vans are just disasters of humanity but and they do a lot of bad stuff but they did a good thing there and that's an example of how to fix it so it's already been addressed and it's because again humans had historically done bad things urban renewal was inherently racist and it led to some of these things that

Amazon just picked up without realizing it and the machines then played out the racism of our desegregation and segregation policies yeah thank you [Applause]