A.I. | The Future and ESO

StihlReign

I've often considered the benefit to ESO that AI could offer. Here's a small wishlist of items I believe could be transformative and offer years more enjoyment, immersion, and an overall better experience for the user and developers.

How does AI improve gaming?
AI can analyze player data to uncover patterns, preferences, and pain points. This information can improve design decisions, helping developers create more engaging and enjoyable gameplay experiences. AI-driven analytics tools can provide valuable insights into player demographics, playtime, and in-game behaviors. Oct 19, 2023 (Source)

How will artificial intelligence revolutionize the way video games are developed and played?
...Exhaustively playtesting complex games requires massive human effort. AI simulation tools powered by Machine Learning algorithms can play through games far faster than humans while accurately modeling human behavior. This makes it possible to extensively test games in mere days rather than weeks or months. Jan 27, 2024 (Source)

How is AI propelling the gaming industry into a new epoch?
In-game AI enables dynamic and adaptive environments, offering players personalized challenges and experiences tailored to their unique interactions. AI also plays a pivotal role in character behavior, generating lifelike opponents with human-like decision-making capabilities. (Source)


1. I'd love the see AI solve the challenge of PvP server performance and limited population caps. ESO is an awesome PvP | PvE game, one feeds the other. I continue to consider ESO a PvP game fed by an amazing PvE adventure, with Vet Trials and PvP the true endgame experience. I'd like to see the cities full and bustling and have the population increases 'feel like' less of an ESO event phenomenon. I'd also like to see less loading as we transition to the next zone, and have a more seamless experience (speed is not the issue here - most zones load fast).

[AI can also dynamically adjust in-game resource allocation on the fly by performing real-time performance analyses and delivering resources to game elements as needed. This load-balancing act means games utilize available computing power in the most efficient way at all times for optimal operation.] 1

2. Solve the challenge of dynamic interaction with PvE enemies. More immersion and intelligent dynamic interaction as a player travels throughout the world based on each character. Deep learning - the game knows where you're at in your character development, solo or with friends, and adjusts the enemy NPC accordingly - level 39 with a 1600 CP friend (grouped) and questing -encounters a WB or even and Alit, halfway through the fight a 2800 healer jumps in (non-grouped) - enemy NPC adjusts and so do the rewards.

It would be awesome to see similar logic applied to trials, and Vet trials in particular - where you might have a list of abilities the boss has available but the fight is much more dynamic and adaptive. The end result being great fights but fewer youtube videos of how to beat the boss, and more videos of "we beat the boss This Fight Was Epic."

3. Remove the Battle Spirit limitation from the game to create a more seamless experience between the armor types chosen and utilized, for PvP | PvE gameplay, by adjusting the minimum underlying base stats for player characters and NPCs. AI offers an incredible amount of computation/simulation/testing capability. This issue is mostly math and probably more easily solved than not, by running the simulations and utilizing item Number 2 above.

There are a ton of other items AI will help solve and address in gaming not included but considered, for a comprehensive list - AI IN GAMING: TRANSFORMING GAMING WORLDS WITH INTELLIGENT INNOVATIONS

EDITED | Titles for clarity

danno8

The pessimistic (or perhaps realistic) side of me tells me that LLMs (I dislike the term AI for nearly all that the term is currently being used for) will be used to primarily drive down costs for companies by replacing as many employees as possible with inferior LLM systems while hoping most people don't notice or care enough to notice.

DinoZavr

LLMs already can solve "write me 100 AD&D quests. each 1000 words minimum"
There are good models (like) ElevenLabs for TTS (text-to-speach)
Well-trained Stable Diffusion alike models can generate characters look and interiors/buildings/landscapes/biomes
so it is more likely, A.I. to help/replace writers/sound actors/art designers
as for "clever" NPCs - this is highly unlikely, considering servers load and VRAM prices
Gina told old servers worked for 10 years before upgraded recently.
players trends analysis can be done relatively inexpensive, the question is how much this could increase profits?

M1SHAAN

AI is a useful and exciting developing technology, but it is not a panacea. There is a trend of companies implementing "AI solutions" to their problems because "AI is the future" when in reality the AI technology they have selected because it is new and flashy is a poor fit to their problem and will take more resources to produce a worse solution than a simpler method. I hope that ZOS will consider the strengths and limitations of any AI they are considering implementing rather than getting swept up in the hype wave.

I have a few miscellaneous thoughts on some of your points in particular.
Re: AI in analytics: data analysts working with large amounts of data already make use of various machine learning techniques such as k-means clustering. I'm sure the analysts are already actively looking into potential applications of more sophisticated deep learning techniques, if not implementing them already.
Re: AI in QA: I share @danno8's concern that companies would try to use AI tools to replace QA engineers rather than to assist them. I am uneducated in the videogame QA process but I expect existing AI tools need human direction to be effective and may not be useful for problems that can't be brute forced.
Re: PVP server performance: dynamically allocating resources doesn't necessarily require AI sophistication, and I suspect that if resource allocation would solve the Cyro performance problems they would already no longer be an issue.
Re: dynamically scaling enemy difficulty: I don't think this would require an AI level of complexity, it'd be a simple formula with the number of engaged players and their levels (and maybe their roles) as inputs. Also, personally, I'd hate it. An enemy I'm fighting gets stronger or weaker in the middle of the fight if some rando comes along or leaves? It doesn't fit existing videogame logic and I think it would be frustrating rather than rewarding.

Anyway those are my opinions, take them or leave them, I'm just some rando on the internet.

Alpheu5

StihlReign wrote: »

It would be awesome to see similar logic applied to trials, and Vet trials in particular - where you might have a list of abilities the boss has available but the fight is much more dynamic and adaptive. The end result being great fights but fewer youtube videos of how to beat the boss, and more videos of "we beat the boss This Fight Was Epic."

I think we'd end up with the same balance problem faced in some BG3 encounters, that being a DM with a priority of annihilating the group over providing a fun experience. It would be very frustrating to fight a boss in a trial that constantly spams group-wipe mechanics because the AI learned that it was the most effective way to "win". Sure, they could balance it by forcing certain abilities to be on cooldown and restrict which ones can be used sequentially, but then they may as well just do what is already done by putting mechanics on rotations and tying others to health thresholds.

Bo0137

I strongly believe videogames can benefit from AI way more than other areas - specially because of [less] moral dilemmas.

I am absolutely against laying off real people to implement AI.

However, I understand that videogames have different iteractions. For example: NPC conversation: could very much benefit from AI text geneartion and does not mean a real person wouldn't be needed to write some imputs; level/difficulty scaling; bug hunt/fixing; many others

With that being said, I really believe ESO won't take benefit from AI. Simply because it was not built around it from the beginning. It might be too hard to implement at this point. And the dev team might have another mindset approach to using AI, due to the past 10 years of experience.

At the same time, ES6 could and should implement AI to many aspects of the game.

StihlReign

AI is well beyond hype and is a current reality of how companies will operate in all of the areas mentioned, particularly in the gaming space.

This list is simply how I'd like to see "what will be implemented in general", focused and done in ESO. Most of what I've outlined will be announced in the MMO and gaming space within the next 5 years as some standard of operations going forward.

AI will continue to eliminate and replace current models without AI algorithms as cost-ineffective. Machine learning algorithms and simulations are simply way too powerful and possess incredible speed and efficiency compared to models and processes without them.

AI allows the end user decision maker (humans) the ability to make better decisions, faster.


Consider:

#1 AI models use fewer rules
Rules need constant updating whenever a new behavior is detected. Algorithms seek and compare data based on a wide range of options.

Benefit: Fewer rules result in less maintenance
Reduction in Rules using AI 50,000 -> 12

#2 AI models are not constrained by hard-coded rules
Rules are hardcoded and depend on certain conditions. Because rules take a binary approach – if this, then that – they depend on the information provided. This leaves room for human error and bias.

Benefits: Models can analyze multiple data points and detect anomalies
Anomalie detection using AI +200%

#3 Continuous, automated refinement
Rules are static, which is why they need constant manual updating. Your information lives in silos.

Because AI is self-learning, models can use an automatic feedback loop for constant improvement

Benefit: Models learn and adjust automatically based on experience
Reduction in False alerts due to AI 8300 -> 300

#4 Scalable AI outperforms other solutions
Data in rules-based decisioning solutions have to run through thousands, even hundreds of thousands, of rules.

Unsophisticated AI can use such complex algorithms that it takes hours or even days to run.

This timeframe makes it very difficult to stop fraud in progress as it bogs down when the volume of data increases.

To improve processing time, some rules may be turned off but that comes with risk. If you delete or turn off a rule in a rules-based system, it won’t detect a fraud scheme that resurfaces in six months. That results in a drain on resources. Staff have to spend time investigating the fraud or increased risk, while having to comb through and decide which rules to reinstate. Meanwhile, the company needlessly loses revenue.

“A significant amount of damage can be done..."
AI will not only detect the behavior, but also learn any new twists to schemes.


"...distributed architecture allows lightning speed response times (less than 10 milliseconds), end-to-end encryption and traceability. Our AI models make decisions at up to 100,000 events per second..."

Benefit: ...customers benefited from 99.9999% uptime with limitless scalability.


How do AI models compare to rules-based decisioning? Because AI’s machine learning makes multifactored decisions, AI returns higher accuracy rates, and reduces false positives and manual reviews.

SOURCE | Please review for context

StihlReign

Bo0137 wrote: »

With that being said, I really believe ESO won't take benefit from AI. Simply because it was not built around it from the beginning. It might be too hard to implement at this point. And the dev team might have another mindset approach to using AI, due to the past 10 years of experience.

How does AI impact the modern gaming arena?
Click the source link to view a detailed breakout. Gaming companies are and will be transitioning, ESO will be heavily impacted. For example - consider the anti-cheat systems.

The influence of AI in the gaming industry has witnessed an unprecedented surge, reshaping the landscape of interactive entertainment. AI algorithms have transcended traditional gaming boundaries, enhancing player experiences and the game development process.

Applications of AI in gaming
Non-player Character (NPC) behavior
Enemy AI
Pathfinding and navigation
Procedural content generation
Adaptive difficulty levels
Graphics enhancement
Voice recognition and Natural Language Processing
Game design assistance
Quality assurance and testing
Anti-cheat systems
Dynamic game environments
Personalized content delivery
Player-Experience Modeling (PEM)
Data-mining and real-time analytics
Player sentiment analysis
Virtual assistants

SOURCE

Tommy_The_Gun

The thing I find funny is that AI is nothing new. It is just a new term (or "slogan" or "buzz word") to describe basically an algorithm or program. For instance, I remember that even in very old graphic editors like Gimp or PaintShop there were various tools to generate stuff. For example - you could generate clouds. You just typed the amount & type of clouds and the program would generate a texture that is a blue sky with clouds on it. Amazing right ? Today we would call it that it is "AI generated", but the technology existed decades before. Similar example is TES IV Oblivion or Daggerfall. Those games had loads of areias that were "procedurally generated" - meaning that they were generated by the algorithm / program. Again - today we would call it AI generated, just because the fashion has changed and AI is the "buzz word" of the current decade.

The biggest difference though is computing power that we got now vs what we got in the past. So I guess that it is maybe why the new term (AI) is being used ? Idk.

Anyway, the biggest benefit from video game perspective that I see, is that it could take far less time to download a game, because files (like textures & audio) can be "generated" locally instead of being downloaded. Think of it. Imagine that instead of recoding voice for the main character, the game devs only record samples. Player then choses which samples they want their character to sound like, they chose personality tags and... boom. Algorithm (or AI) generates like 20 GB of audio data and then you have fully voiced character. It would also mean that you could have games that are way longer than now.

As far as I am aware, something like this is already happening in the modding scene as there are already some Mods that aim to make TES3 Morrowind for example to be fully voiced.

StihlReign

Alpheu5 wrote: »

StihlReign wrote: »

It would be awesome to see similar logic applied to trials, and Vet trials in particular - where you might have a list of abilities the boss has available but the fight is much more dynamic and adaptive. The end result being great fights but fewer youtube videos of how to beat the boss, and more videos of "we beat the boss This Fight Was Epic."

I think we'd end up with the same balance problem faced in some BG3 encounters, that being a DM with a priority of annihilating the group over providing a fun experience. It would be very frustrating to fight a boss in a trial that constantly spams group-wipe mechanics because the AI learned that it was the most effective way to "win". Sure, they could balance it by forcing certain abilities to be on cooldown and restrict which ones can be used sequentially, but then they may as well just do what is already done by putting mechanics on rotations and tying others to health thresholds.

AI can learn and be taught to provide a challenge and the dev can decide the thresholds that determine what winning is. The AI ability to run countless simulations and learn (know) what you consider fun would be interesting.

M1SHAAN

StihlReign wrote: »

SOURCE | Please review for context

This source, Brighterion (and the one for your next comment, LeewayHertz) is a company that sells AI solutions to other companies. In that context, I am highly skeptical of their claims.

AI is exciting and can do better than traditional technologies, but it also comes with new downsides. It is not a panacea. It is not immune to human bias.

I may be back with further thoughts later.

Stamicka

Honestly this feels like a lot of mumbo jumbo. A lot of these AI "improvements" wouldn't be improvements at all. For example in regard to testing while it may be possible for AI to find bugs in games at some point, there's still very important subjective aspects to the game testing experience that will still require humans. For example, is the tester engaged, do they feel like their time is respected, are they having an easy time navigating menus? The game is made for humans after all, so humans will still need to be involved in the testing.

In regards to dynamic enemies, that's something I am actually against. Having somewhat predictable enemy behavior allows you to make a strategy for the fight. We as the players adjust for the enemy, the enemy shouldn't adjust for us. When it comes to score pushing as well, there would be way more RNG involved if enemy behavior was dynamic.

I also don't really like the idea of AI generated NPC text. Story writers and creatives often spend time on dialogue to tell the story they want to tell, how they want to tell it. In good stories, word choices are intentional and meaningful. I can see the purpose of dynamic NPC text in other games that aren't story focused, but I don't think it has a place in ESO.

In regards to AI solving problems, that's optimistic. Find the most advanced AI you can find and ask it a question about a highly technical subject that you know a lot about. Chances are the AI will just lie to you and make up false information. It excels when there's a ton of resources on the internet for what you ask it, but it does not do well when there's limited information on the internet. To actually see AI solve a problem that isn't already somewhere on the internet (like the extremely complex issue of ESO lag), I think that is years and years beyond what is currently possible.

Maybe AI has a place for generating models and stuff of that nature, but even then I would be hesitant to use it. There's a certain soulessness to AI generated art. It's often missing character and style, it's just boring. Keep the artists please.

StihlReign

Tommy_The_Gun wrote: »

The thing I find funny is that AI is nothing new.

Yep. Machine learning has been an integral part of gaming since the early 80's. Pacman is an example often used.

Alpheu5

Stamicka wrote: »

In regards to dynamic enemies, that's something I am actually against. Having somewhat predictable enemy behavior allows you to make a strategy for the fight. We as the players adjust for the enemy, the enemy shouldn't adjust for us. When it comes to score pushing as well, there would be way more RNG involved if enemy behavior was dynamic.

We can look at the current iteration of the Archive to see this in play already. The predictability of trials is what makes them good leaderboard content. Changing how an encounter progresses from instance to instance makes it very poor for leaderboard content. Until IA makes veteran a seeded instance for that week's leaderboard, where everyone can choose the same verses and visions, and the same marauders spawn on the same stages, success is too heavily reliant on randomness for the leaderboard to mean much if anything.

StihlReign

My 3090 and 4090 use AI right now, in real-time. This is just one of a myriad of ways AI is transforming gaming.

The Ultimate in Ray Tracing and AI.
NVIDIA RTX™ is the most advanced platform for ray tracing and AI technologies that are revolutionizing the ways we play and create. Over 500 top games and applications use RTX to deliver realistic graphics, incredibly fast performance, and new cutting-edge AI features like NVIDIA DLSS 3.5 with Ray Reconstruction. RTX. It’s On. SOURCE

AI-powered innovations in graphics: Transforming gaming realism and aesthetics
Real-time ray tracing
AI technologies have spearheaded advancements in graphics with real-time ray tracing, powered by hardware acceleration like NVIDIA’s RTX series GPUs and software frameworks such as Microsoft’s DirectX Raytracing (DXR) and Vulkan Ray Tracing. In real-time ray tracing, AI algorithms accelerate the calculation of rays of light, simulating complex interactions with in-game objects. Ray tracing denoising, implemented through deep learning technologies like NVIDIA’s DLSS (Deep Learning Super Sampling), plays a crucial role in improving ray tracing performance by reducing noise and enhancing visual fidelity. The combination of powerful hardware and AI-driven denoising results in stunningly realistic lighting, reflections, and shadows, profoundly impacting the visual quality and immersion in games.

AI-driven rendering techniques
AI-driven rendering techniques optimize graphical performance and quality, incorporating technologies like machine learning-based super-resolution. NVIDIA’s DLSS is a prime example, utilizing deep neural networks to upscale lower-resolution images to higher resolutions in real time, effectively improving image quality without the computational cost. Additionally, AI-based anti-aliasing techniques, such as Temporal Anti-Aliasing (TAA), enhance visual clarity by reducing jagged edges and smoothing out graphics. These technologies collectively contribute to a smoother and more visually appealing gaming experience by dynamically optimizing rendering processes based on AI-driven algorithms

This list is simply how I'd like to see "what will be implemented in general", focused and done in ESO. Most of what I've outlined will either be announced, implemented or is already underway in the MMO and gaming space or will be within the next 5 years as some standard of operations going forward.

RicAlmighty

As someone that worked heavily in ML the last 3+ years, what you've suggested would take a minimum of 3-5 years to implement if it could even be done in the first place. You cannot add "AI" to an existing game engine. The entire core would need to be re-written and the LLM constantly regenerated and updated to manage the learning. This is a massive undertaking.

Your first link is to a blog post talking about "AI" replacing human designers, so not sure how that is applicable. Your second link is just a mind map of things that "could" be done with no specifics or direct applications.

Thirdly, the "literature" you are quoting is from a company that has a financial interest in developers adopting these technologies and thusly the marketing rhetoric borders on nonsensical. Nvidia's upscaling tech like DLSS is completely different than what you are referring to in your first post and should not be used as a point of comparison. Nvidia likes to use the term "AI" when they just mean branching algorithms because "AI" makes their stock price go up. Their tech is great, don't get me wrong, but it's not "AI", it's LLM based on frame data that upscales (or interpolates in the case of DLSS Frame Gen) in real-time. "AI" cannot alter game logic in real-time unless the very core of the game engine was written with that as an intrinsic part of its inner workings. Adding some of the things you've suggested to ESO would be more time consuming than Zos would deem reasonable. Not to even mention that a console or a PC would be wholly unable to perform the algorithmic complexity necessary to run the logic locally. So then you'd have to rely on performing all of these decisions on the server cluster which would add an unacceptable amount of latency to the gameplay.

In short, while the points you've raised can be valid areas of interest moving forward for new development, none of them would be able to be easily added to a game like ESO.

Stamicka

RicAlmighty wrote: »

As someone that worked heavily in ML the last 3+ years, what you've suggested would take a minimum of 3-5 years to implement if it could even be done in the first place. You cannot add "AI" to an existing game engine. The entire core would need to be re-written and the LLM constantly regenerated and updated to manage the learning. This is a massive undertaking.

Your first link is to a blog post talking about "AI" replacing human designers, so not sure how that is applicable. Your second link is just a mind map of things that "could" be done with no specifics or direct applications.

Thirdly, the "literature" you are quoting is from a company that has a financial interest in developers adopting these technologies and thusly the marketing rhetoric borders on nonsensical. Nvidia's upscaling tech like DLSS is completely different than what you are referring to in your first post and should not be used as a point of comparison. Nvidia likes to use the term "AI" when they just mean branching algorithms because "AI" makes their stock price go up. Their tech is great, don't get me wrong, but it's not "AI", it's LLM based on frame data that upscales (or interpolates in the case of DLSS Frame Gen) in real-time. "AI" cannot alter game logic in real-time unless the very core of the game engine was written with that as an intrinsic part of its inner workings. Adding some of the things you've suggested to ESO would be more time consuming than Zos would deem reasonable. Not to even mention that a console or a PC would be wholly unable to perform the algorithmic complexity necessary to run the logic locally. So then you'd have to rely on performing all of these decisions on the server cluster which would add an unacceptable amount of latency to the gameplay.

In short, while the points you've raised can be valid areas of interest moving forward for new development, none of them would be able to be easily added to a game like ESO.

Agreed, there's a lot of nonsense in this post. The tech world has these big trends every few years and people just throw around buzzwords and try to hop on the hype train. AI can do impressive things and it can speed up development time. However, a lot of things that aren't necessarily "AI" use the word AI as a buzzword and to get more attention. There's a lot of claims being made of what AI is capable of currently and it's usually untrue or a very stretched truth.

Remember when cryptocurrency was new? Blockchain? NFTs? Web3? 5G? All of those are example of technologies that were supposed to revolutionize the world, but look where we are. Yes all of those technologies have had impacts on the world, some good and some bad, but the hype around what they would be is completely different than how it turned out.

StihlReign

RicAlmighty wrote: »

As someone that worked heavily in ML the last 3+ years, what you've suggested would take a minimum of 3-5 years to implement if it could even be done in the first place. You cannot add "AI" to an existing game engine. The entire core would need to be re-written and the LLM constantly regenerated and updated to manage the learning. This is a massive undertaking.

Predictive analysis has many applications, I'm uncertain why you're discussing the existing game engine, and time is something developers have and use as they see fit during product development. AI (machine learning or any series of acronyms that helps you better understand - feel free to use) is already in use in gaming and has been for 40+ years.

The link is provided to give Source credit to the direct quote. Your suggestion that a consulting company with billion-dollar clients matches (or is beneath) your 3 years of recent experience is...interesting.

AI will have deepening and further involvement in almost every aspect of gaming listed simply because this is where we are.

Applications of AI in gaming
Non-player Character (NPC) behavior
Enemy AI
Pathfinding and navigation
Procedural content generation
Adaptive difficulty levels
Graphics enhancement
Voice recognition and Natural Language Processing
Game design assistance
Quality assurance and testing
Anti-cheat systems
Dynamic game environments
Personalized content delivery
Player-Experience Modeling (PEM)
Data-mining and real-time analytics
Player sentiment analysis
Virtual assistants

The machine is smarter and faster in the areas of analysis where we need it most - now. This will only increase.

StihlReign

danno8 wrote: »

The pessimistic (or perhaps realistic) side of me tells me that LLMs (I dislike the term AI for nearly all that the term is currently being used for) will be used to primarily drive down costs for companies by replacing as many employees as possible with inferior LLM systems while hoping most people don't notice or care enough to notice.

I hope that isn't the case, there's a lot of talent, and the primary purpose of a machine is usually to make human tasks easier...

Danikat

There's a lot of hype around LLMs and other machine learning products that get called AI at the moment, although a lot of it (including the sources for this post apparently) can be very quickly traced back to the people trying to sell it as a service by dramatically over promising what it can do and hoping the fact that it's cheap will balance out the mediocre results so I'm not sure it's even accurate to call it hype.

In most cases I don't think the suggested uses offer anything beneficial (for companies or end-users), especially in situations like the ones the OP proposed. Replacing actual playtesters with an algorithm which attempts to mimic human behaviour is a terrible idea. The benefit of playtesters is that they aren't a uniform mass - because they're people they're unpredictable and do things the developers didn't expect, which allows them to plan for when players will do the same things after release. Machine learning algorithms can also be unpredictable if they're allowed to run long enough, but they're unpredictable in wildly different ways from humans, which can be interesting in some situations but would be a huge deteriment in this case.

Having said all that Bethesda has already made one of the best games to use machine learning algorithms (although the Creatures series is still the best) and I'd love to see them try it again. Yes I mean Oblivion. Yes I know it was often a mess and the "AI" causes all kinds of weird problems. In one playthrough I was unable to buy the house in Skingrad because the butler inexplicably decided to leave the castle one day and got killed by goblins (who also shouldn't have been in that area). But the unpredictable NPC behaviour (and the fact that it often did have a weird kind of logic behind it) was also one of the most interesting things about the game and IMO made it more interesting to replay because it was never going to be exactly the same twice.

I know they had to tone it down a lot due to the crazy things that happened during testing, but I'd love to see them make a game that's actually built around that so it's less of a problem when things go in unexpected directions and so the NPCs can be allowed more freedom.

Not in ESO though, that would be a terrible idea because the game isn't built for it.

Stamicka wrote: »

I also don't really like the idea of AI generated NPC text. Story writers and creatives often spend time on dialogue to tell the story they want to tell, how they want to tell it. In good stories, word choices are intentional and meaningful. I can see the purpose of dynamic NPC text in other games that aren't story focused, but I don't think it has a place in ESO.

I agree with this point especially. "AI" text generators are good at producing a lot of words quickly, they are not good at consistent narratives or matching the text samples they assemble from their database (which is all they're really doing) to background lore, even when they're just supposed to make it realistic.

I've seen AI generated NPC dialogue as an example of how AI could help games a lot and it always seems to work on the assumption that quantity is the only consideration and quality does not matter at all. Yes a chat bot embedded into a game could allow NPCs to endlessly spit out new dialogue no matter how many times players spoke to them. But if none of it is relevant to anything else what's the point?

If you're focused on exploring what the LLM can do it might be interesting that an NPC can, without being programmed to do so, tell you about how Castle Goronhast on the cliffs over looking the vale where the Night Elves live used to be ruled by the Lord of Shadows (who was rumoured to be a vampire) until his daughter overthrew him by forming an alliance with the hill trolls. But if none of those people or places is in the game and the next time you talk to them they're going on about the endlessly burning fire in the forest of Shorrenhai (which also isn't in the game) what's the point? What does it add for the player?

Especially if you're removing actual narrative created by writers based on the games lore and designed to tie into the larger narrative to make space for this borderline-randomly generated nonsense. Then it would actually detract from the game, making it worse than it could otherwise have been.

StihlReign

Danikat wrote: »

There's a lot of hype around LLMs and other machine learning products that get called AI at the moment, although a lot of it (including the sources for this post apparently) can be very quickly traced back to the people trying to sell it as a service by dramatically over promising what it can do and hoping the fact that it's cheap will balance out the mediocre results so I'm not sure it's even accurate to call it hype.

In most cases I don't think the suggested uses offer anything beneficial (for companies or end-users), especially in situations like the ones the OP proposed. Replacing actual playtesters with an algorithm which attempts to mimic human behaviour is a terrible idea. The benefit of playtesters is that they aren't a uniform mass - because they're people they're unpredictable and do things the developers didn't expect, which allows them to plan for when players will do the same things after release. Machine learning algorithms can also be unpredictable if they're allowed to run long enough, but they're unpredictable in wildly different ways from humans, which can be interesting in some situations but would be a huge deteriment in this case.

Humans are mostly predictable, it's why marketers and advertisers make so much money. Simulations can run millions of scenarios and answer tons of questions from inputs that used to take exponentially longer to try and discern.

I'm unsure why you'd replace a playtester - seems like an odd choice when you could simply hand them a better product to test.

sleepy_worm

I seem to remember QA testers having a union. They are most likely more resilient than others to the inevitable downsides of automation.

It would be a shame to see a game that took pride in using human voice actors for every single line of dialogue (a tremendous feat and true dedication to a quality experience) start using AI voices for NPCs.

Danikat

StihlReign wrote: »

Danikat wrote: »

There's a lot of hype around LLMs and other machine learning products that get called AI at the moment, although a lot of it (including the sources for this post apparently) can be very quickly traced back to the people trying to sell it as a service by dramatically over promising what it can do and hoping the fact that it's cheap will balance out the mediocre results so I'm not sure it's even accurate to call it hype.

In most cases I don't think the suggested uses offer anything beneficial (for companies or end-users), especially in situations like the ones the OP proposed. Replacing actual playtesters with an algorithm which attempts to mimic human behaviour is a terrible idea. The benefit of playtesters is that they aren't a uniform mass - because they're people they're unpredictable and do things the developers didn't expect, which allows them to plan for when players will do the same things after release. Machine learning algorithms can also be unpredictable if they're allowed to run long enough, but they're unpredictable in wildly different ways from humans, which can be interesting in some situations but would be a huge deteriment in this case.

Humans are mostly predictable, it's why marketers and advertisers make so much money. Simulations can run millions of scenarios and answer tons of questions from inputs that used to take exponentially longer to try and discern.

I'm unsure why you'd replace a playtester - seems like an odd choice when you could simply hand them a better product to test.

I don't know why you'd want to do that, you're the one who suggested it:

StihlReign wrote: »

I've often considered the benefit to ESO that AI could offer. Here's a small wishlist of items I believe could be transformative and offer years more enjoyment, immersion, and an overall better experience for the user and developers.

...

How will artificial intelligence revolutionize the way video games are developed and played?
...Exhaustively playtesting complex games requires massive human effort. AI simulation tools powered by Machine Learning algorithms can play through games far faster than humans while accurately modeling human behavior. This makes it possible to extensively test games in mere days rather than weeks or months. Jan 27, 2024 (Source)

Cazador

Ok I'm curious OP, are you simply interested in what AI can do or do you work for one of these companies that market it? This thread honestly just feels like a sales pitch.

M1SHAAN

StihlReign wrote: »

I'm uncertain why you're discussing the existing game engine

They are discussing the existing game engine because many of your suggestions would require modifying the engine.

StihlReign wrote: »

time is something developers have and use as they see fit during product development.

I disagree. Development takes time, and for a game like ESO that is expected to have regular updates the developers have precious little of it. They are not going to sink years of their time into rebuilding the engine to incorporate AI elements.

StihlReign wrote: »

Your suggestion that a consulting company with billion-dollar clients matches (or is beneath) your 3 years of recent experience is...interesting.

You misunderstand. @RicAlmighty is not suggesting that they have more ML experience than all the developers at the company combined. They are suggesting that the webpage of a company whose entire reason for existence is to sell other companies on the idea that their AI products will allow those companies to make more money is not a place to get a level headed discussion of the potential merits and drawbacks of a particular AI technology. I repeat, The entire reason that webpage exists is to sell the idea that AI is valuable. The issue isn't that the company lacks knowledge, but that it has an ulterior motive in sharing particular information or ideas. I'm sure the engineers at the company are keenly aware of potential drawbacks to their systems, but they aren't going to be shouting that information from the rooftops. Folks like @RicAlmighty and I don't have intimate knowledge of this company's product, but we know enough about the field as a whole to differentiate between useful information and fluffy marketing-speak.

The webpage is marketing, plain and simple. Its intent is to make people excited about the possibilities of AI. Nuance, qualifications, drawbacks... you won't find them there. If you're interested in the latest advancements in the field check out some research papers, a lot of authors publish their papers on sites like arxiv so they can be accessed for free. It still won't give you an unbiased picture because poor results are less likely to get published, but it's better than corporate fluff at least.

Oh, and also:

StihlReign wrote: »

Humans are mostly predictable, it's why marketers and advertisers make so much money.

This is somewhat true on larger scales, but accurately predicting the actions of an individual is as far as I know still beyond us. People are complex.

StihlReign wrote: »

Simulations can run millions of scenarios and answer tons of questions from inputs that used to take exponentially longer to try and discern.

Nitpick, but simulations are not the same thing as AI, I work with both. Simulations can for example just be the execution of a static algorithm with no AI elements whatsoever.

StihlReign wrote: »

I'm unsure why you'd replace a playtester - seems like an odd choice when you could simply hand them a better product to test.

It's an easy choice when employing a playtester costs money and an algorithm does a "good enough" job. Quality means nothing in the eyes of a company which just wants to make money.

TaSheen

@M1SHAAN - if I could give you a million awesomes, I would. Unfortunately - one is the limit here.

WiseSky

I am currently playing a game, Ravendawn, that has AI voices for quests, maybe some of those quests we written by LLM also. I can't tell if they used it on writing or not.

I think we are many many years from a GOTY made by LLM and AI.

Last GOTY was BG3 which had so much effort put into by the devs.

Tale old as time: Quality VS Quantity, as much as corps want to get away with spending less money on products they will dabble with LLM and AI but in the end who knows... maybe one day we can solve for creativity... let's just hope we dont go the dystopian future where AI makes art and human do the hard labour, we all would like it the other way around.

TaSheen

WiseSky wrote: »

I am currently playing a game, Ravendawn, that has AI voices for quests, maybe some of those quests we written by LLM also. I can't tell if they used it on writing or not.

I think we are many many years from a GOTY made by LLM and AI.

Last GOTY was BG3 which had so much effort put into by the devs.

Tale old as time: Quality VS Quantity, as much as corps want to get away with spending less money on products they will dabble with LLM and AI but in the end who knows... maybe one day we can solve for creativity... let's just hope we dont go the dystopian future where AI makes art and human do the hard labour, we all would like it the other way around.

Interesting. I... feel the same way, though I didn't put it into words....

chessalavakia_ESO

I think much of the "AI" in gaming discussions miss that "AI" may very well end of wrecking significant sections of gaming.

What happens to multiplayer games when someone produces a device that watches your screen and then based on what it sees interacts with your mouse and keyboard to play the game?

What happens to singleplayer games when "AI" can produce tens of thousands of hours of content?

What happens to the social interactions around gaming when the sites are totally filled with "AI"?

How will players being able to more easily "create" their own games with the help of "AI" impact the gaming marketplace?

With that said, I think a big part of the reason "AI" has so much potential to threaten entertainment is because much of the success in today's entertainment is execution, social media, and marketing not creativity and the content creators reflect that.

runa_gate

Imagine if AI could argue in zone chat for all the people who no one will talk to IRL so they need a captive audience! They could act out their inability to achieve parental approval without bothering with typing their misspelled almost-sentences

pardon me... I was doing a survey in Auridon a little bit ago but I'm doing better now

StihlReign

Danikat wrote: »

I don't know why you'd want to do that, you're the one who suggested it:

No. I didn't.

M1SHAAN wrote: »

StihlReign wrote: »

I'm uncertain why you're discussing the existing game engine

They are discussing the existing game engine because many of your suggestions would require modifying the engine.

StihlReign wrote: »

time is something developers have and use as they see fit during product development.

I disagree. Development takes time, and for a game like ESO that is expected to have regular updates the developers have precious little of it. They are not going to sink years of their time into rebuilding the engine to incorporate AI elements.

StihlReign wrote: »

Your suggestion that a consulting company with billion-dollar clients matches (or is beneath) your 3 years of recent experience is...interesting.

You misunderstand. @RicAlmighty is not suggesting that they have more ML experience than all the developers at the company combined. They are suggesting that the webpage of a company whose entire reason for existence is to sell other companies on the idea that their AI products will allow those companies to make more money is not a place to get a level headed discussion of the potential merits and drawbacks of a particular AI technology. I repeat, The entire reason that webpage exists is to sell the idea that AI is valuable. The issue isn't that the company lacks knowledge, but that it has an ulterior motive in sharing particular information or ideas. I'm sure the engineers at the company are keenly aware of potential drawbacks to their systems, but they aren't going to be shouting that information from the rooftops. Folks like @RicAlmighty and I don't have intimate knowledge of this company's product, but we know enough about the field as a whole to differentiate between useful information and fluffy marketing-speak.

The webpage is marketing, plain and simple. Its intent is to make people excited about the possibilities of AI. Nuance, qualifications, drawbacks... you won't find them there. If you're interested in the latest advancements in the field check out some research papers, a lot of authors publish their papers on sites like arxiv so they can be accessed for free. It still won't give you an unbiased picture because poor results are less likely to get published, but it's better than corporate fluff at least.

Oh, and also:

StihlReign wrote: »

Humans are mostly predictable, it's why marketers and advertisers make so much money.

This is somewhat true on larger scales, but accurately predicting the actions of an individual is as far as I know still beyond us. People are complex.

StihlReign wrote: »

Simulations can run millions of scenarios and answer tons of questions from inputs that used to take exponentially longer to try and discern.

Nitpick, but simulations are not the same thing as AI, I work with both. Simulations can for example just be the execution of a static algorithm with no AI elements whatsoever.

StihlReign wrote: »

I'm unsure why you'd replace a playtester - seems like an odd choice when you could simply hand them a better product to test.

It's an easy choice when employing a playtester costs money and an algorithm does a "good enough" job. Quality means nothing in the eyes of a company which just wants to make money.

Future indicates future. I'm assuming with ESO's success the engine will be rebuilt at some point. Sooner than later...it is aging.

Simulations fit directly in the machine learning box.

Business owners understand quality is of critical importance if they want to make money.

Accurately predicting humans in the environment we're discussing is complex but not that complex. Most of the decisions are fairly linear and limited. Marketing, advertising, and search are just one area where these tools are making astounding progress (to the tune of trillions at last check).