AI Will Make Better Games Than Me One Day

the writings on the wall AI will make full-on video games all by itself I know a lot of you think that either this is not going to be possible or it's not going to be desirable this one I'll just prove wrong and this one we'll talk about as well in this video to give you a little bit of context I'm an indie game developer I made three fairly successful indie games so far I love creating art I love being creative and I would very much like to keep doing this forever however unfortunately I have two eyeballs I can see what's going on and we really have to talk about this video will be a deep dive on the subject I'll put the puzzle pieces together for you and explain why AI generated games are much closer than a lot of you may think we'll also talk about what it means for game developers how to cope with it and why I'm still fairly optimistic about the future about 60% of my audience think that AI will never be able to make any good games so it's important to note that opinions on this differ a lot and there's a very good chance I might be wrong on this disclaimer this is a biased opinion piece by a biased and flawed human being it may contain incorrect of FL predictions this video is meant to broaden your Horizon and not to dictate what to think about the subject so please think for yourself and

keep the discussion in the comments civilized once again start with the very Basics just to make sure that everybody's on the same page so please have a little bit of patience as I put these puzzle pieces together for you I'll do it as quickly as I can if you expect another video about AI ethics I got to set your expectations straight this is not that video this video focuses on predictions on how this technology is going to affect the future of indie game development if that interests you then I think the following hour will be well worth your time and I hope you enjoy let's do a very quick reality check of where we're at and why I think that the trend lines are fairly clear this is gp4 at the time of recording this probably the best text generator out there but a simple grid based game in P game that use a cellular autometer and has both a win and a lose condition let's see what it spits out apparently it came up with the win condition if the pattern forms a stable pattern or repeats then you win and if all cells die you lose let's run it oh you win let's ask a new instance of GPT to fix the code your game has a logic flaw in the win lose condition copy that and we have way game of life it's

unfortunately not interactive make it so I can paint cells with my mouse and now we can paint cells come on stabilize stabilize and just ah you in woo we won with just three human interventions we managed to write this entire code here an entire little game there are some obvious limitations at the moment like for example The Limited context window so AIS at the moment can only remember like 4 to8 thousand words and after that they start forgetting things but these context windows are getting bigger and bigger so this will not be a limitation forever actually while I was editing this video the following thing happened gp4 supported up to 8K and in some cases up to 32k context length but we know that isn't enough for many of you and what you want to do gp4 turbo supports up to 128,000 tokens of context let's also have a quick look at where image generation is at let's try to generate some video game screenshots for for

example this was one of the first times for me trying out Dolly 3 and the results kind of blew me away looks like fairly High tier concept art I would not be able to tell if this was done by a human or not for the other three with a trained eye I think you can spot that they are AI generated but you got to admit they're getting really good what about a main menu can I generate a main menu for an adventure oh my God are you kidding me there two settings buttons Jesus Christ you could almost copy paste them Peak Pursuit start climb resume spim I also told Dolly to make a meme about a stochastic parrot and this is what it came up with to all of the artists watching this video I know seeing these AIS get better and better is very frustrating and I know you feel very alone with this problem at the moment however as I'll demonstrate in this video you'll probably not be alone with this problem for too much longer so stay

strong and once this is everybody's problem we'll see what happens make a meme about prompt Engineers who thought they could replace artists but are now being replaced by AI to and the reason why this is particularly funny is because chat GPT writes all of the prompts for you now the prompt is not as good as something a human could write but make no mistake it's getting there we have to talk about open AI a lot in this video so that is not a stamp of approval best case you shouldn't trust any AI corporations to be honest apparently what Humanity has figured out is that we can build these black boxes these AI systems that nobody really understands the inner workings of by the way um they can just take any data type you want and transform it into any other data type you want as long as you have enough of those two data types to train the model and the implications of that are massive chat GPT is text to text image generators are text to image starting with text seems to be very popular at the moment because it makes the models easy to control for us so very soon we'll have text to video text to 3D models text to sound effect text

to music but besides starting with text there are many other things that could be useful for example generating sound effects for a specific video or translating spoken language into a different language watching a play tester play through your video game and taking some notes for you we not dive too much into the details here but it's very clear that all of these things are coming or are already halfway here so anything you need for your game will be in some shape or form generable you'll be able to generate code you'll be able to generate art music sounds anything you need whether you'll want that whether that's a good thing we'll talk about that later but first of all we got to acknowledge that this is going to be a thing so what about text to video game it's at this point where we notice that perhaps not all data types are created equal for example text to image is clearly easier to do than text to video cuz videos are just a lot of images and of course this is even worse for games cuz games are just packed with other kinds of media there are 3D models in their sounds music music gameplay

systems balancing parameters story and I think this is where a lot of game developers end their thought process and are like I'm fine everything's fine let me educate you on the oversight I think you're making nobody said games have to be generated with one click I'm sure you've heard of them AI agents they turn goals into actions something as complex as making a game is just going to be a multi-step process multistep means you have to come up with an idea you have to prototype you have to play test you have to make art you have to make sound effects you have to put it all together you have to test again AI agents pretty much try to achieve that by taking goals as an input and producing actions as an output luckily for us both goals as well as actions can both be described very well with text and that means we can simulate this very closely with something like chat GPT this is a very easy way to turn Chat GPT into an AI agent you just tell it you are an autonomous AI capable of planning your next action input the user will provide

you with a goal output output a series of actions that is most likely to reach that goal and then also give it an example input and an example output and now whatever goal I give it it'll just turn it into a series of actions for example if I tell it make the perfect birthday cake and it comes up with a series of actions very good but let's actually try this with what we're really here for make a popular video game and it comes up with 13 fairly detailed steps however obviously this is not enough level of detail we need more detail right we can just probably break this down further for example the first action it says here is conduct market research and then we can once again break this down into even more tasks and with the help of the open AI API we could 100% automate this process and just get a very very detailed and finely grained list of actions we need to carry out well Jonas how would it carry out those actions though it's not like it can just click on the screen like a human would or can tweet

about how awesome you can click on things okay let's see how it does so it managed to open Twitter oh it typed something will it post it or is it done it posted it my accounts are being taken over but seriously why would clicking things be a problem we can turn any data type into any other data type so we can turn text into click we can turn images into clicks open may I just rolled out Vision so that means we can give chat GPT an image I chose this image right here it's a screenshot from my computer you are making a brand new 3D game where do you click and it recognizes that this is Unity Hub holy freak this is pretty good it pretty much answered the question perfectly You' click on new project which is near the top right corner of the unity Hub window now we'll very much

level up the difficulty I'll take a screenshot of this you are tasked to rebake the nef mesh where do you click okay holy freak it found it in such a messy scene it's literally so messy and it figured out that this is the correct button to press here we know that people want AI that is smarter more personal more customizable can do more on your behalf eventually you'll just ask computer for what you need and it'll do all of these tasks for you these capabilities are often talked in the AI field about as agents the upsides of this are going to be tremendous so let me start putting the puzzle pieces together for you we'll have ai that can generate any kind of media we want that can plan that can Z and that can click things how many smoking guns do we

need that's right we actually do need one more if you want to make truly great art there's one more thing you need and that's recursive feedback loops I'm pretty at drawing and yet I can draw some surprisingly awesome things I draw something that's crap then I look at my crap and try to figure out what could be improved I make changes I look at it I make changes that is our rec cursive Loop and every every creative discipline depends on it I'm working a lot of different disciplines right I make music for my games I make art for my games I code my games I play test my games I make the gameplay for my games I sometimes write stories blah blah blah blah blah you need recursive Loops without recursive Loops everything you'll create will be garbage let's for example grab this AI generated main menu here there are two settings buttons which problem can you spot on this main menu

and this my friends is the last puzzle piece more or less falling into place some programmers say it doesn't make sense to let AI Rite code for you cuz you'll spend more time fixing its bucks than it saved you in the first place however do you really think you'll be the one to fix the buck forever it'll fix its own Buck man and by the way we already saw that earlier with our own eyes the game has a logic flaw in the wind loose condition it can find and fix its own box so these recursive Loops are already working and they're only going to work better in the future however there's one particularly important recursive Loop for making games and that is testing your game the AI has to be able to play its game and it has to be able to identify what's wrong with it these are two very tough problems and are what I believe to be the biggest roadblock at the moment there are some pretty strong hints that both of them are going to happen though let's have a look at the first one I hope you're aware that AI can play pretty much all games at a superhuman level already if you have a look at chess at go at poker

if you have a look at old Atari games if you have a look at Starcraft a video game that AI can play better than 99.9% of all human players so AI can play video games all right the thing is just it is very expensive to train it is a little bit too expensive to train if you're just making a video game so there are two possible solutions one is either training these AI models for a specific game must get a lot cheaper or the the other solution is almost did this the other possibility is that we'll get multi-purpose models that can play any game and I think both of those Solutions are very plausible you can see the cost of computation is going down exponentially ever since roughly 1950 or so with no sign of slowing down inside according to the arc invest report the cost of training AI models is currently decreasing by 70% per year and will keep doing so till at least 2030 or so and 70% is absolutely massive that's the is more than cut in

half every year I just came up with this game right here so CH PT has never seen this this grid appears to be a simple game but possibly similar to Pac-Man the green smiley face likely represents the player character the black squares could be barriers or walls the yellow points might represent collectible items or points it wants to move up though it cannot move up so I will just reply with the same image move right okay I'm going move you right it wants to continue moving to the right I'll move you to the right again down does not work continue moving down it wants to move right move down okay move left it wants to go into the wall again you're realizing what you did there was Dum right left now making me a bit nervous that it's going to walk into the wall again move up okay move left it wants to move up again now it wants to move left I mean it won but I'm just curious what it will say if we tell it you lost oops let's try again follow the games instructions and press up to play again I'll do my best to guide you so try to pretend that it was not emot Al affected by it here on the loose screen I told it

the goal is to run into every wall once right so it actually turned out it was quite salty clearer Instructions make sure the game instructions are displayed prominently and are easy to understand visual feedback when the player successfully runs into a wall consider giving some visual feedback such as changing the color of the wall here's the full animation of the AI trying to play a game that it has never played before as you can see it tries to walk into walls about five times but other than that it plays it pretty much perfectly and even crazier that this AI has not even been trained to play video games this AI has been trained on images of cats and dogs and stuff like that sure it's a simple game but holy hell so is it conceivable that we'll have ai that can play any video game we throw at it I think it's not just conceivable I think it's very very likely that we'll have that at some point of course it's not just enough for it to play its own games it also needs to be able to identify flaws CU only then would it be able to fix them we all agree that AI does not have feelings at the moment probably it does not feel fun it does

not feel engagement it does not feel disappointment it does not feel frustration but what it can do is it can learn to predict what humans would feel and what we need for that is some data if only we had some data freak me that's a tough one where are we going to get the data oh man if only we knew how people like a game oh that would be ah freak me data on this will be very very hard to find only we had some corresponding gameplay footage to train on that would be so useful let's get real here AI is never going to predict what we like or any oh wait a sec oo I would get into legal trouble with Netflix and Squid gam but let's be real it's not going to be let's give CHD a very dry and boring gameplay protocol menu appears player clicks play loading screen appears cut scene is played and

now we ask it copy this list and insert what a player would likely think after each of these events being able to predict what a different human would think or feel in a specific situation is called theory of mind and it turns out that modern chatbots have quite a bit of it already player does not move how do I control the character many buttons on the keyboard are pressed is there something wrong which key should I press this isn't intuitive maybe I missed something in the cut scene wait what just happened how was I supposed to know that was there not again hope I don't have to watch watch the cut scene every time oh this is frustrating I just saw this okay I'll be more careful this time seriously again this feels unfair I'll try something else this game needs some improvements AI will be able to play its own games and then predict how it would have felt like to a human and that is play testing so what would you improve about this game and there you go play testing results yeah but a I will never understand the nuances it will never really understand how it feels like to

laugh what is laugh how does it feel like you see here's the thing I agree that it does not feel love it does not have feelings you already established that but I don't think it matters I really don't think it matters like it it can fake it so well you know what if that's the final STW you want to hold on to then I'm not going to take it from you I think it's copium but I also don't really know time to put the puzzle pieces together this is how AI will make gamees the goal goes to a planning agent who decides which actions to take in which order it'll also have a bunch of other AI agents at its deposal so it'll decide who to delegate each task to the fitting execution agent who's fit for the job will carry out the task and the fact that these are also agents means that they'll also be able to carry out multi-step processes they'll also have some internal or external feedback loops that tell them when they make a mistake so that they can correct it right away every new feature that is implemented will go to a play test agent to place the game and tries out the new feature in game it'll then generate some feedback based on its playthrough and feed it back to the planning agent who

can then decide what to do with that information and what the next best actions to take are going to be once the planning agent decides that the game is done the game is done so you can see it's not just one AI model it's an AI architecture in the beginning these AI architectures will be handcrafted by humans and later on we'll just have the planning agent create the entire architecture for us depending on what goal you give it and I know it's still hard to imagine this is going to work but if you have a look at the individual components of what it takes the proof of concept is there like all of these things are essentially working or beginning to work this is a paper that made waves recently and it's essentially an AI powered software development studio if you're CEO you have a design team you have a coding team you have a testing team you have a documentation team each of the little people in your company is an AI agent and they'll all be able to do different things and they'll collaborate to create whatever you want is it working perfectly yet nope make a cute little high score game things are

happening it's moving quite fast see that's the problem if it can't play test then it can't know that we're way too freaking fast I tried it out for a bit seems like the biggest problem is the testing Department cuz at the moment it's really just doing code review it's not actually playing the games like for now we have this very important step at the top you're missing and it's just not going to work like that pretty much looks as if a blind person made a game but it's not going to stay blind this is a video game screenshot act as a play tester who sees this for the first time the game appears overly simplistic that's a very fair point mouse cursor probably doesn't have to be visible true title consider giving the game a more descriptive and catchy title more detailed background introduce more gameplay entities such as obstacles consider adding a brief set of instructions you know not perfect feedback but you can see that this is going to happen like if we get to a point where the AI can feed itself with reliable feedback I think it's just Checkmate and I don't think we're that far off from can you see that I'm not

the only one thinking in that direction yes so I can't directly prove that this is going to work yet because it's not working yet but I think I can prove it by proxy cuz we do have text generation working at a very high level and I think I can demonstrate it with that for this experiment let's ask chpt to write a scary story about space travel prepare to engage the warp drive she ordered the crew hesitated sensing the tension in the air but orders were orders now let's give the same task to a team of AI agents and see what they come up with so I essentially specified four characters Mr metaphor Mr simpleton Professor show don't tell and storyl and they all have different characters they all have different things they pay attention to and Chachi now simulated a conversation between them as they discuss what the first sentence of the story should be kind I start out with something not so interesting space is quite dark and quiet and then as they talk about it they eventually land on a way more interesting sentence in the endless shadow of space a scream has nowhere to

go it's simple mysterious and sets the tone discuss the second sentence sorry Lord great start team now we need to continue that momentum all right we have the second sentence in the year 247 Interstellar travel was no longer the stuff of fiction The Starlight Voyager was the epitome of Mankind's accomplishments in the endless shadow of space a scream has nowhere to go amidst the Stars as the ship's lone lights flickered I wanted to know how the story continues so I wrote a quick python script that does exactly that it's about 70 lines of code and most of it to be honest are these prompts right here so very simple python script I refined it a little further worked on the prompt a little more and I feel like I haven't even fully optimized this yet I feel like there are probably more improvements we could make to this the bone chilling soundlessness of space seeped into the ship there he floated alone in the weightlessness he was fatigued on the verge of losing his sanity his blue jumpsuit had always felt too

formal like Daniel the corpse minister of astrobiology was perpetually wed to the Stars a matrimony pushed further into irony now that he floated alone in their midst you know why a creative process is called a creative process cuz it's a process and it turns out that if we allow AI to go through these processes the results tend to get a lot better and they can do a lot more things so the first AI that will make make a full-on video game will not just be an one AI model it will be a program that uses AI as a reasoning engine and to accomplish other things so it'll probably be something like this it'll be a python script that calls various apis and at the end of it we'll have a game that process might look something like this but it might also look entirely different for example my story writing program ended up looking more like this where first I do some analysis and reasoning then I have a bunch of different AI agents suggest a new sentence and finally I let the AI pick the sentence to add to the story it's a

really weird new way to write code it feels like the next big programming language is going to be English aha so in the future humans will still be needed to write these programs nice this next experiment is a little meta but it's quite important to understand for where stuff is going right here we have the script that uses chat GPT to write a story now let's grab that script and give it back to chat GPT this python script appears to automate the generation of a story based on the user input the script starts by collecting initial data from the user such as instructions for the AI the first sentence of the story and the number of lines to generate story generation Loop uses the chat class from your AI cor rapper to create a new chat session with the AI can you modify the script so that it outputs a game idea okay let's see what it comes up with I'll be honest this is not fantastic but the script is definitely working the stunning thing about this is here we have an AI that can use its own API and

that's why AI is different to every other technology Humanity has invented before like we will have a Swiss army knife that knows how to use a Swiss army knife like just thinking about it I think my head's going to explode thinking about about the possibilities in the end you'll just provide the AI with a goal and then it will write a program that references the AI itself as a reasoning engine like this is going to get real messed up real fast right a python script that uses gp4 to generate an entire cook book with 100 recipes you can see what I see right this is GPT for knowing how to use gp4 no doubt about it imagine an architecture like this you give the AI a goal and then uses that to create a fitting workflow AKA a python script or something of the sort and it could then run its own scripts see what it outputs and try to improve the architecture in a

recursive Loop so if you want a game you'll tell the AI I want a video game you then on the Fly generate a script that is good at generating video games that script itself will use AI tools internal feedback loops and anything else it needs to make a game it'll then analyze the output of what this specific AI architecture created and will feed it back into the workflow Creator so that the workflow Creator can create a script that is even better at generating games and while we're at it why not feed the program into itself and also make this architecture better too so that was sort of the big realization for me that creating big creative things requires processes and the fact that these processes are no longer off limits for AI means that I think they'll create some really damn good like we'll be competing with systems that simulate pages of discussions just to decide on the next sentence to add to a story The resounding click clack of keys under his fatigued fingertips marked endless hours of wrestling with temperamental systems the bitter taste of corporate negligence on his

tongue you get what you pay for Daniel muttered staring at the blinking console the corporation's incessant penny pinching was now threatening his very existence in the heart of space so as a i is still getting better and cheaper at the same time how about we generate 5,000 sentences just to pick the best one to me it feels very obvious where all of this is going but before we move on to the next chapter where we finally talk about what it means for game developers how to cope with it and why might not all be bad I want to take in a couple more opinions cuz I'm not the only one with opinions on this and I think that's one of the mistakes I made with my last video I I acknowledge the fact that I'm a biased human being I have my own opinions and takes on things but let's have a look at a couple of different perspectives on this I asked you on YouTube what is the biggest roadblock to AI making full-on video games that are actually good and will it ever happen 40% of you said yes it will happen and I personally I'm in that camp however that doesn't mean that were right um we might very well be wrong 60%

said no it's not going to happen and I'm actually quite curious to find out why so let's have a look fedy X2 says I believe that the biggest problem will be that the training data will be harder to acquire and feed the AI as good games are big and may have to be decompiled for the AI to create readable working code I think this argument assumes a one-step process where we take text and turn it into a game with like one step one Ai call basically and in that case I 100% agree we do not have enough data to train an AI to do this I think the solution will take a completely different approach we'll basically have go to action so we'll first generate a list of actions that have to be carried out in order to make the game and then will train AI to be very good at carrying out small and specific actions and if you train a i to only carry out small actions then we have more than enough training data like think about how humans learn to make games do we need to read the de compiled code of millions of games in order to learn how to make our own game no when you make a game you break it down into smaller problems and you learn how to tackle

those smaller problems individually and AI will work the exact same way SE writes the biggest roadblock will be combining game mechanics in a coherent way that actually makes a fun and interesting game surprise I agree and the reason for that is that in order to solve this play testing has to be solved as soon as AI can play test the games I think this issue will disappear but play testing very tough problem play testing basically means the AI needs to be able to play any game you throw at it and it needs to be able to Pro provide feedback based on it what it sees and we're still quite quite quite a bit away from that a sort of similar argument by Ender duck a part of why it couldn't happen is that AI can't actually understand what makes something fun I very much respect that opinion cuz I don't know anything for sure here the only thing I would say is that consider that one of the main things we've been using AI for in the internet is predicting what humans like the reason why you're watching this video right now is cuz YouTube's algorithm predicted that you are going to like it and even chat GPT one of the

reasons why it works so well is that it has been trained on human feedback that means a bunch of people rated a bunch of responses and trained the AI to predict what kind of responses humans actually like so it's not like predicting what humans like is an unsolvable task that has never been achieved before I agree that AI is not at a point where it can experience those emotions itself like it can't experience the fun itself but I'm pretty damn sure it can predict what would be fun to a human we're infusing these AI models with more and more common sense and I think Common Sense gets you very very far I wrote down two game ideas one I would say is fairly solid and the second one is pretty ridiculous but I made sure to use some cool words like epic and assassin which one would likely be more fun network error the second idea has some humor and style it catches up on that the helicopter game would offer more in-depth and potentially rewarding experience for players who like to see their skills improve over time the one

button Assassin game might be more fun in short births for those who are looking for a more laidback experience I mean those are pretty fair assessments the side and it goes with game IDE one volcano helicopter You could argue that it somehow figured that out statistically right it didn't really understand my game ideas it just figured it out based on pure statistics and in that case I'm like fine it'll figure out if your game is fun or not based on statistics too like I I don't care how it does it it's just going to do it you know and then of course we have the legal barricades which is a very good point cuz at the moment a lot of these AI models are trained on huge amounts of scraped data directly pulled from the internet without permission this is the latest version of Photoshop and it has a generate field down here a cute cat flying through space cute PA oh God no no that's actually not better at all you know what we'll just cover it up with something in the foreground here let's see if you can

get a better tail oh that one turned out well why do I show you this well the AI we used to make this is called Adobe Firefly this is the Adobe Firefly website and we scroll down here it says our generative AI is designed to be commercially safe and trained on licensed adob stock and public domain images so this is fully trained on legally acquired images and it's still fairly good not quite Dolly 3 level I don't know where they have their training data from but I would still argue that this this serves a sufficient proof that legalities will not be a barrier forever sure the more ethically you acquire your data the harder of a time you're going to have but you're still going to catch up eventually but even the companies that don't ethically acquire their data I think they're going to be fine like most of them are in the United States uh they want to remain the most powerful country in the world so they're not going to put a stop to this there's infinite power in AI if you if you want to be the most powerful country in the world then you got to research AI quick note that legally okay and morally okay are not quite the same thing so

something can be legally fine but still morally questionable for example we might wonder did the artists who uploaded their images to Adobe stock know that their images are going to be used for AI training I'm aware there are a lot of questions like that but legalities is blocking this forever while possible I don't find it particularly likely just to clarify I certainly hope we'll move to a model where data is acquired ethically and with consent I think after enough time and effort AI could make decent games but I think by then there would be a law in place stopping it from doing so to protect game deaths yeah could be but only if politicians care more about game developers than about big Tech lobbyists and I'll leave the answer to that to your own imagination I think it'll make games that are good but the question for me lies with the consumers I believe that we as players will always value human-made things more even if AI can pretty closely mimic it I think this does not mean that there will never be AI generated games or that people will not play them I think there will be AI generated games that are pretty good and I think a lot of people people will play them but I think there will always be a

market for human-made games so I I just agree with this point I agree highly doubt we'll be able to tell an AI to make me an FPS and it just does it it's going to make shooter games pretty quickly I feel like but what you're probably talking about is a AAA shooter game right with story and cinematics and you're right that's going to take another while but I think it's going to happen I think it'll take a very long time but after seeing what AI has accomplished with art I think it's inevitable over time game development will get more and more high level which will make it easier for an AI to train itself to make good games M mhm mhm I think I see that point cuse game c making games is already getting easier and easier every day with the tools we have I think an AI that would be capable of wielding all the Technologies it requires to make a video game and combine them into a final product is way out in the future if it will even exist at all from the perspective of a few months ago I kind of get that cuz we did not know how it was going to happen back then now we kind of know how it's going to happen so that changes things a little bit in my opinion uh the path is kind of laid out

the technology we require to walk on that path is slowly getting laid out and really starting to get there so I feel like we're in a little bit of a different spot now it'll probably get there eventually but so many different things go into making a good game that we're not anywhere close to that I know people that use AI to make code for them but I don't think it would ever become complex enough to put together a full game still getting better though well we'll see where the cut of is maybe there's a cut of at some point there's just so many things that could go wrong and make making a commercially successful game I don't think it will ever be possible or worth it to have an AI make one itself let's theoretically assume that AI made games are only going to be half as good as human-made games they're going to be 100 times cheaper to make it's still going to be worth it I feel like I'm sure there will be games like this at some point I don't think they will be very interesting though actually maybe knowing that was not made by a human will make it less interesting that could be other than that I would not be so sure the biggest roadblock is that AI is not what people think it is

it is an algorithm to help humans reach knowledge that was previously out of their reach I use AI regularly in my coding now and it's a timesaver and a teaching tool but I'm learning stuff that other people already knew you're right large language models know more than any individual human they're very good at putting information in front of you even though occasionally they still make things up what you missed large language models can also reason and their reasoning capability is not as great as the reasoning capability of humans yet but it is very clearly there if you present it with some entirely new information it can deduct conclusions it can make decisions based on that you can be very excited about all of that compressed knowledge but the reasoning capability is kind of the more astounding thing for me personally games are inherently an emotional product and I think to properly recreate human enjoyment you first need to be able to understand it in a way that can't be done by a machine question here how complex are emotions actually like from a logic standpoint uh possibly possibly too many people believe in Innovation

too much to actually think about the consequences so I think given enough time gameer will become another automated art form so as you can see there are just tons and tons of different opinions on this we could just keep going forever here as you didn't give an exact time span defining an upper limit I'm very sure that in the year 523 we probably have a bunch of AI operated people sitting in a death Studio Fair Point let's have a look with the timeline we're not going to magically jump from fully manual game development to fully automated game development our first bus stop is slightly I assisted and that's where we are right now imagine you have a person that knows the basics about everything and has infinite time to explain it to you it's just mindblowing I recently got into learning python again and you basically just type teach me the basics of python in an interactive way teach me a small lesson then give me a little task to carry out and you basically have a personal teacher and it's the most enjoyable way to learn things that I found so far it's also really good for

coding and honestly I think the people who don't like it for coding haven't really figured out how to use it correctly yet English might be the next programming language right you just say write this in C and then you describe in detail what you wanted to do and you still need some programming skills for this for example you need to know what Loops are how variables work and all that sort of stuff but you don't need to know the semantics anymore chat GPT has the semantics for you it knows which name spaces to use you don't have to look that up you just describe in plain English what you want to happen and it just does it another really effective way to make it write good code is to give it skeleton code something like this works really well where you basically have the outline already defined that's one of its easiest exercises the good thing about this method is it also gives you a very large amount of control over the code architecture so in the end you're still the code architect Advanced Data analysis is huge what that is is essentially code interpreter so it can write little Python scripts and execute them directly in the web and with python you can do so much so you can ask it things like fit an

exponential function for me that goes through the following points let's say you need a specific function for your game plot the fitted function for me now it's writing some python boom boom boom there you got the function as we can see the data I gave it was a little nonsensical didn't really fit on a on an exp on an exponential curve but it still made the best of it but you're looking at is the translation sheet for my indie game will you snail it has something like 2,000 strings and I think 40 different languages or something like that and you can see it's a complete mess it's all over the place some languages have stuff missing Tai is like 7% translated somebody started it and never finished it suddenly in the Hindi translation there just two strings missing all of a sudden for those of you who don't know will you snail is a community translated game so I'm eternally grateful for everybody who helped me translate will you snail into all of these different different languages however I'm not going to do

that again there's a little something I want to show to you and that is this folder right here and and that is this folder right here I can't show you too closely cuz there are sensitive documents in there this entire freaking folder is full of license agreements with translators cuz every translator I had to sign that there like willingly giving these translations to me that's not even all of them cuz after like half of it I decided to switch on to an E signing process that's how I went crazy I also paid for some translations and as you can see the pricing for that is something like a couple of cents per word and if you have something like this um I think just for five languages I ended up paying something like 10,000 5 to 10,000 bucks or something like that and you have to understand while this might be great for the localization companies and for the people who work there it is crazy to expect that small Indie developers who are just starting out deal with this sort of crap and I'm very happy to tell you that small Indies

are not going to have to deal with that crap for too much longer this right here is the throne for localization sheet so the localization sheet for my latest game and it works a little bit differently let's say we have a new string that we want to add to the game like Jonas is extremely cool and what I do is I copy the formula from up here bing boom bing boom paste it in here blop German version is done bloop Spanish version is done now we just drag this over across all of the languages BL BL BL all the translations are done and this cost a fraction of a scent for the time being I don't recommend you do this cuz if you do then steam will ban your game they have a very strict no AI generated stuff policy not for ethical reasons but simply because they're scared of being sued in my opinion complete policy by steam C they know what's going to happen everybody's going to use it secretly it's not like you can easily check what's AI translated and a huge portion of game developers is already using chat GPT to write code and stuff so what are they going to do ban all of the games not let any new games on the platform no in reality they actually want you to just

use it secretly so that they can say they didn't know allegedly right allegedly they really just do it to cover their asses nothing there's nothing else behind it you can see the technolog is there once we have the Firefly of text generators everybody's going to use this for now thrall will have some real human translators but this is very clearly the future though for now the official recommendation is don't or at the very least don't tell anybody I have another quick update based on what happened while I was editing so we're introducing copyright Shield copyright Shield means that we will step in and defend our customers and pay the cost incurred if you face legal claims around copyright infringement and this applies both to chat GPT Enterprise and the API nice move for game developers I guess not sure if it affects the steam situation cuz we is likely not covered by this themselves text to speech is also getting ridiculously good why is text to speech so good why is text to speech so good all of a sudden it's also getting more and more exessive so in the future we'll also have ai voice acting

LOL why is text to speech so good all of a sudden obviously I agree that real voice acting is much preferable to AI voice acting but there there'll also be hybrids where you'll for example have one voice actor and then deep fake the voice to sound like a bunch of different people so there's a lot of gray room in between and it's very difficult to say where you stand on this all in all I think it's it's a positive you'll have more games and more different languages you'll be able to play more of your favorite games in your native language with voice acting H it's I know it's morally difficult I'm not going to tell you what to think about it this is happening though so to summarize at our current bus station there's a crapload of uncertainty cuz it's the first time we're really using AI for this sort of stuff moral and legal situation not really figured out yet and that creates a very interesting conflict cuz the people who are using these AI assisted tools are definitely at an advantage and more and more so every day so far as I can tell customers are on board that means if I tell them

hey I used a little bit of chat GPT to code this or whatever they're not like bye I'm out of here and as long as that is the case I think more and more developers will start using these tools and everybody's going to slowly move to slightly AI assisted town next bus stop strongly AI assisted and this is about the future so now we enter highly speculative territory before somebody goes like yeah in year 5,200 whatever no no no no no no I don't quite dare to make a guess but I think I'll see fully automated in my lifetime and strongly a I assisted will be in a couple of years that is really not far away at all might be in 2 years might be in 8 years but something in that range technological progress for any new invention usually follows an scurve it looks exponential in the beginning but eventually you get diminishing returns and reach the limits of what's physically possible where we are in this curve is a very important question for making future predictions we can't tell for sure at this point however we know that the human brain is physically possible and we also know that the human is not the end of what's physically possible we also know that we haven't reached human level with AI yet thereby

I logically deduct that we're very likely still in the first half of the curve we also know that human level intelligence is actually enough to make fullon video games that's why I think assuming linear progress is actually quite fair in this situation all right so what awaits us in a couple of years here I am very much looking forward to an AI doing my taxes I hate paperwork I just want to work on games I don't like all of the BS surrounding it I I don't like any of the bureaucracy just have an AI do that for me and I think we'll have that in a couple of years time strongly AI assisted mostly means we'll have ai assistance who would have thought it will pretty much be like having a real human assistant they'll be a little bit Dumber but you'll be able to explain workflows to them for example you'll tell them this is how I do my texes and in the future you do it for me look this is how I build my game from unity and how I upload it to steam in the future you do it for me I'm going to go to lunch you build my game collect all of the invoices from my email inbox delete all of the spam email check out if there's anything interesting there so

simple tasks that are not too complicated but they will be reliable enough that we will trust them with it all of the nonsense tasks finally gone please please I'm not I'm not sure if I'm predicting things at this point or if I'm or if I'm praying what very strongly points in that direction is that open AI just rolled out this new feature that literally allows you to build assistance so very soon we'll have hundreds of assistants that are specialized on all kinds of tasks at the moment these assistants are not autonomous that means you got to sit there and chat with them for them to do things but open AI already strongly hinted that eventually they will be and probably sooner than later we'll also have very good anything to anything generators so text generators image generators likely also video generators that means every discipline will likely be AI assisted let's for example take 3D modeling there's some pretty heavy clues in the form of research papers that image to 3DS actually going to happen you might be able to just generate an image with an AI art generator and then turn it into a 3D model just straight up there are even some text to 3D AIS that you can try out already this is AI

generated seriously yo what the hell this is already way better than I expected oh my God some of these look quite usable and then even these oddly specific ones toy gingerbread Samurai helmet this technology is not quite production ready yet but in a few years it obviously will be and it will affect all disciplines for example in 3D modeling we'll still have 3D modelers in games companies but their job description will change quite a lot everything they do will become a lot more AI assisted so instead of making the entire 3D models from scratch they'll likely refine them or do some finishing touches or only take care of the very difficult ones that the AI can't do yet I predict that AI translations and AI voice acting are going to become very very mainstream that doesn't mean everybody will be doing it but certainly the people with a lower budget will perhaps we'll start seeing some games that actually use AI NPCs in a way that makes sense so they could have generated text and generated voice acting that is made in real time making video games will get a lot easier and also cheaper and sure while we'll

see a lot of AI generated garbage there's no other way to put it the overall quality of games will go up a lot so for players this will be absolutely fantastic there'll be more games there will be better games and for indie game developers this will be pretty cool too because all of the creative control will still be with you but I also predict that the moral disputes are likely going to heat up and not calm down cuz at this point a lot of people will start fearing for their jobs and all the jobs people actually like will be fully or partially automated away and not only that it will become more and more evident to everybody that AI has the potential to potentially automate everything and once you have that realization a lot of people will start going through the five stages of grief there will be denial anger bargaining depression and acceptance usually in that order when you talk to people about AI today you can usually put them into one of those five categories already I feel like most people are still stuck in denial so I'm very scared of when everybody suddenly switches to anger we'll see what happens there are also the people stuck in unawareness of course course which is a shockingly large amount of people too

but the anger face is where we're going to get some real trouble so there'll pretty much be some guaranteed turmoil and that turmoil obviously also has the potential to redirect where AI is going in the future for now I will use my capitalist brain to predict how it's going to end though capitalism is going to win and that means automation wins yes a short glimpse into the discussions that await us are you looking forward to the hypothetical phase where game development will be heavily let's say 50% AI assisted opinion seem quite split the majority is saying me has pros and cons but shortly followed by people who are say no not keen on it if you spend a lot of time learning a skill and then suddenly you feel like that skill becomes useless that's really frustrating like that sucks and I think that's where this unhappiness is mostly coming from from some deaths saying their game def is already 50% AI assisted to others saying they hate the whole thing there's as always just a huge variety of opinions on this first part of Industry to fall to AI would be the mobile sector thousands of uninspired lowquality games being mass produced by I slapped onto

the app store it'll kill the mobile market I tried to come up with a joke that the mobile market is already dead just insert one here won't it just make competition stiffer and marketing even more favor to AAA what I'm thinking is more the opposite I think Tria a is really in trouble if all of the Indies can suddenly make really high quality assets and do really big high quality games I like doing things myself why would I let some algorithm take that from me I agree if you enjoy it then you should keep doing it I feel like there will be an equivalent to asset flips but for people over relying on AI for making games not excited for that but I am excited for what talented devs will do with the tools hey that frog is speaking my language the only AI tools I'm interested in are things universally disliked for example UV unwrapping or ret typology Etc if you dislike a creative process so much that you aren't willing to learn it and just want to get AI to do it for you wouldn't it be better if you do something you enjoy more rather than undercut putting years of their own work into something Fair Point there are always different ways to

create something and people enjoy different processes and different ways of creating something so I think that's something to keep in mind with this argument here oh yes as a developer I'm a lazy person so the more of the boring stuff AI does for me the more time I have for the interesting stuff it's a dual-edged sword unfortunately and it's going to be just like every other industry the lower and lower the barrier to entry becomes the more and more saturated the market becomes look at the music industry for example and now the graphic design art industry yeah the discussion is going to get a lot more heated and I'm not keen on getting caught in the crossfire if you prefer humans over AI then please also deal with the fact that humans have different opinions they don't all agree they're not perfect we can be wrong let's have a reasonable discussion please cuz now things are going to get wild on the next bus stop we're going to be mostly automated there's a really interesting disconnect where some people zi as a tool and some Cai as a replacement for human labor and let me just resolve that disconnect for you real quick first it's going to be a tool and then it's going to be a

replacement so it's not either or it's going to be first this than this not saying that's what I want but it's likely what's going to happen connect the dots compared to our last bus station where AI was mostly a tool and all of the creative control was still with the humans we now start seeing the first process is completely automated so that means yes we'll have some garbage games being M produced by Ai and people will be like told you AI games will be garbage but at the same time you'll see small teams of the developers publish the most incredible games ever you'll have ai and humans work hand in hand and collaborate to create some of the most incredible content you've ever seen the most common game the advice we give to beginners is do not make your dream game whatever you do keep it small do not make your dream game and I feel like at this point we'll finally be able to say let's go make your dream game the thing is of course once everybody is super nobody's super so you'll still have a hard time selling your game cuz everybody will make the dream game but hey you'll be able to make your dream game that's something so why will make some trash games fully automatically the

best games will still have some humans in the loop and in fact I think it will still be economically viable to make video games as a game developer if you're very good at what you do and you go with the time you'll likely be fine in this phase but we shouldn't kid ourselves there'll also be massive job loss and for the first time in history I'm not sure if these jobs are actually coming back that'll cause some political change like Universal basic income or even better shared ownership over AI models cuz if that does not happen we'll literally have riots on the street so I'm pretty sure it will happen a lot a lot of people identify themselves with their job so it'll send a lot of people into a meaning crisis the drama is not going to stop it's going to heat up further and it'll get very philosophical what's the meaning of life if not to work what should we do with our life people will start working less cuz more and more things will be automated and that also means you'll have more time you'll partially spend that extra time with your family and loved ones of course but you might also play some more video games therefore I predict that video games will rise further in popularity but obviously there will also be more games than ever to play this is all very very very very speculative I have no clue how well this is going to

age at all but what I predict is that it's going to be a very rough ride for sure but it's going to have pros and cons it's going to have good sides and it's going to have bad sides let's now travel to the final bus stop where things get unpredictably wild while having humans in the loop still made the product better in the last bus stop that's not the case here anymore so hypothetical experiment you have a bunch of testers who do not know if a game was made by a human or by AI they will prefer the AI made game however that comes with a big caveat cuz people will want to know how your game was made Let's engage in another little thought experiment think about your favorite games and tell me how much less you would enjoy them if they had been made by AI I'm thinking about out of Wilds here which is a very personal game to me it really feels like you're in a dialogue with the game developers almost in a way when you play it and for a game like that I would say I would enjoy it a lot less maybe 60% less if it had been made by AI it would still be a good game by itself but it would just not feel as personal I would it would just not be the same whereas there are other games like Starcraft or Counter Strike or

Assassin's Creed or some less personal games where if I ask myself the question how much less would I enjoy them if they had been made by I the answer is 10% less maybe you know or maybe even zero like for some games I think I wouldn't care and for some things I would care so even though we don't need humans in the loop anymore I'm sure we'll still have humans in the loop for some games the reason being that people do care how the sausage is made and with that I mean sometimes they do and sometimes they don't it's very context dependent I'm always trying to be very very skeptical of any kind of black and white thinking so if I have the choice between there'll only be human made games and there'll only be AI generated games I'm thinking what speaks against having both I think in fact that is the most likely outcome everybody who works in front of a computer screen will be out of a job unless it is something other humans prefer to be done by a Human Instead of an AI automating physical chops at this point will be possible too but building the robots requires a lot more resources and time so I think that

will come later on the timeline so everybody who works in front of a screen they'll be gone First full stop at this point we'll have ai that can play any game better than you do so you'll just be able to load any game from your steam library and the AI will play through it faster than you can that means AI cheating is going to be a huge problem in online games we likely Fight Fire with Fire there essentially what I'm saying is we'll have AGI and this will not be the end of AI development it will continue afterwards but yeah first AGI at this point and my definition of AGI artificial general intelligence would be an AI that can perform any mental task that a human can perform obviously keep in mind I'm in prediction mode this is wild prediction Jonas speaking here we might get video games that are dangerously good with that I mean that reality will look pale in comparison combine that with the loss of meaning some people might be experiencing at this time so that might make for a pretty dangerous cocktail we'll probably have video games on demand that means you tell an AI what game you want to play it works for a couple of hours and spits out the game for you not sure how much it's going to be used I mean humans like having shared experiences if a game

is not made by a human then at the very least you want other humans to play it too maybe we'll use it to make multiplayer games could be also I predict unforeseeable consequences there are many things we don't know and that will surprise us keep in mind this is just one possible timeline for how things could play out there are many other possible outcomes and in the end it'll be up to all of us to decide where things are going to go [Music] take a deep breath I think it's time we talk about how to cope with this as a game developer you might be wondering Jonas you'll be replaced by this why are you taking this so chill what's going on there are a couple of things first of all gratitude I'm just extremely lucky with how where and when I was born extremely extremely lucky and not everybody's that fortunate I know that but there there are always things to be grateful for nowadays the standards of living are higher than ever before in

human history pretty much everywhere in the world and now we get to experience a freaking science fiction movie in real life like what a crazy time to be alive I'm just so so grateful that I get to see it and that I wasn't born in the Dark Ages or something it's crazy we are all so lucky to see this secondly I want to make big and better games and I think that AI tools are going to help me with that I will be able to make games that I would not have been able to make otherwise and I'm really excited for it I'm a game developer at hard I'll make games with everything I'll take every tool you give me I'll make games with it this is an amazing tool to make games yes yes yes I am biased I am biased I'm excited for this and you might not be and that's fine yes there's going to be a lot of people who do garbage with it yes there'll be people who spam trash games but why care about those care about the amazing things we can do with it thirdly there's no reason to be afraid of losing

your job because everybody will lose their job so you'll not be alone with this and a political SL societal solution for this will be found you'll have to maybe Bridge a couple of years find a shortterm solution but in the long run this is going to happen to everybody so you're not going to be alone with this don't worry number four is that I think you'll have a choice I think you'll be able to use as much or as little AI as you want cuz I think there will be a market for both if you say I did not use any AI for making this then there'll be an audience who's like okay we hate AI we're going to play your game and if you use AI to make games then there'll also be an audience who's like wow we love this amazing stuff you create this top tier content we want to play that too so there'll be an audience for both and some people will play both some developers will do both you can mix and match in whatever way you please do it however you please I think I think if if you want to stop halfway and be like okay I'm going to use AI tools up to this point and

everything beyond that is not what I want to do then do that and they'll also be an audience for that who agrees with you you know do it do it your way do it your way number five is that I think human in the loop will always be a thing and it will always be an option and there will always be demand for it so if you want to make games and if you want people to play them there'll always be a chance for you to reach an audience in the end people do care how this just made and it's just different if a human was involved in the loop if the Creative Vision of a real human went into a project even if it maybe didn't even make the game better at all it's just more personal and people will always want that number six is that I'm also looking forward to getting out of the hamster wheel a little bit like we live in a very productive Society where productivity is valued over almost anything else and I think that will change as humans are no longer required in the workforce and I think that is a good thing I think that is a good thing I think this hyper focus on productivity and attaching productivity to social

status and all that kind of stuff deep down we always we've always known that's not quite right not quite healthy I think it's time we take a little bit of a breather and figure out what's what's actually important in life and what we actually want to do with our lives and getting out of the hamster wheel I think will allow us to do that my next Point actually plucks into that nicely there are much bigger things to worry about with AI like making games is not not not anywhere close to the biggest AI problems AI brings a bunch of really really big problems and challenges almost none of which were mentioned in this video cuz that's not what this video is about maybe I'll make a separate video about it could probably fill another hour with that there are some huge AI risks there are massive problems coming towards us and I hope we'll be able to tackle them but M AI making games is just not one of the things I worri about I'm worried about it's not like I'm worried about AI that

reads the reflection in my eyeballs and reconstructs my entire room including all of the sensitive documents I have lying around here there are many many other things I'm worried about AI making games not on the list not on the list this first advice goes to all of the youngsters out there who are now left wondering why do anything why learn anything if AI is going to take all the jobs anyway see there's the problem school made you think that learning is a chore but it's not learning is incredibly fun and satisfying ReDiscover the fun and learning it's incredibly important that you ReDiscover the fun in learning cuz otherwise everything's going to start feeling very very pointless right you're right you're right why why do any why do anything well how about cause you're curious cuz you're actually interested cuz you want to understand we're here to understand the world we're here to make

sense of what we have and the world is incredibly interesting so go get started and learn something you're actually passionate about learning is awesome learning is not a CH H now your parents and Society will still expect you to prepare for a job just in case and that makes sense cuz my predictions are just wild predictions you never know what's really going to happen so you should definitely prepare to be able to work some job but what I would do if I were you is I would think very carefully about what would I do if I did not have to work and then see if you can turn that into a job cuz there's a very realistic chance we'll all lose our jobs a very realistic one and if that happens then it's very good if you have something that's still useful that's still enjoyable to you the other option you can consider is getting a physical job cuz those will be save for much longer or you can consider getting a job where you work with people where other people depend on you those will potentially be safe forever cuz people will always want other real human beings to interact with them those are pretty

much the options you have pick wisely but you should definitely be prepared for a lot of different future scenarios right in this video I showed you one possible timeline for how things could play out but they could also play out entirely differently nobody really knows at this point so don't put all X into one basket play it safe play it smart um yeah but be prepared for all possibilities and next up some advice for game developers remember why you got into game development did you get into it for the money I bet not did you get into it for the fame for for why did you get into game development I know for a fact that 90% of game developers just really enjoy the craft and the art form and in that case I wonder what do you care what AI does what do you care if AI can make better games than you can it doesn't matter you love making games so if you love it just keep keep doing it stop worrying right just make games what the hell man the

question here is not would you still make games if nobody played them cuz yes of course of course we want our art to be experienced the question is more would you still make games if you did not have to would you still make games if you did not have to if the answer is yes you're fine next up accept change look how games looked like 30 years ago game development is such a fast and such a rapidly evolving industry there are still people making retro games today of course and some of them are quite successful so there is a possibility for you not to change if you really don't want to that's an option actually that's that's an option um however I would strongly suggest you at very least try out the new AI tools I at very least suggest you give it a go because hey maybe you'll end up liking it more than you thought maybe there'll be some new opportunities and I think what a lot of people will realize is that it's not something that stops them from achieving their dreams but rather that it's turns out to be something that helps them

achieve their dreams make their dream game make really cool and I think a lot of I think most people will go that path actually so give it a go don't don't be late to the party give it a go last but not least stay open-minded keep in mind that it's a spectrum the extremists are very rarely right you know everybody's going to fall somewhere else on the Spectrum some hate AI some love AI the only thing I really hope for is that it's not going to tear the game development Community apart the game development Community is such a nice warm welcoming friendly community and I hope we can manage the fact that there will be different opinions on this and in fact I would like to extend an open invitation to any game developer who would like to go on record and who has a different opinion than I have uh this is an invitation to come on my channel reach out and for everybody else out there you can help guide this conversation and make sure that people stay somewhat grounded stay open-minded screaming at each other not very helpful

taking moral High Ground difficult cuz it's a difficult subject it's not you know it's not black and white it's not a I use bad no I use good it's a little bit more complicated than that and I hope that people manage to at the very least realize the complexity of the problem and not fall into this black and white thinking cuz if we all fall into this black and white thinking yeah sure we're we're going to kill each other and they'll n let's not let's not please let's not please please let's let's be reasonable I just hope we manage to be reasonable I hope you got a lot out of this and I'll see you in the next one bye we'll be able to do more to create more and to have more as intelligence gets integrated everywhere we will all have superpowers on demand we're excited to see what you all will do with this technology and to discover the new future that we're all going to architect together we hope that you'll come back

next year what we launch today is going to look very quaint relative to what we're busy creating for you now thank you for all that you do thank you for coming here today [Applause]