[Release Date: June 21, 2022] It’s hard to imagine games without AI, but having AI in games also raises a host of ethical questions involving their use by players and game companies. We explore the possible use of AI for cheating and manipulation, the importance of transparency, and responsibilities of AI’s programmers.
SHOW TRANSCRIPT
hello everybody we’re here today with dr jillian smith who directs the interactive media and the game development program at worcester
polytechnic institute where she’s an associate professor of computer science dr smith’s research interests are in
computational creativity game design computer science education and the intersection of traditional crafts and
computation her interdisciplinary work merges technical research and ai and hci
uh with creative practices and textiles and games with with a view towards addressing social issues and broadening
participation and perspectives on computing she earned her phd in computer science from uc santa cruz where she
studied in the center for games and playable media uh gillian welcome to the show
thank you i miss andy this is another show our second episode without andy andy’s
usually my the cheerleader at the end of welcomes our guest to the show once again andy is helping with his mother
but he’ll be back with us on our next episode um all right jillian we are here to talk uh
ai so um there was a whole lot when we thought about uh
what this ai ethics uh episode should be it was a little all over the place because there are so many directions we
can go to with ai and in fact this may just end up being our first episode and
and an ai ethics one of the ways that this actually came out and me and andy kind of came at it
from different directions but uh my uh my brother-in-law is a poker player he’s
a professional gambler and he was talking about how uh ais uh are being used to to cheat uh in
online poker and how it’s like a really really big deal in online poker so i kind of wanted to start with a cheating
idea if you have any idea about this how are ais being used now by players to cheat
against other players or um in any kind of online games oh that’s a
it’s it’s tricky right and some of it comes down to what’s the definition of cheating right and where does
like where does tool assist end and cheating begin um in terms of like what a level playing
field is so i i will say i’m not super up to date on on like the latest online
games and exactly what’s happening in the news around ai and cheating but for sure there’s there’s ai players in in
multiplayer games i’m absolutely certain that some people are doing tool assist play or trying to do tool assist play
um so yeah it’s it’s there so let’s break it down into two things
right ai players as essentially i’m playing someone that i think might
be a a person but it really actually is an ai and it’s controlled by another
player is that the idea uh so it’s controlled by so so definitely
online games will have ai controlled players
that may not immediately be obvious to the player that they are playing against an ai it presents to the player as
though they’re just playing another character in the game right so notice right um some of that
obviously is just uh you know the game is designed to make you you know to simulate playing other
players so you can enjoy a multiplayer experience uh at what point so at what point can
this be let’s say morally problematic yeah i guess to to answer that i would need to know like why
why would people want or not want to be playing against ai what is it doing to their overall play experience in terms
of like their stature in the game right so um i know that some games will use ai
players for um you know you want to be able to pick up and play right away and that aren’t players who are available to
play against so here’s something eyes to play against right right but i don’t know for sure whether that’s going into
or or how that’s being treated in in sort of game rankings right i i would assume that it’s not having an enormous
effect on sort of like player ranking and in the game right if you’re if you’re playing mostly against ais but i i don’t
know it’s it’s not very transparent right so which i guess is part of the ethical issue right like
does the player have the right to know right you know as there are some games where
you know i really want to play multiplayer and i’m tired of playing with the with the ais i think i’m
thinking of battle polytopia which i play on my phone constantly but it just seems impossible to ever play
with a whole set of players because i end up waiting forever yep i wouldn’t mind a mixture of player and ai but uh
they don’t do that but if they did i would i really would want to know who the ai is so why would you want to know
who they are because i played the ai all the time you know and if i’m going to play the
players i want to feel like you know i know that i’m actually playing the players because i already have like a
single player option where i play a bunch of ais right uh that’s this is a turn-based
strategy game right so you know ai’s are are really easy but i was thinking actually of other situations so the
example uh my uh my brother-in-law uh uh essentially told me uh was um that in
poker if let’s say i’m playing poker uh if some of those people around the tables are sharing the deck
uh if i had confederates among those people we could essentially pass information to each other right and if
those people are ais right then they’re by confederates then i can get information we can share information
about let’s say uh what cards we have and that gives us an advantage over other players
uh in in the game right i’d imagine there’s other uh i’d imagine there’s other games uh
where this is the case where essentially if uh you’re supposed to get a bunch of random
players but really some of them are ai controlled by a player oh i see right so that’s a different thing
than if they’re ai controlled from the company right which i asked first which is right i understand the
confusion has another set of things um i would say i don’t totally know if that’s happening or to what extent it’s
happening um i’m sure companies have a lot of rules against it um i’m sure it’s happening in poker
i’m sure it’s happening in in non-competitive esports there’s a lot of systems that are set up to make sure
that that can’t happen in like competitive play esports um right right like the team has to all
be in the same place it has to be known who’s playing what right like there’s there’s league rules in place to
prevent that from happening but the sort of unmoderated i don’t i’m sure it’s happening somewhere right yeah it’s
interesting that what kind of uh things game companies really can do to stuff like uh to stop stuff like that from
happening so okay so yeah being in person is an obvious one yeah um yeah
how do you how do you check if uh how would a company check if there are
ais being used by players to gain advantage they go beyond let’s say the aimbot right so right
be the tool assistant thing that you were talking about yeah that would be tool assist right so and and that’s that that sort of dividing line of like when
does it stop being a tool right you know what skill do you want to be measuring right is is part of being able to play
and being able to play competitively having like your own really nicely written set of tools to be able to play
um right which which is is different um i mean there’s some things you can look
look at for for detecting ai play right like hyper repetitive behavior spamming
spamming yeah so so thing things happening in the game that are happening faster than a human would
typically be able to do it right so um you know an ai can take actions
relatively like faster than a human can right so right so can you like try to detect that that’s what’s happening and it’s it’s
similar in some ways to trying to detect bots on on twitter and social media right like you’re looking for
non-human-like behavior right right uh i think uh i’m thinking about elon musk
talking about the butts on twitter and how there’s too many posts that came out way too quickly and couldn’t have been
done by humans because they were posted they must have been algorithmic because of the speed that they were
posted in which is interesting because it seems like a kind of easy fix if you want to make your ai seem more human
just give it a like yeah in in one of my classes at wpi i
actually teach uh ai for games and one of the assignments that the students have to do is
make a like a touring test mario so write an ai control of a mario that when someone
watches it play people will think is human um and then we run a little competition
of like who has the most human-like mario ai controller uh that they’ve written oh my god
and i’m just thinking you’re helping people you know that that entire uh you
know prove you’re human i think you’re you’re helping people it sounds like you’re helping people kind of beat that
in in some way in some way or i’m teaching them about human behavior
which is a lot of what ai people do right um but yeah some of the students in that class they figure out very
quickly that one of the best ways for them to make their bot seem human is to insert mistakes
to slow it down is to put in hesitancy right and a thing that
sort of takes away from the it looking like mario is competent right right because
people will assume that incompetence is a human trait well i mean in my case it is and in my
case it is as well yeah so okay so let’s let’s say the ais are
not trying to uh appear like people right um uh and let’s say they’re uh easily
accessible they’re smarter too what are some ways that you see uh uh players potentially using ai uh to
cheat uh either now or in the future and so so certainly any kind of tool-assisted
play um speed running is actually an interesting place to look for this right so
um where where tools is to play of like you know there’ll be there’ll be some runs
that that you’re allowed to to have tool assistance on that have to be just human play
um wait that you are allowed to have tools you are allowed to have tool assistant so if you look at um oh my god
there’s a speed running festival that runs every year um game’s done quick i think it’s called it’s like a marathon
of of speed runners and okay and people will there’s tracks for it where you’re
allowed to have ai assistance in your speed run or you’re allowed to have a tool assist in
your speed run to like put in uh do the do the key presses for playing the game faster than a human would ever
be able to that sounds like a potential spectrum so i’m speed running this game um on one uh on one end of the spectrum
i’m just doing it all alone on the other end of the spectrum it’s completely and ai is doing this for me yeah uh right so
what’s what’s in the middle here what kind of tools i i i’m even having trouble imagining the kind of tools
that’s why i’m asking oh so it might be like you
putting in the series of commands to be played and then the tool
you know spamming them fast enough to be able to actually get through the game is one example right so you tell you
kind of tell it what to do and then it it does it um or maybe it like you reach certain points in the
game and you’ve got a tool that will kind of like get you past the comeback
quickly or get you past some aspect of the game quickly so that then the humor will go on and do more of it
ah okay oh i see so interesting so specifically like there’s a point
where i would need to slow down and i could have an ai tool uh essentially help me pass that
specific point like a battle potential yeah so in competitions they
might agree that hey let’s just skip all the uh the battle part and we’ll have the humans do the rest well
and not necessarily that it’s more about like like where where is the skip right like if if we know that there are
tools that are out there and right and like using the tools can be a skill in itself right
sure and this is where this is where it kind of comes back to what’s what’s the definition of cheating right okay is it
cheating if everyone has access to the tool or is it cheating if we’re open about what the rules are for when you’re
allowed to use it or not right yeah interesting right so right so we can say look here’s a our competition allows you
to use any of these ai tools as long as you’re sticking to the assign tools uh you know they’re just
fine you can use them if you end up sneaking in any of your own additional tools uh that’s a no no a lot
of the time these tools are actually made by people in the community so maybe it’s okay if you’ve modified yet so you
know a lot of times when we talk about cheating it says and you’re getting an unfair advantage and of course the big
question is well what does fair mean how does fair go in here and you might say look it’s as long as i did it it’s fair
and as long as i programmed my own ai uh then that’s fair but as long as i
tell you about the fact that i’ve done it or as long as i tell you uh the fact that i’ve done it yeah that that itself
is interesting right so you can get an advantage but um you know uh just as if um let’s
say we’re doing a uh a car race uh a traditional car race uh in the real world and you’re like yeah my engineers
came up with this great uh you know with this great uh thing for my carburetor uh i’m not gonna pretend what
a carburetor does i just gotta throw that in there you know but that allows me to shave a
few seconds off of off the lap uh i might say yeah uh you know our team came
up with this new ai right and then you’re using ai you’re being transparent about it so you’re not necessarily
cheating but if there are potential rules about what kind of ais you might use
um and this is violating those rules then maybe maybe you would be cheating
so where it gets really interesting is in things like ai competitions for
for players so so every year there’s a there’s actually a couple of different starcraft ai competitions
and the idea is that you want uh the researchers who are entering the
starcraft ai competition are trying to create an ai that can most competently play
starcraft against other ais okay um
and they had to set rules in the competition of what’s allowed and not allowed in the
ai play of like what’s the what’s the level playing ground well because
it’s it’s fair rebecca what what’s fairness right like
what information is the ai allowed to have access to how often is it allowed to make commands right is it allowed to
try to knock another ai out of the competition through non
non-ludic play right um like it has it has to you gotta give me
an example of that yeah i gotta get an example that just sounds fun yeah uh so in the early days of ai play
testing i remember this is like over a decade ago um i went to this conference and there was a paper that people were
presenting about their ai their ai play testing tool um and they were trying to you know have
have ai characters play the game to be able to identify where the bugs are in the game
and one of the the learning agents learned that
it could beat the game by forcing the game to crash under certain conditions
that would count as a win against the against the player
and so it learned how to like engage in activity that is outside of what you would typically do
as a player in the game in order to force the test to
crash right like a ddos attack yeah and essentially it like found a bug in the in the game
server code by like forcing too many actions all at once
wow wow that’s cool so yeah if the goal is that you’re writing an ai that can
participate in competitive starcraft play or competitive anything play it stands to reason that you would want the
ai to only be taking actions that count as playing the game yeah
interesting i haven’t really thought a whole lot about the idea of cheating in ais playing each other i won an episode
just on that at some point and we might might get back to you on uh on that jillian
let’s go back to an an mmo right you got we got an mmo with both players and npcs
um can you think of any potential reasons we might want to know uh we want to make sure that the players know who’s
a real person and who’s an ai yes real real real people have real feelings
right uh and so you you want to know if you’re talking to a real life human being on the other end of things for
like mitigating abuse reasons right um people will people will talk to or
interact with something that they know or suspect to be an ai differently from a human
or not right like maybe they treat them the same way and that’s horrifying but but for the most part yeah you’ll
you’ll just you’ll interact differently with with someone who you know to be human versus and yeah i mean that seems
like a very good thing to yeah to keep in mind right uh and presumably
i would assume it’s interesting because part of me is really uh interested in the possibility of
really not knowing uh being in a world where it’s really not clear who the player control characters are and who
the npcs are but if i get a level interaction with them that could have some sort of emotional repercussions
like this it makes it makes sense that sounds stressful to me
to not know and like weirdly a little dystopian right like how how do you know if the person you’re talking to as a
person i think it depends too if you’re talking to them right as opposed to going around just shooting people right then in that
case it really doesn’t matter yeah maybe but the more you’re able to have interactions the more you’re
actually able to which is interesting this partly this means uh how much are we gonna enable the ai to be able to to
talk with you like a person would and how much we’re even keeping the ability of players of talking to each other
directly versus uh you know uh choose responses from a menu or uh right
um because we’re already so worried about players uh being brutal to each other
and being toxic to each other in the first place yeah it’s um yeah i guess it’s like the problem matters
like predicated on the extent to which you’re you’re expected to behave as though you have a relationship
with the entity right so yeah right i mean to the extent that
shooting someone is a relationship i guess like it’s not which is like a whole separate
conversation that i’m sure you can can have right uh we’re in a competitive relationship if we’re in a competitive shooter it’s
just not a very you know it’s not a very rich relationship it doesn’t have any emotional
uh well i mean as long as our ability to communicate with each other doesn’t get past that
right it doesn’t have any emotional context that goes kind of beyond that we don’t know anything else about each other and people will always find like
one thing that’s always striking to me is people will always find ways to communicate with each other if they possibly can through games
right and so you know if nothing else let’s say you’re in and this is like some imagined game right let’s say you’re in
some imagined multiplayer shooting game where like the the whole thing is just open arena
like run around shoot people uh they respawn after some amount of time into the arena again run around
shoot people again right there’s a number of games that are in this genre if you want to harass a player you camp
on where they are and you target them and them alone and you right like that’s a form of communication that comes
through even this like really non-verbal just shooting people way of doing things right like there’s a
way to be able to interact with the human and so it matters whether whether the entity that you are camping
on and harassing is an ai or a human yeah because it seems there’s no point why would i waste my time harassing an
ai so as the harasser i need to know that i’m harassing a real person sure
that’s one way as the person picking harassed you also probably want to know right and
it would be different if it’s an ai that keeps killing you over and over and over again versus another character like
another exactly yeah because there are real human relationships it’s not like you it’s not like you stop being human
beings when you go into a game even if you never talk to each other out of never interact with each other outside
of the context of that game even if there’s no voice channel even if there’s no no nothing right like there’s no
identifier that makes it so that you would know who that person is outside of this game you’re still being a human
with other humans right i think it matters let’s let’s move on to um there’s a
concern in almost every field about ai replacing human workers um and what their moral response to to this should
be um how do you think this applies to the game industry i mean we have game designers programmers artists producers
play testers etc um whose jobs will ai be able to
competently perform in the near future do you think perform right perform versus change right so this like this is
the eternal conversation about ai automation right i feel pretty confident
with this every field within the game industry people who work in the industry are already interacting with ai
systems in some way in the course of doing their creative work
many of which a decade ago we would have said oh it’s going to replace the people right so
this is like the core concern of procedural content generation right if i’ve got an ai system that’s going to
generate all these levels for me um will i still have a job right is there
still such a thing as a level designer when there’s procedural level generators it turns out the answer is yes because
there still needs to be human creativity in the process the job just looks a little different um
it’s not always the case that it’s like nai helping right um but but it just
changes the work that you’re doing or like um you know any artist who uses photoshop
and uses like context aware editing tools right highlight this person delete them
from the scene fill in the background with something that’s an artist who’s interacting with ai in some capacity
where yeah like two decades ago they would have had to manually and painstakingly go through that process right and now
they work off of what the ai gives them and then like move on to the next task it’s just it changes the tasks that
they do right it changes the work environment that they’re in right the creative though though it’s
interesting i want to see the creative choices still entirely in the artist’s hands but
more and more part of me is like well there’s all these little choices that you know we make along the way oh sure
uh you know that uh you know we pass on to ais you know a
little bit at a time but we still live ourselves some creative choices that we might think are maybe the more
important creative choices more important or harder to do or outside the ability of
the tool that you’re using also it’s not like the ai is some
agency carrying entity right the creative choice is actually being made by the people who wrote the ai system
that you’re using right sure in many ways the the the most
interesting ai ethics in creativity question to me right now
um is how have the tools that we’ve got how are they shaping the kinds of
experiences that people can make right the kinds of creative agency that they have
either in adding new possibilities or removing possibilities right like the medium is changing clearly i should have
started with that and that was fine i think i’m glad i didn’t have this is like my
core expertise so i’m glad we didn’t have to follow it with a bunch of things i know that’s about
so yeah tell us what are the things that you think are uh really the the core concerns that we should have
about uh aifx or what you think are the most interesting uh aspects of it uh
happening right now or in the near future um uh a few things so so let’s look at things like how do you decide
what game gets to be made in the first place huh okay certainly and so the industry is is
enormous and diverse right and there are always going to be studios who sort of ignore market trends
ignore market forces right or kind of like do their own thing and have breakaway successes in unexpected spaces
right but increasingly things like analytics on the app store are going to
drive creative design decisions right right what like we’re watching a bunch
of players play um like we’re running some kind of analysis on what players are currently doing
and using that to decide what makes the most sense to implement moving forward in changing the game
right what are the games that are already in the marketplace and can we analyze like
the different factors that go into making those games and use that to decide what gets made next
um you know it’s interesting because for for for me the i i i this this brings up two things uh
number one it brings up the idea that uh the focus then might be on
you know uh what ends up automatic what ends up getting players to spend more money um
you know rather than you know what uh would make for a good game because uh yeah it’s much easier to track
and get data and correlate essentially you know game functionality to
you know spending versus to enjoyment because you know it’s how you attract enjoyment and the other is the uh
uh remember the uh that job algorithm that facebook had yeah that turned out to be sexist yeah
amazon has also had these amazons also have these like built-in bias to the ai
system so that so that’s happening all over the place right and not just in games by any means but certainly in
games as well underlying bias in the ai system that’s leading to different kinds of experiences and and
really like any game that you make that that incorporates ai in some way as a
as a player experiences it the player is going to be experiencing the biases of
the person who created the algorithm right in in play all right so
in procedural content generation we’ll talk about things like um
what who gets represented and who doesn’t in in ai sim society type games
right um mike cook who is a wonderful ai researcher and
game designer um just has a couple of simple examples
right of like let’s say you’ve got a tombstone generator for for a game and you’re like i need like an entire
graveyard and the graveyard needs to have tombstones in it and those tombstones need to have text on them and
like i don’t have a writing team to be able to write all of the text for all of the tombstones so i’m going to write a
little generator that just generates all of the things right whose names
are in the generator right right what what data set do you work off of
for that how do you decide uh what a spouse is on the tombstone
generator right like do you represent same-sex marriage do you
right like there’s so much interpretation that goes into these
these little generative systems right the that the designer of the system
bakes in either intentionally or unintentionally right that’s a big deal uh yeah that’s that that’s interesting
and and that seems like such a i want to say almost benign uh kind of example but
it’s a very clear example can can you think of how this might apply can you give me a how this might happen written
larger i guess or you know in a kind of uh you know larger you know the little
names on the tombstone is really easy to understand it’s a really good example but okay what could be like you know a bigger
problematic let’s say also in many ways simcity is a really big policy modeling game
right okay um well so and and this is what so some of this is
going to get back to like what’s your definition of ai right well because some people will move some people would say
it’s not ai it’s it’s design right but typically like anything that involves
like complex simulation and modeling that has emergent effects i would say is is part of ai certainly part of game ai
so look at what kinds of policies you’re allowed to create in a game like simcity
um what are the assumptions about what a city is that’s baked into the design of simcity
right it has to have roads it has to have right it has to like exist in this
capitalist model right um it doesn’t let a player explore
a simulation that’s outside of the bounds of what the design team decided
counts and then that informs some larger conversation within a community around
like oh look at look at simcity right like as a colloquial example of
of like how i learned about city management right and then it’s like this this
model that is not inclusive of all potential cities that we already know about in the world okay interesting so
essentially we have our uh our design team our design team essentially brings their assumptions of what something is
like uh they fit they feed a data set into something that crea into an an ai that
creates um whatever city you will have in uh in sim city is is is is that the idea or
because that’s it’s it’s been a while since i played sim city yeah so simcity is more manually authored than that and
so this is also like these days i guess when we talk about ai a lot of the time we’re talking about
machine learning but ai and machine learning are not synonymous right machine learning is a tool or approach
within ai right so simcity is not like this um
it’s not like learning a city from prior data right it’s a designer’s sort of
very explicit uh authored model of of a city right with all of the different
components and what they do to different parts of the system for a long time games really shied away from machine learning because it doesn’t
give you as much authorial control oh that was the that was the argument right like they needed
game designers wanted to be able to have like tightly crafted experiences and so they needed ai systems to be
like handwritten and testable in a way that the machine learning approaches don’t don’t always permit let’s go back
to the previous question which was uh again tell me what sort of things uh interest you what sort of things are you worried about
or concerned about yeah i don’t know concern worry it doesn’t need to be like there’s some sort of
tragedy about to happen but yeah uh sure so a few of the that are gonna sound a little tangential to games but i
promise they circle back um okay so there’s all of this work in like ai
generated art these days right so um disco diffusion is is one of
the uh like pieces of software that people are using a lot but it’s all and i should not minimize it like they are
meaningfully different in many ways but the essential idea is like ai system trains on a bunch of images
you give ai system prompt it generates new art right like new image based on
prompts that you have given it right um and people are already seeing things like baked in aesthetic biases right and
um you know you you put in words the the acute is as very masculine in our
society and you’ll get these like gritty masculine reading things back right um
we don’t have a good language yet to be able to describe the
authorship and responsibility for ai created content and people will argue about it really
people people will fight about it like the ai made it therefore
i don’t need to be held accountable for the offensive thing that it made right or
um oh that’s interesting yeah which certainly is something that can come back into into games right
right oh definitely and actually did so so with
some of the earlier procedural generation work which part of me can’t believe is earlier right so look at spore for example
they have these procedurally generated worlds the qa team ended up needing to
make a list of what the acceptable random seeds were for generating worlds
so that the game will only generate worlds that the qa team deemed to be
acceptable before shipping what uh acceptable in terms of playability or in terms of uh
offensive offensiveness of some kind in terms of playability uh i’m told also a
little bit in terms of like is the shapes of all of the continents uh
appropriate or could it be interpreted as [Laughter]
um [Music] on some level it doesn’t quite matter what what is
meant by acceptability right it is the sense that like does a human need to be the arbiter of
like the final arbiter of what goes out into the world like does does the human
a human on the design team need to be the arbiter of what is acceptable how do you allocate credit
if you’re using so let’s say you like take one of these ai art ai art tools
use it to generate a bunch of like art right
uh slap that up as paintings in some virtual gallery that you then ship as
part of a game right right and let’s even take away corporate america from it like let’s assume that it’s some little
indie studio that’s doing this right right still corporate in some sense but not answerable to like the lawyers of
blizzard or ea or whatever okay um is it okay to do that right is it okay to generate art using one of
these tools claim it as yours put it into a piece of software that you
then ship uh yeah it’s interesting because uh this is this is part of what you said
earlier where uh if the ai does something wrong you can say well look it was the ai but
that means uh how can you take credit for the benefits you get from the ai and
the money that the that you’re getting from the ai but refused responsibility for any of the negative consequences
from the ai and in this case right it seems that uh i mean you’re selling it it’s you know you’re making money from
it it seems that you would be respon but it seems to be different if you’ve
if you’ve essentially procedurally generated it and then it’s kind of a set thing as opposed to let’s say every time
you run the game it gives you a different configuration why is that different well because you
have control over the final product you’re shipping so if you did it once where you can essentially your qa team
can look over the product but if you’re doing it every time there’s no way you can control what’s happening there right
so it seems that uh you know the in the first um in the
first case uh the ai is a tool for creating the product but then the product is kind of fully yours
where in the second version the ai is creating the product anew
every time in some way yeah and that that seems to me to to to make so at
least some sort of difference with responsibility because when it comes to responsibility we essentially you know
our common excuses are i didn’t know uh and i couldn’t control it right
and you know with uh the first product if it’s just an art gallery and i’m thinking of something like uh instead of
a regular art gallery i’m thinking of something like in um bioshock 3
where you have like the racist museum let’s say and let’s say you let the ai like generate some stuff for the racist
museum right um if you had it generated once and then you saw it at this point you
know right you’re you’re in control of it if essentially any anytime everyone uh
played bioshock 3 they would get a different uh racist gallery
created by an ai the producers of the game
they couldn’t know what was going to be shown and they would have no control over what is shown
well so but they know that they will have no control right and they know
that someone does have
control over the space of things that can be created right right right and
that and so and when we talk about procedural generation or generative design
really what a designer is doing when they create one of these systems is that they’re creating a space of potential
things right right and and it’s sort of the responsibility of the person doing that
to understand the space of potential things that they’re creating so that they feel decently confident
that some enormous corner of that space isn’t going to be horribly racist right unless it’s intentional but in that case
it needs to be racist in the way that you intend it to be for exactly effect of the like the social commentary of the
game and that isn’t that is a really difficult line even for humans to follow with one oh definitely like no one would
be stupid enough to let an ai actually generate an argument and yet you would be surprised
right because like i think that like that’s one of the one of the tricky things about this
right like there’s some level of of sort of promise and hope for what procedural generation will be able to bring like a
different experience every time like maybe it’s it’s interesting that it’s procedural maybe it’s part of the game
that it’s procedural where you know you you go you go once and and you see one thing but like another person who’s
playing sees a different thing and that’s important because then the players get to have a conversation about
how they saw different things i don’t know whatever it is right like that can be a really important and an
intentional piece of design if you need that as a design element in the game for
the kind of experience that you want to be able to create you also need to be able to accept
the responsibility for what happens when that procedural system goes wrong right because you have
given it to someone to use and play with like you have put your name behind it and said yes this is like this is
something that i stand behind as appropriate and so you need to be pretty confident that it’s going to be appropriate
right yeah yeah yeah and it’s really interesting how uh and again i’m thinking about the kind of
uh the sexist job uh things on facebook how you know
you know i definitely think those programmers are responsible for uh for the outcome
though there’s a question with that one of whether you know they should have known whether that was negligent or not and i think
the question of kind of you know negligence and recklessness uh right is is one of those things where
knowing what the potential space of the outcome is seems really really
important uh and something that uh programmers would need to give thought to right and so i guess
i my argument would be a sufficiently large portion of the
population could have easily predicted that as the outcome right like i expect
i fully expect that if you take an ai system and you train it on
something that we know to be biased it will replicate the and amplify those
biases right it’s like it is simply the nature of machine learning to do that like literally that is its strength
right it identifies patterns that we know about and it identifies patterns that we don’t know about
and drives all decisions made in the system based off of those patterns like
literally it is what machine learning does right let me ask you based on this
uh would you say that uh social media companies like facebook and twitter
were negligent or reckless in uh essentially the impact they the divisive
impact they had uh on the world given uh what those algorithms were that they should have known from the very
beginning they should have known they should have had an audit process to be able to look for it
even if they didn’t know right even if you’re like oh they had this like idea in mind right they should have known
what they were doing there’s a real people um the like the technical
the technical approach is known to have these kinds of of emergent effects right even if they
didn’t have even if they didn’t have teams of people whose job it is to study how people behave on social media which
they do right in their research arms even if they didn’t have those people i
would argue that they should have had those people because it is part of their job right to facilitate human
communication so they better be working with people who understand human communication right like yes at every
step of the way i think those i think social media platforms are responsible for the impact on society that the the
very design of the platform is like intended to create and but by the very
desired platform we’re talking about the use of ais in particular to kind of do this sure any kind of recommendation
system yeah right do you see anything like that uh in the game sphere that would be
remotely comparable i mean there’s kind of community management type stuff right for multiplayer games and there are some
um you know how do you design a platform to write how do you how do you moderate a
platform how do you control for hate speech on your platform um there were for a while i’m actually i’m
not as sure about the current state of things but like some ai uh
like ai tools that are used to kind of like monitor and detect toxic behavior in online community spaces
um that’s definitely an analog yeah it’s interesting how you know the ai tools
being being used to identify versus changing the game or something in response to
identifying those patterns which is what social media does right right
yeah like you know a really interesting thing and i don’t know if anyone’s doing it
but now i kind of want to like after this call find out if someone’s doing it um
is so that there’s this idea in um in cyber security
of like detecting bad actors and then like intentionally honeypotting
them right like um and so this is actually and i’m not a cyber security researcher
so my knowledge like i may be explaining this wrong um but the idea behind uh
platforms like cloudflare right which now everyone like routes their internet traffic through cloudflare um for for
security purposes right they initially set up some system of like what if we could detect
the bad actors and like route them this way right okay and then let everyone
else go down the main like the main path but like we’re gonna find the people who are misbehaving
and redirect them so they think they’re still going down the right path but actually we’re like honey potting
them into this like cat like labyrinth essentially that they’ll get kind of trapped in
huh so this is a kind of virtual entrapment yeah virtual entrapment but like for good
yeah yeah yeah yeah yeah but yeah i mean yes virtual entrapment um
yeah i wonder if this if people are doing things like that for for social media or for like online
online game playing right if you can detect that someone is being a tremendous [ __ ] in your game
all right can you find it and then just have them play ai players
so that they still get to be that right but like not to other human beings
this takes us back to the earlier point of identifying or you know whether the
person is a player ai maybe sometimes we don’t want to identify it because
yeah that’s that’s that’s a great way so people can those people who want to grief can essentially
still grieve or still feel like they’re griefing but they’re griefing against ai right and no one’s getting like no one’s
getting hurt i mean except to the extent that you’re permitting the the anti-social behavior right which is
longer-term harm done um but but maybe actually maybe relatively less than
because people no one’s getting no one’s getting hurt by it it’s interesting this is definitely manipulative oh it’s totally
right manipulative for the power of good right we can use ai’s for that no you know i mean manipulation is one of those
things that we often talk about is bad but you know i’m a parent i mean i manipulate my child like every other
parent yeah let’s look at let’s look at behavior charts like i have a number of those spread across the house of my kids
uh yeah right yeah um jillian i i think we’re i think we’re about done uh so let
let me ask you this what final words do you want to live uh our listeners with when it comes to ai ethics and games i
say that someone is always accountable right you don’t get to hide like any i
would say anytime anyone hears someone say something that seems like it’s shifting
blame onto an ai system don’t believe them okay so like someone someone made that thing
right for good or for bad right it doesn’t get to be the ai
uh the ai’s fault or or virtue right it’s it’s always a human behind it like
ai systems aren’t they’re not magic that a bunch of numbers people can learn how to make them it’s it’s not that hard
um yeah it there’s a there’s always a human wherever there’s an ai
so look for the human all right keep looking for those humans jillian smith thank you so much for
coming on our show uh good podcast gp all right play nice
everybody you can subscribe and listen to all of our episodes wherever you listen to podcasts
