Writing with Technologies - AI and Writing for Games - Revolutionising Play and Design

hey welcome everybody I can see there's a couple of people still coming in um but I'm going to talk for a couple of minutes so that's fine really great to be here I'm really excited for I know is going to be a really fantastic session today um I'm Sarah Campbell I'm from my world and I'll be chairing the session so from generative writing tools to augmented publishing processes artificial intelligence is rapidly changing and challenging the landscape of creative writing and Publishing so in this free four-part webinar series which is a second series of webinars from my world and barbar's University's Center for cultural and creative Industries and narrative and emerging Technologies lab we're really taking an in-depth look at ai's emerging influence across writing and Publishing in multiple Fields so I'm really excited to welcome three fantastic speakers today inen Joseph and Richard and this is the third of the four-part webinar series focused on writing with Technologies so today our focus is on AI and writing in games with a specific focus on how these tools are revolutionizing play and design so we're going to be exploring The Cutting Edge intersection of artificial intelligence and the gaming industry bringing together game developers AI researchers and Industry experts to discuss how AI is transforming both both game development and player experiences with a focus on accessibility and inclusion it's really great to see some people um introducing themselves in the chat please feel free to introduce yourselves to each other um each we've got three speakers today each will speak for about 10 minutes and then we'll take questions um for all the panelists at the end so feel free to put questions in the chat at any point um but we'll leave them until the end so without further Ado it's my absolute pleasure to welcome inen e so invent is a play Alchemist creative technologist facilitator educator and tedex speaker specializing in how play is a powerful tool for exploration and transformation so with nearly two decades in immersive entertainment and digital transformation her expertise spans interactive design game mechanics Workshop facilitation and creative technology making complex transitions accessible through playful methodologies she focuses on play in adulthood to address social political and environmental challenges she's the founder of new party rules Labs chair of my world's Advisory Board which is a 30 million pound creative technology initiative in the southwest and co-organizer of bxn black and extended reality Network in 2024 inven spoke at the UN Geneva permanent forum and G20 Rio advocating for the transformative power of play in governance and Society previously she's worked with shunt punch drunk taught at iscom Paris University of the Underground and Bristol University so rooted in both London and Bristol inven continues expanding approaches to cultural play and technology so I can't wait to hear um your talk inen so over to you thank you Sarah a um great introduction okay we're going to get started you might bear with me because my screen is a little bit sticky but we try I can see that that right here we go okay so a few years ago there was an AI driven game called AI dungeon which let players create infinite text experiences I don't know if some of you are aware of it AI would generate the stories in real time based on Play Input but something alarming and revealing happened players realized that AI was was n just following the rules it had started making up its own logic creating unexpected and highly offensive narratives that included sexual assault and child abuse themes exposing the potentials and the risks of AI learning through play players weren't just playing the game they were co-writing it they were teaching the AI through their interactions and some of these teachings led to concerning narratives that reflected human biases and harm ful patterns the meaning wasn't just in the words it was in the complex relationships between human input and machine learning which raised critical questions who shapes this learning who is responsible for what emerges and how do we create ethical Frameworks for AI that learn through play hello I'm inen I won't repeat this because um Sarah gave me that fantastic introduction but just to say I'm a play Alchemist I think that's the most important bit and I make games with and without technology and I'm genuinely genuinely excited about ai's possibilities in games and play but I think that we need to be thoughtful about its implementation we often think about AI as something that generates words the fing of black s on a script but AI isn't just writing it's learning as we're learning from every interaction AI is inherently generative it doesn't just output Tech it outputs it adapts it evolves patterns and anies them like our bodies in play faes it responds to the energy we bring so think about how your body naturally adapts through play leaning in stepping back matching other people's rhythms AI can be like that too responding not just to our words but to how we move and how we interact in space and we need to be intention intentional about what patterns we want it to learn and amplify in AI dungeon players weren't given a fixed story they were shaping narrative as it was co-created in real time and they were contributing to ai's understanding of narrative behavior and interaction and this is the power and responsibility of AI co-creation so we need to think about what does it mean for game writing and Ai and I think it means creating systems that are not only that not only generate Dynamic meaning but we must do this with awareness of the impact of learning AI in games isn't just a tool par participation in creating meaning and we need to design a of how it's learning from play but before we dive deeper let's clarify something important and this is something that I always say when I'm teaching um mechanics as meaning in games there's a difference between story and narrative design a story is what you tell your plot your character the events a narrative design is how you tell it through game playay using mechanics systems player interactions to create meaning in games narrative or narrative design is the art of creating meaning through game play I'm repeating myself but it's good through game play through systems mechanic and I think this goes beyond the written story and I find that when most people think of narrative than games they think and they imagine video games cutscenes dialogue and text but games themselves already tell stories through the language of play video games or the many other types of games that exist all right meaning through mechanics creating narratives that emerge from player interaction let's use game portal there are some images from portal portal is a video game and the game is about power control and escape and every mechanic in this game tells a part of the story the small rooms are about confinement the portal gun equals agency and problem solving because the portal gun lets you break through the rules of space itself the hidden walls in Portal signify rebellion and Discovery and even the antagonist GLaDOS who is an AI operating system tells her story through how she manipulates the spaces and mechanics this relationship between AI space and mechanics raises interesting possibilities games already create spaces where movement tells stories but how might AI enhance these mechanical narratives through the ability to learn and adapt through play my first provocation for everybody listening how might AI enhance the stories of our game enhance the stories our game mechanics tell in this section I was going to talk about is a playful narrative partner learning from player behavior and shaping emerging stories using NPCs as an example and for those who don't know NPCs are non-playable characters so the characters in games that are controlled by the game software but I had to stop because NPCs are problematic for many of us for as black and brown people for indigenous people for women for working class for queer for Trans MPCs hold significant racial and ethical concerns the idea of mpc's learning from player Behavior quite likely could amplify the existing toxic interactions and reinforce existing problematic patterns I decided to get us to think about AI as a thoughtful partner in play AI is gr games isn't about replacing human creativity or learning uncritically from player Behavior instead we could design AI systems that support intentional play patterns that enhance physical storytelling while maintaining boundaries AI could amplify positive Collective experiences so let's consider how AI could support Collective movement and play and it could do this by recognizing group dynamics in physical play Spaces by supporting facilitators in creating inclusive environments by helping scale positive play patterns while preventing harmful ones this approach could move us from AI as an uncritical mirror to AI as a thoughtful partner in creating meaningful play experiences let me share cognition this is a game that I originally created in 2017 that demonstrates how simple play mechanics can embody complex social dynamics we recently took cognition to the G20 in Brazil as a was part of a conference that was an official Civil Society event of the G20 so we had officials Community leaders and Civil Society groups experienc ing democracy through physical play I can get a sign off for the workshop photo so we have to look at the logos so cognition as a game is a physical multiple choice game where players navigate real world ethical dilemmas unlike AI dungeon where meaning emerges through AI interaction cognition creates meaning through Collective decision making and human movement the core mechanics of cognition are majority where players vote on dilemmas and scenarios and experience the weight of collective decision making feeling what it means to be part of or outside of the majority freedom of speech this is a mechanic where players can step forward to advocate for different solutions their physical PR presence becomes their voice and the power to challenged and flayed is embodied mechanic three minority rights challenge teams can initiate physical challenges to reclaim power and physical efforts represent the work of advocacy four proactive points players and teams can predict how others will vote by positioning themselves early and that way people get to experience the tension between leading and following and understanding how power shifts through anticipation so the power of playing a game like cognition lies in making abstract Concepts physically real so at at the G20 we watch people from different cultures and positions of power physically experience the weight of decision making the efforts of advocacy the Dynamics and power of influence with AI we could explore how it might enhance that physical learning by understanding things like cultural Nuance modern context how ethical and racial concerns show up in the body how physical needs and patterns look through the lens of movement and my second provocation what truths about power do we learn only when we have to move our bodies to play I'd be remiss if I didn't talk on what's behind the curtain of AI I know everybody knows or most people know but it doesn't hurt to reiterate because there is an invisible infrastructure and human cost to AI technology AI is not neutral it's not magical and it's not in the cloud it's built by people and it's powered by low wage workers tagging massive data sets physical infrastructures that require Mass amounts of energy and water and this matters for game design because AI systems often amplify and normalize existing biases and we still have the opportunity to embed design in Co cooperation in care and in ethics but to do this we need more diverse voices in AI at the discovery design development and delivery stages all with greater awareness of ai's real costs come to my conclusion writing in games isn't just about what we tell players it's what we let them do AI can be a tool for dynamic emerging and inclusive storytelling not just by generating words but by shaping systems of meaning my game cognition is an example of how physical physical meaning enacts through decision-making mechanics and with thoughtful design AI can enhance this process while respecting human agency AI should be a co-creator cretive partner and not a replacement for human imagination the future of AI and writing in games isn't in just crafting dialogue it's about creating possibilities for play while being mindful of who creates those possibilities and who they serve and I'm going to leave you with the last provocation what if ai's greatest gift isn't what it can create but how it teaches us to play differently I'm in V you can find me on Instagram at of course I talk of course unor I talk to myself or via email inven newparty rules.com thank you thanks inven that was fantastic I really loved how clear you made some of the ethical challenges and I love that idea of actually AI is an opportunity and something to be excited about but it's also something to be wary of and it's what we bring to the tool rather than being slaves to the tool basically so um yeah and I love that idea about how can we be thoughtful partners and sort of co-creating how these tools were applied and um you know what ethical Frameworks are needed to help shape the learning in positive ways and constructive ways and um yeah I found what you highlighted about NPCs really thought-provoking so thank you for that we have a request from Jake in the chat if you could um repeat um the mechanic oops about freedom of speech in cognition but I'm going to ask you to write it in the chat if that's okay so we can move on to Joseph because I'm mindful of time so our next speaker is Joseph wil so Joseph is an artist and programmer who uses the digital to explore disability and uses disability to explore the digital he deconstructs misuses and repurposes software and Hardware to challenge Notions of ownership narrative and visibility so over to you Joseph cool thank you just share my screen I'm hoping everybody can see my presentation yeah it's looking really good thanks cool brilliant thank you so hello yes uh I am a an artist and a programmer um I have a long 20 year background of AI when it was cool and then it wasn't cool and then it was cool and then it wasn't cool and maybe it's cool again I'm not sure um I just wanted to share some of my thoughts and feelings of my artistic work with AI and uh my technical experience with AI I use the word AI a lot in this talk and I think it's very important to be specific when I say AI I mean large language models or large uh image generators so things that are trained on vast quantities of training data there aren't actually that many of these things things like chip PT Gemini um Claude anthropic um or Del some of these generators and when I say AI please consider that's what I'm talking about and most of my critique is about that because most of my spam box in email is just full of people trying to tell me how AI can fix all my problems from my dishwasher to fixing a wheelchair so it doesn't mean much anymore uh I am supported by my world uh the Watershed based media studio uh created with a group of disabled artists a tabletop role playing game called cryp ship if you don't know what a tabletop role playing game is um it's a form of telling stories together really of Make Believe of being creative um rolling dice to decide outcomes which again just lead to new stories um usually there's a Storyteller who's kind of describing the world and players get to say what they do in the world and the purpose of this game the tagline is disability save Society from Big Tech um a Revolt this game is about um resisting Ai and kind of comes from a place where I can already see uh AI being used as a as a presented as a solution to not having to fix any of the social problems that cause disability and so this game is a reaction to that um it's about breaking large language models and very specifically in the real world so this is you know we've played this game where we uh used a customer sport chatbot on a booking website for a hotel where we play the game about breaking it so it exists very much in the real world and you don't have to be disabled to play it but it comes from the perspective of disabled people uh destroying systems of power and being new World building new worlds from disabled imagination because that doesn't exist in popular fiction and it's fun to explore what disabled fantasy actually means not the first to do this two very cool references if you're interested in Reading disabled writing and narratives um oncy magazine disabled people destroy Science Fiction and Fantasy when you don't see yourself within the popular fiction um these are opportunities to break it down and present unique perspectives if you want to talk about creativity these are very very creative and very good the core kind of like Drive behind how you resist AI is um a very long-winded research paper but effectively talks about that when people see AI as magical um they're more likely to adopt it so if you demystify it uh they actually don't tend to like it so our game is a lot about creating AI literacy um in order to demystify it hence discourage people from using it if you want to resist something you have to cost money to those who make it and that's where I'm coming from the game is orientated around what we call AI slop slop is really the word that's become adopted for AI generated content that's just not very good is is the best way to put it um and our game is really all about people discovering this AI slop um so it's a role playing game so it's orientated in people telling stories um mechanic and the only thing that matters in this game is that everything leads to people telling stories um there are aspects of large language models within it but they should lead to people telling stories there are dice roles they should lead to people telling stories so human interaction is everything AI is the theme within the game so it's the players work for the ministry of AI spillage who are responsible for cleaning up the the slop that AI companies have been generating um there's a lot of Illusion to the real world there um the game has make believe and storytelling and a fantasy that players create whatever their background dictates what that fantasy is um but there does end with people coming towards interacting with a large language model in terms of how you structure that interaction there's a lot of engineering work in terms of like trying to prompt kind of prompt engineer that to have interesting outputs because humans detect very quickly patterns in AI output but also we're in a game where humans are telling stories together so it's impossible for the L agage model not to be dull because humans are just amazing at telling stories together but there is an aspect where they have to integrate with an llm and there's no point in doing that without providing also the tools to Break llms um so we provide options to like you know run the models in unstable conditions to jailbreak them so that they work and reveal hidden training data and we do that with allies because sometimes can lead to difficult places but also how do you resist Well Community is is a clear aspect of resisting if I have to make one admission of um most like llm content I find fairly boring but when you run them broken is the most creative most interesting output um this is an example of running an LM from one of our games where uh it's running with a very high temperature which effectively without going in detail means you get to like surface a lot of the H training data that's hidden away you get a lot of the commercial tone in the training data you just discover really weird and wonderous stuff that's hiding in there um if there's anything that's interesting about llms they're most interesting when they don't work which is funny how do you structure this game so it remains critical when we break that magicalness of AI because you know we play games with people who have never seen AI before they come into it it's hard not to feel like it's magical um so we try and structure this game mechanic around having kind of challenges for the players to undertake in their interaction with it um very much focused around the idea that everyone who comes to the game has their own lived experience and we want them to use that to try and discover for themselves how large langard models don't work so that's things like you know why when you ask to draw a picture of a scientist does it draw a white young male um when you talk about uh your pronoun why does it ignore pronouns um why does it contain copyrighted content why does it present facts that are not facts um why does it have a value system encoded within it kind of providing the framework for a critical engagement with AI um while still providing space for the fact that some people are still going to be positive about AI after it that's the nature of games and playing that sometimes it goes into interesting places but that's also fun in terms of the writing of our game so we had a you know game is about comes from a disabled perspective is about critiquing AI so we had the rule that we weren't allowed to use AI in the generation of any content within our game but we did use it to check for example when um we named the book cryp ship uh we wanted to check that an llm could not generate that because that for us is a sign that it's a bad quality name so a lot of our story Concepts went through like let's make sure that an llm can't generate it because that's the sign of low quality in this in this current Modern Age also um this book comes from a place it means something right like we we took a team of disabled artists we had to fight through so many AI Bots responding to like calls for illustrators just to find some disabled artists who could create work that comes from a meaningful place that all the decisions used in writing the book in creating the pictures come from somewhere and I could give a disabled artist like a commission a prompt almost and I know that their lived experiences is going to take it in an interesting place as soon as I talk about using llms or any sort of like offloading of writing I'm talking about giving up decisions to an average of authors Based on data on the internet and uh it doesn't really come as a surprise to many disabled people that um llm simply cannot imagine disabled people with anything other Through The Eyes of nesa people right um and why is that well that's because popular fiction also can't represent disabled people just look at any Hollywood movie and look at a representation of disability it's not for disabled people it's for non-disabled people to to relate to and for their enjoyment so it's it's kind of unsurprising that llms are fairly discriminatory against disabled people can't represent disabled narratives because popular fiction cannot either in terms of uh sales um I think it's also important to quantify that um we have ai content in our book explicitly as examples of how AI does not work so the writing is us the pictures are us but we have these examples but in selling the book everyone requires us to disclose whether we have ai generated content in the book or not and we do so we have to flag this which means within the writing and the fiction and the tapes or role playing World there is a massive resistance to AI from a purely consumer perspective people don't want to buy AI generated content from my experience so we had to release a redacted version of book hiding how large language models don't work so we could tick that it didn't contain any AI generated content in order to sell it more effectively which is a I guess a good thing but also a clear thing that there is a consumer there there is a resistance to a already in certain filter fiction and even something that's resistive of it has to you know abide by that uh what am I doing for time okay there's also a lot of research I haven't got detail to go into this but there is a lot of very questionable research coming out around co-writing with large language models to summarize this entire research paper Microsoft researchers decided to do some research using Microsoft tools which outcome was that using Microsoft's tools is good and profitable and readers want it um there's a lot more there to unpack um but I'll leave that for another time but please be very critical of any research that's coming out because the quality is questionable so in conclusion um if you're deciding to use llms in your creative process in co-writing um you you have to be very aware that you become Holden to an ideology um the US and China both acknowledge that using AI as part of their trade policy to enforce their beliefs that their control and by using one of these tools into your creative process you're you're living under that control and much like game developers using things like Unity discovered that can be really harmful to your creative process when suddenly Financial models come along that are really harmful to even Indie creators uh um Dr timnit has talked a lot about this ideology called test creel um but without going to too much detail there's a lot of belief systems um that are present within the developers of AI that are rooted in Eugenics um probably the most relevant at the moment is long-termism with the idea that there are more people in the future so what we do now um could be justified we can pretty much do anything because as it will make the lives of people in the future better and this is being used to rationale creating harmful technology that is hurt that is that is hurting people specifically disproportionately marginalized groups to justify not fixing the social issues now but building space rockets by building general intelligence that effectively will supposedly solve all possible problems um I highly highly recommend because Dr Tim Nick can talk about it a lot better than I can about some of the belief systems that deep seek talked about in the research paper open AI talk about subscribing to long-termism Elon Musk talks about long-termism Sam Orman talks about long-termism it's important to understand these are rooted in Eugenics um and as disabled people we know how the what's that phrase the the ends justifies the means is not a great way for humans to go we also have the whole a military Association it's not surprising that these companies are dropping their support uh sorry are adopting military applications um I am not okay with this I don't want to use creative tools that are associated with military applications it's very important to be aware of this Association that's present in all of the AI tools I has struggled to find one that isn't involved in a in a military Contracting I think inw talked about this really well so I can kind of skim it but um AI replicates colonialism um uh if we look at where the data workers are where the extraction of Labor is coming from we have the global South that is effectively being extracted for the privilege and profit of the global North if we look at the pollution aspects of data centers and pollution it disproportionately affects the global South um we have massive monolithic models um that are trying to dominate every possible thing um and it's really really important to not repeat the the uh horrific consequences of colonialism and ask you know these tools are creative and they are useful sometimes but really be aware of who suffers I highly recommend checking out the data workers by the debt Institute which has interviews with data workers and they're talking about their their lived experiences with these tools working with AI please inform yourself about the people who are doing this work and what their lives are like so if I would give you some recommendations to end on the Dare Institute is an amazing Institute distributed AI Research Center um run by um Dr timnit among many others um mystery AI hype Theater 3000 is an amazing podcast a stream um effectively mocking uh AI research in a very uh very academically rigorous way I highly recommend it and cryp ship is a game you can buy we wrote it and you can buy it so thank you thanks so much Joseph as someone living with a disability and also always up for resistance and Revolt I'm definitely going to buy your game so thank you and it was really great thanks for defining kind of what you mean by AI in terms of llms because I see that a lot the word the term AI is used but very rely is that defined what we mean by that and I love that idea of a sort of design principle of actually if an llm can create it or generate it then it's not very good so let's not use it so I just thought that was really interesting as part of your design process and yeah a great kind of summary and introduction of some of the more critical work that's going on and research in this area and time posting um us to to where to look at if we want to be more informed so thank you thought that was great so um we'll move on any questions you have for Joseph please add them to the chat or the QA Q&A and I'll pick them up after our third and final speaker so delighted to welcome Richard Cole who's a lecturer in digital Classics at the University of Bristol and co-director of the Bristol digital game lab so Richard's research focuses on history and culture intersecting with New Media in particular video games VR and artificial intelligence he's published leading research on Classics and video games and his latest work is looking at how questions around authorship of Iliad and odyssey can help us better understand AI generated texts through the game lab he's leading the research component of a collaborative R&D project with industry partner meaning machine on generative AI systems for video game dialogue with the aim really to understand how players respond to conversing with AI powered characters so over to you Richard thanks Sarah wonderful to be here and and fantastic talk so far um hopefully I can follow on a little bit from uh some of that um I will pick up with a little bit of History um the historian of me couldn't resist this and I I hope it sort of further contextualizes some of the um fantastic sort of discussions I think we can have at this intersection um so a very very short introduction to Ai and games um because again as we've already been hearing when we throw the word Ai and games together it actually mean a vast array of different things depending on how we Define what AI is and how we think about its interrelation with games um if I can break it down very briefly I think there's two main ways it's sort of fallen out one is developing AI systems within games of which generative AI is the latest iteration but this can be traced all the way back to pretty much the 50s and 60s before AI research even really took off um so there's been a sort of very longstanding relationship there which I'll pick apart in a minute um the other one is AI systems being trained or sort of developed through playing games um or AI is being put into um game world and sort of tested as a result um and one of the really interesting examples of that is um is with the game Doom from from the early 90s um and I think the reason I wanted to raise this in this context was we're all probably very familiar with games but many of us may not be so familiar with the actual inner workings of AI systems but when we start to see AIS playing games it actually becomes much easier to understand what's going on to dispel some of that magic as as Joseph was giving us and with doom um researchers were able to to develop systems that could play the game but really what that means is developing a series of individual systems that understand the Fireballs in the game World um the space around them how you move in that space and then tying all of that together to A system that can actually play the game so for me anyway it's very much helped you can use games as a framework for thinking through how these systems are built um and that can also mean you can use games to Think Through the problems and the biases and the inherent um you know underpinning sort of limitations of these systems as we've done through the game lab running um kind of game jams to explore uh algorithmic bias for instance so games I think are really productive tools here to think with as much as thinking through how how AI can can help us um create new games and on that point of bias there there there is actually some funny research that open AI did in this space which I think can really draw attention to those limitations um uh so they they got some of their systems early on in in 2016 so this is long before the release of of chat GPT um to uh they were exploring the reward function um of certain algorithmic systems um and machine learning systems and they they they got um their systems to play uh this this sailing game and of course humans would intuitively try and um complete the map in the quickest possible time to get the best results um the AI system decided to uh crash itself into surrounding boats go on fire um because this actually it meant that the the system would achieve a higher result but it basically reveals the sort of problems of reward function uh so again games useful as a way of thinking through the the inherent um challenges in building these systems um and their limitations moving more towards kind of though how AI systems have been built for games in the sort of mid um 2000s 2006 Oblivion um Elder Scrolls Oblivion by Bethesda came out and this was one of the first games that um developed what we would now think as a kind of large language model system or generative system within the game for powering the non-playable characters um we've already heard that this is a an area of work um from inven um and this was interesting because when they first built the system uh again severe limitations um the AI MPCs in the game especially if they had um sort of preset you had to be say a drug dealer in the game then what happened is all of your um uh people you were supplying to went and killed you um as the NPC so it sort of revealed the limitations of this system um because it it it would be too creative almost in its own in its own sort of way of working so they had to tone down basically the the complexities of the system to actually make it much simpler uh because it broke it broke the game that system then has been developed we've already heard about AI dungeons so I won't talk so much about that but the other way the reason I wanted to to highlight it was um it's where we see plugins so existing large language models the earliest version of um AI dungeon used gpt2 then it used gpt3 3.5 so it started to sort of use the system as it was developing and again for those who sort of got into the AI space before chat gbt burst onto the SE sort of scene there was a lot more interestes in creativity that was do sort of you could you could do with these really early systems sort of GPT 2s and threes where where there were sort of prompting words or individual phrases or sentences and you can kind of pick that up challenge it take it somewhere else move with it you could do quite a lot of really interesting work as it's gotten more and more complex and and generates more and more we actually lose a lot of that but I still like AI dungeon for one one thing that you can do with it which is sort of to go back in and edit and change and move the narrative around as inone was saying you can do a lot of creative work in collaboration there of course it also raises all the other problems and questions we've been hearing about and then we start to see some of that come together so we've already seen how Oblivion um sort of experimented with early AI systems with the the game Skyrim that's still being modded to this day we now have um modders developing fully interactive uh AI MPCs that you can converse with in natural language and will deliver a natural language response back to you um and sort of romantic data we know this has been sort of framed as a as a as a sort of huge step forward in the possibilities of role playing games uh especially if you're dealing with a kind of companion character um again there's there's an interesting problem problems there um and I'll come back to that so near the end of the talk around what this actually means for players um and then sort of moving even more towards the present we then have people using all of these systems that people been experimenting with testing using mods to explore the potential for um and now we're seeing them actually sort of pres put put into the building blocks behind games that we may very well be familiar with already so the sort of game worlds like The Sims here's a kind of a game like The Sims that's being built from the ground up but with these systems ing it now I have particularly sort of I'm I'm I'm not sure this is a really good way to go because it feels like you've got games that we're very familiar with and then you're just adding AI into it um not really sure exactly what you're going to get out of it and and the true benefits um something else I'll come back to in a moment just wrapping up my my quick history here to sort of Go full circle um so we saw how Doom was used as a as an sort of training space for AI systems well there's now ai systems can completely hallucinate the game Doom simply by playing it um the reason I want to of end on this point is because so some of the wider commentary when when this paper by um by Google researchers came out was basically well that's a bit useless because you know you're never going to actually be able to play a good game because all these systems if they can start to generate game worlds completely from um the the pro the prospect of just playing them um you you're never going to generate something new so actually for me it sort of it reveals the inherent problem s of of generative AI from from from the ground up which is that they'll only ever be able to parot some level of what um what has gone into them and here here is a you a very very clever system parting a game that was built back in the early 90s that is you forged and generated an entire genre of Game World entire moding communities um Etc um you it's hugely important culturally um here AI is just hallucinating that again it's not going to generate new it's not going to go beyond that it's not going to give us something thing um to to to sort of really work with there um but that's sort of tying the tech development together in a circle obviously there's so much more there to say um I haven't barely even touched on the fact that kind of machine learning deep learning is something that's that's really now very much present in all games in some shape or form and has actually um played a huge role in informing the kind of the way we play games um and again it's kind of a hidden re Revolution that's sort of been happening for decades um but it's only now that we've got generative of AI bursting onto the scene culturally that we're starting to question what does Ai and games mean um so in a way there's been a kind of continuous long process there going back for for a long while um and now we're starting to kind of think about that more in the Limelight as well as thinking about the generative side of things so yeah the generative side really just sits as a very small part of what Ai and games means more broadly um but I do think it sort of allows us a way into that the wider debates in a way that that perhaps we um we weren't so interested in before but the sort of second part of the talk before I turn to the research side and the third part I just wanted to touch on the work that's been done around kind of what what players want because I think that's the other interesting question to ask here you know what what do actually players say they want when it comes to to Ai and games and there has been some research done in this space um again big caveats here it's been done by uh companies that are trying to market the tools they're trying to put into games so a lot of This research this kind of research um less the academic side more more the kind of commercial side um can be um challenged on that front but it is interesting because of the sample size and what they um say they that what they've been able to distill from that and and that's largely the players seem to definitely want in this case AI MPC so it's that it's a sort of subset of a subset of a subset of what generative AI can do um but that was their kind of um particular response um with that in mind so players say they want this and that's what sort of the industry um you how how is how is the industry responding to that is is the next sort of Step I wanted to to take us on and inor have presented some further research since they released that the white paper I just quoted in 2023 and they released another one last year in 2024 which suggested that developers were from about a sample set of 500 or so and were really excited about this and that this was something that they were going to to definitely go ahead and do in some shape or form AI MPCs as you can see there was sort of the the third most likely thing that people help developers help they would go on and work on what's been interesting interesting is that since that came out in 2024 the game developer conferences state of the industry paper for 2025 released at the start of this year suggest actually there's a a hardening or a souring against the technology more broadly bigger sample size here 1500 people although again there are some problems with that data um the the percentage that are from AAA Studios or even able a or Indie is actually quite low um so sort of about half of this is from a kind of more hand gential Market or people who work in and around games rather than sort of perhaps what we think of as as core developers um so perhaps that's that's coming through in the data as well um there are some other sort of interesting caveats that the people who say they use it the most actually tend to be again from the business side rather than from the creative side um so there's I think there's there's places where we'll start to see AI being used just routinely in automating certain tasks and processes but less so in the kind of the creative spaces the more complex one um definitely um and and that's also because as I've highlighted here issues around intellectual property energy consumption quality of a generated content the slop that Joseph was talking about potential biases we've already heard about and also regulatory issues which may come around and basically just make all of this much much harder or at least very difficult in certain markets in the world and I wanted to finish this little bit of this section by looking at um uh Tommy Thompson who's done some fantastic work been working in this space for for for decades um has an academic background in Ai and games and is now running a Weekly Newsletter which I highly recommend um he's been highlighted some interesting case studies that I think speak to where we will see this being used creatively generative AI is what I'm referring to here and he gave a recent example um for Capcom that are using tools to develop kind of early ideas for particular sort of assets in the world or ways they might be visualized the idea is that once they fleshed out they would no longer use the tool in an would go away and build all of that which sounds great in many ways um the sort of flip side of that and this is what some AI law researchers been looking into is that that can often mean that the sort of Junior tasks that early stage artists designers might work on um are no longer the things they can work on because U they're being taken by the kind of automation side they're being taken over by automation um and that could have a longterm negative impact on the industry it's not doesn't seem to be being picked up in the context of the kind of massive problems around um layoffs that the industry is having at the moment in in indeed in the GDC report people aren't sort of linking Ai and job losses together because of the wider problems the industry has um but it could be the long term you have a sort of Junior cohort that don't have the S of same skills the more senior people doing the industry now and that could have long-term impacts on how the industry runs and works so you it can be fine to have them as sort of early stage creative tools but there is still an interesting context there to think about as well the final thing I wanted to come to was some of the work that we're actually doing um to sort of pick apart some of these assumptions in the kind of more commercial research and indeed in the academic research around games and Ai and writing um and I wanted to frame it with this quote here from um uh from inor again uh in their latest white paper from last year they say player engagement in immersion giving players the agency to impact the narrative in meaningful ways through your actions and also to talk to NPCs as they're real people play real people will inevitably make gaming more immersive and engaging this is their kind of assumption and prediction um again based on actually quite flimsy work that they've done or or kind of just perspectives rather than actually being able to point and show that this is indeed happening they have gone on to do some more work with with people just with industry people trying their tools that I think is more interesting than this more kind of high level stuff um but that's what we wanted to to poke at as it were when we um started work on a my world collaborative R&D project with IND lead meaning machine who are doing very different things in this space and I think the model they're working with is one that that I'm very interested in seeing develop because what it is is it's about telling new stories and developing even what you might call new genres of games indeed their Prototype game Dead Meat um which is the one we're using for our current study has been termed by Thomas Keane the co-founder of mini machine a first person talker so a new kind of genre of game made possible by gen of AI but hugely and highly constrained by human creativity that basically tailor every part of the kind of conversation that you can have here and Demi is a murder mystery game um that is uh currently playable as as a demo and we're um using that in our study um it requires text input or sometimes voice input um and you get either text or Voice output from it um and we're using it as our as our model to sort of really try and plug a bit of a gap in the the academic literature as well as poke at that um kind of more commercial expectations of AI MPCs this is far from that kind of throw some AI NPCs into a Sims like game or into a massive open world game this is instead you know a really creative highly original very new form of engaging with kind of traditional modes of Storytelling so we're all familiar with murder mysteries here's a very new way of of potentially designing one where you're in the seat of an interrogator in this case and you can ask quite literally anything and the system will respond in any um completely tailored to to those inputs um so what we've sort noted in the academic research is that a lot of people have been putting again chat gbt and building their own sort of tailored AI systems and putting them into existing games but I don't think that's really the future of the industry there are huge problems there as people like Tommy Thompson have have drawn attention to so really I think a lot of the creative potential sits at this sort of intersection of Highly creative and original human Le ideas that perhaps use generative AI systems to power some elements or some aspects of that uh and if that does end up being a kind of new space that Indie is work in again other Indies are kind of going the opposite way and having are developing currently um sort of markers to say this no generative AI was used in the creation of this game so I know there sort of there's two way ways this this is sitting but for those Indies that are interested in working in that space creatively and Co in a co-creative way then what we wanted to do was provide a research base to show the actual impact this might have on players uh and so what we've done um is we've recruited over 100 participants to come to Bristol this February and March so this study is happening right as we speak um to test out dead meat we're using uh a completely mixed methods approach um again for those academics in the room this will make a lot more sense than perhaps those who are less familiar with this but basically what we're trying to do is capture as much as we can of the kind of gold standard measures in human communter interaction and psychology I'm working with a colleague Dr Chris bon in computer science at Bristol do this um so we want to really understand the people that we're putting through this study uh their personal tendency to immersion for instance um empathy their curiosity um and indeed other really important information such as gender which we had a hypothesis would have a role to play here um and then we um have players play dead me we individually observe them we get them to think aloud as they play to sort of tell us what their mental model of the game is as it develops and then we follow up with kind of surveys that are the the most rigorous and and highly tested surveys on on on on games generally to apply them to this kind of new genre that's emerging along with a semi-structured interview can't really give yeah so I'm just wrapping up there um can't give the results of this stage obviously it's all in process but like just to sort of give a very high level gender we are the results are loading on gender which is really interesting players respond and approach these games in a completely unique and individual way which I think is fascinating um and I look for to be able to share the results of that as we write them up um and a final note sorry there was one more slide um this is relevant though because Sarah and I are working on a a project together as well that's outside of all of this that's about broadening access to AI tools to games and Exile storytelling and this is about bringing together early stage creatives um with academics working in this space to rigorously and critically evaluate the ethics the implications and the ways of potentially integrating and some of these tools that have been released onto the market or or turning um you sort of resisting them in certain ways um in the process of creating new games and XR stories so we're really sort of very much interested in in in seeing this the scope of this work um develop thank you and apologies for running over thanks Richard I thought it was really interesting we've talked before a lot about actually as these tools become more sophisticated they actually become less creative in the ways that the sort of responses you get which is certainly something we've experienced um and I think they're at the point now where they look so accurate but are so inaccurate and that's really problematic um and yeah really interesting perspectives about actually what these tools mean for people entering the industry and that sort of career development and skills Gap that they may create in um the future years so we're very short of time there have been some really interesting questions about how can we make sure we can use ethical Frameworks practical steps to tackle the issues of bias um and around you know experimenting with these tools in a way that's ethical and so I'm going to ask you the three of you to end um with a sort of two-part question I'm constructing um the first that I'd like you to address is what are the practical steps that people thinking of using these tools can take to ensure they're using them ethically or to to minim the bias um and then secondly in an Ideal World and if you're in charge how would AI be used in games if at all so um we'll go backwards so starting with you Richard and then move to Joseph and finish with inen okay wow um I'm gonna try and break down some kind of meaningful answer to those two really really really good questions um practical steps this is what we've been trying to explore with the AI tools and XL storytelling project and you can we come up with some kind of toolkit community of practice that can continually query and test and work with the assumptions uh and the limitations of each of these systems and talk and work out together you where where a Way Forward might be or um prototype together that's that's sort of what we were able to do with that project so I think and that was just a small example of a project that I know other people are doing other things in that space so I think that's going to be one One Way Forward um I think there will also be more directives coming from sort of Industry like I said there's there's um the company that developed mithre recently coming up with a a kind of a sort of um what the word is but it's a a sort of stamp of approval that this hasn't been you worked no AI has been sort of used in the process of making this um so I think there will be depending on whether you want to use these tools or Not There Will kind of ways forward there'll be guidance and practical kind of um yeah guidance that comes out I think there as well as um definitely a kind of push back against it and then the second question was around um how would AI be used in games or it's a big big question I'm really I'm I'm a definitely a critical Optimist in this face I think that the work the meting machine are doing but it's not just them there's lots of other interesting examples and demos coming out kind of right now um that I think test possibilities the human creative possibilities enabled by these tools again sort of challenging almost the fact that these tools are being built to us is that we need more data to make something more intelligent or more creative yeah I think the opposite is is true um it's about how we can you know be more creative with the kind of the smaller small language models is what they're calling them bringing models onto device training them you within um uh a developer an industry developer find a way of of customizing your own data and using a rigorous kind of approach and methodology there you know I think there's so many ways that we'll see this side of things develop in exciting ways that moves us away from the kind of big Tech uh off the peg apis uh and kind of you models that I think will actually encounter a lot of legal challenges going forward so I think I'm I'm looking forward to seeing what what more emerges in that space and if we can do good good research there um to sort of understand how it's changing and shaping player behavior and what kind of opportunities that opens up thanks Richard and same to you Joseph how to use these tools in Practical ethical ways and how would you see AI used in games Okay um I did I did enjoy the the human talker like genre of games and I think of the like the vampire game suck up um where you like the LM is not really part of the core mechanic it's just like this fun bit and I actually genuinely enjoyed that and I feel like okay there's there there is something interesting when it's not forced into existing structures and game mechanics that I I do find very interesting um I think uh there's there's there's a lot to unpack in my answer and I'm still thinking and processing um I don't think um large language models can ever be ethical um because uh to to achieve the higher order reasoning that they demonstrate has not been achieved without vast vast uh sets of data including harmful data um every attempt to try and alleviate this is produce produce models which don't have that higher reasoning ability where they can take external uh unseen circumstances and reason about them um so it's very hard to imagine how llms could ever be ethical like that um I think if we're going to look for suggest questions again Dr timnit has talked about small models um very specific like Niche slices not trying to create the one model to solve them all um she GES some great examples around different languages um uh lots of different very specific Niche trained languages which outperform um things like chat GPT when targeted on a very specific language um again the fact that American content dominates CH gpt's training set um means that it pretty much can

2025-02-26 12:05

Other news