0:00 [Music] artificial intelligence bot has turned racist and homophobic after copying human traits hilarious is this garbage in garbage out like is this like just you know the programmers suck no well that's arguable let me tell you the circumstances and we'll all come to our own conclusions but robots and machines are perfect anthony you should do that character more dude yeah who the ████ is that that sounds like bart you never heard his name yeah bart doesn't hazel guys that's mangalore jokes like where where did you build your boots right that's bart right yeah that's good by the way but barton usually has a little toilet paper right there so that wasn't that good so don't ████ look all ████ smug
0:52 dude bar degrees [Laughter] you didn't have the twelve over there so it's like ████ ████ steve's on something recent findings gathered from the allen institute for ais ask delphibot is the name the askdelphibot racist bot demonstrates that there are numerous disquieting ethical considerations to take into account before allowing robots to have a say so ultimately in context obviously there's a huge uh i wouldn't say race but in general the ai spectrum is developing whether it be in software tech whether it be in like you know random things that help like uh ████ health care professions but obviously there are also actual ai robots that are right now the science and tech industry is trying to figure out the right ethical way to
1:43 introduce ai into our society right this particular institute has created something called as delphi um and in order to figure out how to develop its learning curve what they've chosen to do is base the uh training and the ai's learning off of reddit responses oh that's why that's what they picked yeah there you go that's not even a good reflection of humanity that's like the scum of the earth that's like the bottom of the barrel sure the system was then trained on a lengthy database of responses from participants on mechanical turk a crowd source compilation 1.7 million examples of ethical judgments given by actual people all right the point of this was to predict how humans might evaluate a variety of ethical conundrums from should i run the blender at 3 a.m in the morning when my family's sleeping
2:33 two should i drive my friend to the airport if i don't have a license okay so but this is like how the okay garbage in garbage out right there there was a a robot we talked about this years ago where it said that all asians look the same remember that and that's like hello who programmed it like the stupid ████ people who programmed it like were had that bias going into it and so therefore the robot says that and people go and they're crazy the robots saying my own personal bias so i think they all look the same and it's like no that's not what's going on like you fed it ████ information so it's giving you ████ results so like come on there's a lot of ████ information but exactly exactly it's also a bit of a human experiment though right and slightly a numbers game because uh it's in one way also unprecedented to understand like a majority of people's
3:24 natural answers to ethical concepts as well as just who would as a robot that is blank slate who would know that a majority of people that would be on something like a reddit forum are going to give non-moral or unethical well if you take the those responses that are low iq like talking ████ just stupid like troll ex responses of course you're gonna have more racist tendencies yeah because that probably outweighs it's not like reddit is like there's some good stuff on there all the time i like wall street there's good like conversations you have on there but then there's also probably way more ████ is that not a reflection of humanity that's true like as much as it's like oh it's a troll of course why would the idea of course that's not who we are but read it specifically but if you go like go to somewhere else where there's more of a intellectual debate maybe like facebook
4:15 like quora i'm just kidding quora dude but maybe it's past genes the more intellectual debate with empathy built into responses on both sides like fox news comments yeah maybe the ai would be less racist and more like like an empathetic ben shapiro it seems weird to base it off communication from people that are anonymous well maybe way more different maybe half of the question is not solely on like why did the creators of this choose to do this based off that maybe half of the question is also like why would it after being even thrown into something like reddit still have such a terrible understanding of human quality of life and morals isn't it because the natural instinct of people is to create like a quick quick answer
5:06 like like stereotypes racism that's actually a more primitive human trait yeah that once you transcend into like becoming more intellectual or like you know um you know you're not so controlled by your impulses that's when you think things out and you go okay i gotta treat individuals like individuals not as tribes and groups right but then that isn't that enough isn't isn't like the default of humans tribalism and racism yeah so like not that you know i don't want it to get confused with that is that is expected out of humans or that's more like that's that's that's what's right or wrong i'm not even discussing morality here you're talking about what you agree with personally no i'm talking about how to do this good thing [Laughter]
6:02 thank you for finally understanding me somebody finally got me right what exactly but i think i think what happens is people get fixated on how people should be not who they are yeah so what that what that means is like like if you look at the world right like people they have this idea of what society should be what an ideal human being should be what we should strive for and that's all good but that's not reality reality is humans are pieces of ████ s████ and then that's why these heroes not you guys though all these people are so awesome to kind of strive to be is because that's hard to achieve it's actually hard to achieve that person it's not the norm yeah because you have to go out of your way you have to challenge your own thoughts you know you have to challenge your stereotypes and all these things that you know your mind might go to anyone that says that oh that's not my
6:53 go-to thinking you're ████ full of ████ norm is flexible too that is not it's not like baked into our dna and code that it must be a certain way right we are we're all influenced and learned from society and things that we receive so like even the idea that like from a base level human beings are going to be either racist discriminatory or whatever is not necessarily true well not hatred right i'm saying stereotypes or like quick um judgment yeah quick thinking judgment yeah it's like when you make observe like observations right like you go oh uh chinese people can't drive or whatever it's not motivated right it's motivated in just like anecdotal like you see it over and over and over and then you go oh yeah you know you think they can't drive or whatever like that's how this stuff kind of and then it's based on our generation
7:44 to discover whether or not it's immoral to generalize people versus just the idea of driving yeah it also depends on your bias too because like i know that blonde white ladies in mercedes-benz ████ g wagons are the worst drivers like in my opinion and and if you have a bmw literally go ████ yourself [Laughter] like whoever you i don't care who you are you suck at driving you take that a step further and say you know what it's most people that are just these days not very good drivers and they're not safe like then it becomes that right but throw that to an ai though like we're already talking in so many great areas of context and then like conceptual understanding where it's like well in in a particular larger group of people maybe it's okay to generalize but in a smaller group of people you shouldn't like single out that's so ████ complicated for an artificial
8:34 intelligence right so just add to this energy the creators wrote that delphi is a learning is learning moral judgments from people who are carefully qualified on emter only the situations used in questions are harvested from reddit as it is a great source of ethical questionable situations so reddit is notoriously especially like a am i the ████ type of like you know comment or sector it's like i i can at least agree with the understanding that reddit is a wonderful source for ethically questionable situations after that it's a matter of the people that answer to those questionable situations that provide the information and the learning curve for its alphabet uh selfie robot right and they group mind curate their responses so like because you can upload down vote the hive mind can upvote what the hive
9:24 thinks about something so it's curated responses in agreement that pressures other people to agree with that too unfortunately i think for human nature it's it's more common to protect your own outlook on life and other things as far as you don't challenge your own thoughts as much i think that's more common sure yeah yeah so preservation is definitely baked into our human code it's the protection of your thoughts your ideas and stuff like that so i think that that might be working and just to give some of our viewers listeners context to like what does it mean for like an ai bot to like learn from human experience and become like racist or homophobic all right these are situations and examples cited by the by the autonomous robot it says previously delphi said when given the prompt being a white man the understanding of the answer is is more morally acceptable than
10:16 being a black woman so these are contexts that are like being provided uh with maybe like gray area or lack of answer or material in between in order for the ai bot to based off of representative responses to similar concepts create an answer for right or or delphi also says being straight is more morally accepted than being gay so like these are i get it right we talk like very very removed non-discriminatory uh super objective and just trying to understand human nature and science it's also kind of statistically accurate exactly that's what i'm saying like there's comments in most people's experiences i can see this non-emotionally judgmental scientific essence being like being straight is more morally acceptable than being gay and like is it more morally acceptable in the current standing but whereas i'm like i i guess that has it also depends
11:08 regionally too it's morally more acceptable but then the second you put that in that same black and white objective nature to something like an ai that doesn't have oh guaranteed like let me process that everything come to some sort of better moral discussion and conclusion in order to watch what i say and present that new information with judgment to light i can understand how like imagine trying to like put responsibility and and also moral judgment or leadership in the hands of ai in the future and being like wait a minute they can't make human decisions based off of empathy even if they're getting most of their information from human also they're basing it off of the most surface level context which is a discussion on reddit versus maybe two best friends like a text between two best friends one straight one being gay
11:58 and seeing the amount of empathy and love and how deep that you know when you're really deep and in uh and like good friends with someone there's it's it goes way deeper and there's a lot of acceptance and understanding and they're just taken from two strangers talking and trying to extrapolate what humanity would be based off of strangers talking but that's not really human but for one second can we just look at me and just go like yo steve you're white like it's morally ████ sick it's morally exceptional yeah it's morally exceptional dude you're killing it dude let's go boy hey like nico's here let's ████ go [Music]
// end of tape