AI Will Make Better Games Than Me One Day
ฝัง
- เผยแพร่เมื่อ 19 มิ.ย. 2024
- Artificial Intelligence will have some massive impacts on game development, not only as an incredible tool for game developers, but we're also getting closer and closer to AI agents making full on video games without any intervention. This is a deep dive on the subject. We'll talk about why AI generated games are much closer than a lot of you may think, what it means for the future and how to cope with it as a game developer. Enjoy! :)
The Basics:
0:00 - Intro
1:31 - The State of Text Generation
2:54 - The State of Image Generation
3:39 - Prompt Engineering is Not Safe
4:25 - AI Black-Boxes (Anything to Anything)
5:40 - Why Generating Games is Hard
How AI Will Make Games:
6:13 - AI Agents (Goal to Action)
7:51 - Agents Taking Action
8:31 - Agents With Vision
9:24 - The Future of AI Agents
Recursive Feedback Loops:
10:05 - Why Feedback Loops are Crucial
11:27 - How AI Will Play Its Own Games
14:49 - How AI Will Feedback Its Own Games
AI Architectures:
17:25 - How AI Will Make Full On Video Games
18:34 - Prove of Concept: Chat Dev
20:03 - Prove of Concept: Writing Agents
22:12 - Creative Processes
23:06 - Self-Improving Creative Processes
25:28 - Why It Will Be Difficult to Compete
Will AI Ever Make Great Games? - Community Opinions:
26:46 - Most Think AI Will Never Make Great Games
27:06 - Lack of Training Data
28:05 - Making Coherent Games
28:34 - Understanding Fun
30:33 - Legal Barricades
32:23 - Laws to Protect Game Devs
32:40 - Consumers Will Decide
33:04 - Capability Doubts
34:26 - Profitability Doubts
34:58 - Information Retrieval vs. Reasoning
35:44 - So Many More Opinions On This...
The Timeline
36:28 - The AI-Assisted Phase
36:37 - How AI Can Help Game Developers NOW
38:26 - Example: AI Localization For Games
41:47 - AI Voice Acting
43:14 - The Strongly AI-Assisted Phase
44:15 - AI Assistants
45:36 - Advanced Anything to Anything
45:46 - Example: 3D Modelling
46:41 - More Predictions & More Drama
47:28 - The Five Stages of Grief
48:16 - How The Moral Debate Will Continue
50:42 - The Mostly Automated Phase
53:09 - The Fully Automated Phase
Coping
56:30 - How to Cope as a Game Developer
56:46 - Gratitude
57:24 - Making Better Games
57:58 - If This Happens, It'll Be Everybody's Problem
58:18 - You'll Have a Choice
59:10 - You'll Always Have the Chance to Connect to an Audience
59:39 - Finally Beginning the Real Search for Meaning
1:00:20 - AI Making Games Is The Least of My AI Concerns
Advice
1:01:18 - Rediscover the Fun in Learning
1:02:12 - Consider All Outcomes
1:03:24 - Figure Out Why You Make Games
1:04:16 - Accept Change
1:05:11 - Stay Open Minded
My game Thronefall:
➤ Play Thronefall on Steam: store.steampowered.com/app/22...
➤ Join the Thronefall Discord: / discord
My last game "Will You Snail":
➤ Play Will You Snail on Steam: store.steampowered.com/app/11...
➤ Join the Will You Snail Discord: / discord
For the game developers among you:
➤ Join our creative game dev community on Discord: / discord
Hope you enjoy. :)
#gamedev #indiedev
![แฟนแนวใด๋ - ยูริ โตเกียวมิวสิค [ SyncVersion ]](http://i.ytimg.com/vi/Fm8oDwos2uc/mqdefault.jpg)
Not really a video I wanted to make, but I hope to spark some constructive discussions around this. Please keep it civilized in the comments.
Why did u make it? Do u feel like it was needed?
The fact that you didn't want to, but still did, only goes to show that you care more than most people. This is why you're an inspiration to me :D
@@Tholuc98 I was hoping to get the conversation started and to get some more people out of their "denial" or "unawarness" phase.
I think you bring a great perspective to this conversation. Lots to think about here, thank you!
no i will not
Me: Can you make a profitable game?
Ai: Can you?
Me: *Cries
😂😂
Lmaoooooo
Is that a iRobot reference lol
Haven't finished the video yet but I wanna say I really appreciate you making a video like this, just to at least get us thinking about the future and not be so surprised when all this stuff inevitably becomes reality
Yes, agreed. The discussion oh this has to get started.
what if AI turns out to be absolute garbage and actually, you are so surprised after all because you thought all these predictions would become reality?
Absolutely loving the more journalistic, long-form videos. This was really fascinating!
Happy you found it interesting.
As a "youngster" who kinda had that existential mess you talked about toward the end of the video back when gpt-3.5 first came out, the thing about why bothering to learn anything anymore, your message about finding a way to enjoy learning, and finding a profession you would want to do even if you didn't have to, hit directly on what I eventually came up with for myself. Funnily enough, playing Outer Wilds right then, among other things, also showed me that I could find learning fun, it just had to be the right thing. Since then, I think I've found a study I truly enjoy and would pursue even if I won the lottery tomorrow (though probably a little bit more relaxed about it haha). And just in time for College Applications here in the US no less! While I'm lucky that it's nuclear physics, a generally respected and well paying pursuit, I hope anyone else like me reading this finds something like that for themselves, no matter what it is. We're gonna have an interesting next few decades! Anyway, good video, thanks Jonas!
Your advice around 1hr3m hits the nail on the head for me: from a rational standpoint, there are already humans and massive corporations capable of making games I'll never be able to make, working in teams hundreds of people strong. If the point that I won't be able to keep up with AI is a dealbreaker, it would have already been in play in the face of those super-talented individuals or massive production studios. AI being widely accessible could theoretically level the playing field a fair bit.
You always surprise me Jonas
I'm extremely impressed by how well-made this video is
Among the probably hundreds of video I've watched about AI over the years, this might be straight up the best one in terms of "general" quality
the way you separated what is true and real now and your speculations was perfect
focusing on a small part of the discussion instead of trying to just put everything in there was just brilliant
and the way you communicated what makes this not depressing to you and why it doesn't have to be depressing was done perfectly
This shouldn't stop anyone from making games
Agreed.
True
@@JonasTyrollerI'm scared
Because no matter what, you can show people what games used to be
AI content has a flaw: it can only generate content while human content exists.
Future AI-generated content will need to be "flagged" somehow so that future AIs don't use it as reference.
otherwise, stagnation will happen, since the AIs will be caught on a loop of self-feeding information.
and when they are flagged as AI and easily recognized, people will dismiss them, filter them out if possible.
I've definitely been in stage 4) (battling) depression concerning AI for months now. Like you say, I'm experiencing grief/fear. I'm a bit older already, nearly 40, but relatively new to game dev as I started switching careers just a few years back and only this year am making some money with it. I feel like I waited too long, I'll be swept away by this AI tide, and I'm not sure about UBI coming in on time to save us. But...
Then I think that this must be similar to how the previous generation felt, doing all kinds of things that then got automated and/or digitized. I'm sure a lot of people liked working with the pen, paper, books. But it turned out fine, mostly? Work got more efficient and in that way probably a bit draining, but people can still write if they like doing that, and because then they do it because they just want to, it's perhaps a bit more special now than then. And then I too wonder if perhaps I'm just too focused on what we're losing. But if I feel this way, how must people feel that spent years of their life specializing in making art, like especially concept art. I dabbled in art a bit, and then I saw a friend of mine use AI to generate stunning, mostly professional looking images for a custom card deck he's making in relatively no time at all, and already I feel such a pang of loss. Feel like I'm just worth less. But is that the meaning of life, being "worth" something?
There's potential out there for a hopeful future, with less work, more time for family and friends and for doing things we like because we like them (like making a guy move across the screen and have adventures). And I so hope that that's what's waiting for us. But fear is telling me something else, and it's giving me trouble. I feel like each day for the moment is a battle to try and search out the positive, to reconnect with what's important. So as when I got to the cope and advice at the end of the video, I very much appreciated you taking time to add it to already such a long video. I think there's others who need that. Even though I'm nearly 40, I'm taking your advice for young people to heart as well. Thank you
I think manual human made things are not becoming less valuable. In the future you'll just have to mention how it was made for people to really appreciate it. Just like people will appreciate your beer a lot more if you tell them it's a home brew. If you make a card game with crappy art that is entirely self made, people will still very much appreciate that, especially once the first wave of AI hype dies off a bit and people are more familiar with the technology. I understand that feeling of loss, but I also feel like it might not turn out as bad as we imagine it now. Humans value handcrafted things if they know that they are.
And other than that it is certainly time to stop tying self worth to productivity and attach it to something else instead. A healthy skill to work on seems like social skills and fostering your social contacts. We don't have to do that immediately (it's difficult to do), but if you start with that journey now, you'll actually be way ahead of almost everybody else. And yeah, keep learning things even if it does not seem necessary anymore, just for fun. Life is interesting. All the best to you man!
AI is not likely to take over in any of our lifetimes; it won't be there for a few hundred more years at least. However, AI, like the invention of writing systems and even calculators, will be a very handy tool. For the foreseeable future though, it's very gimmicky and only good and hyper focused specific niche tasks. It's completely overhyped at the moment. It's all very impressive, but it's a massive farcry from replacing the genius of humans. The only humans who should be worried are the stupid sheep who lost all their intellect and creative thought to the school systems... well, I guess that is a lot of people actually on 2nd thought...
@@OxygenBeatsA few decades without regulation more like it.
Yeah our jobs won’t all be gone right away, but it’s already been incorporated into human ideology. The information that we see daily is controlled using AI that is getting better and better at pushing screen time.
Best case scenario you buy a new toaster you’ve been thinking about, worst case scenario you change your ideology and slowly become an extremist of some sort because of a constant push you don’t even feel.
This might sound dystopian and dark but seriously we are living it. The deeper and darker the rabbit hole the more time you spend in front of a screen. Flat earthers didn’t read books to start believing the nonsense, they spent hours reading/watching videos recommend by AI on their phones/computers. Flat earthers (while crazy), are a tame bunch at least, but not all of these ideologies are.
We are playing with fire and most people don’t even see the potential to burn the whole house down.
It feels so good to hear someone such wise talk. This propably made so many people think again and widen their toughts :)
I hope so. I hope so.
The problem isn't really the ability of AI though, but the unethical sourcing of training data
EDIT: ah, seems like some of the later chapters might talk about it? Let me check them out
THIS IS INSANE QUALITY
Thank you Jonas, you definitely know what you're talking about - I've never seen this topic demonstrated and explained this well
Happy to hear it. I gave it my best.
Very glad to see someone else voicing the concerns I've been thinking for over a year now.
Yes, discussions on this need to get started asap. I agree.
@@JonasTyroller unfortunately this concept has kind of consumed me into a depression blocking me from any desire for creative work. It's so hard to get the motivation to work when I know I'll be out of a job in a few years time. I just graduated last year, two months before chatgpt was released and I have a job currently but I am so demotivated to improve and work on new skills
Being someone who makes video content and has been for the past like 8 years, I've been learning various different skill sets to augment my abilities. The most recent avenue I've gotten into being 3D animation and some slight modeling. And I'd be lying if I said I didn't have a bias against A.I lmao.
Generally I think there's two reasons why anyone decides to learn how to do anything. The first is - obviously - to earn a wage. The second is to have an impact in society in some way or another. Leaving your mark, so to speak. Any job does this honestly, no matter how small. In regards to creative mediums like music, art, games, ect, that impact is you showing and telling a story to the world.
And A.I, kinda takes both of those reasons away. Or atleast it will.
You mentioned how you believe that there will always be an audience for games made by human hands. And whilst I agree to some degree, it's only on a very, very small level that I do, and I seriously doubt that it'll last. I believe that over-time, more and more people who are stubborn and refuse to engage with A.I will be converted one way or another. I think the biggest factor that'll drive that, is these people seeing *everyone else* convert. The more people engaging in something, the more an individual wants to engage with it themselves - that's just how our brains work.
I think there *will* be hold outs of people that only care for human made titles, but it'll at some point shrink to such an insignificant number of people that it just won't matter. I'm not saying that like "Oh, art only matters if *thousands* of people see it!"
No, what I mean is that the majority of media will be made with A.I, meaning future generations will thus be more likely to grow up with A.I generated creations and have less attachment to human made creations. Which results in that generation adopting A.I more, shrinking the already small pool of human 'purists'. Their next generation will grow up with even fewer human made content, be raised with even more A.I created content, and the cycle will repeat.
This is the part for me that's the hardest to accept. Because when I make something in a creative medium, it's because I want to share my thoughts or feelings in a unique way besides just writing down words onto a sheet of paper. And I think the reason people feel the need to go to lengths *besides* just speaking their mind, is because to do so in a skill set that not everyone has, makes it more special.
But when that threshold is reached - where the vast, vast majority of people prefer A.I generated content, and where anyone can get to an A.I to do something by just *talking to it*, a skill set with a skill ceiling about as high as a slice of bread - there will be far, far less reason for anyone to pay attention to whatever it is that you may have put your soul into. No one will care about each others arts or work. Because A.I will nullify the vast majority of effort that would have otherwise gone into creating it. When everyone is super, no one is.
At that point, creating entertainment media in any form will be almost pointless. It won't earn you a living (though, hopefully you won't NEED to work to have an enjoyable life by that point), and any stories that you'd want to tell would be drowned in an ocean of A.I content. And hell, your idea would probably already be created anyway. Like a Library of Babel situation.
Of course, everyone has different motivations for learning things. Money, to help people, to entertain people, ect. Heck, I'm sure some people like learning skill sets purely to say that they *know the skill set.* But for me, as I said, I like learning creative skills mainly to express myself to other people. And that motivation I believe, is on the chopping block.
Generally, I consider myself an optimistic person - in general, outside of the talk of AI, I think the future is bright. Even though I think most people think the opposite. But A.I and its relation to creative mediums is one aspect of the future that I am deeply worried about. I like creating things, but other than being able to live off of that, the biggest reason why I do so is for other people.
And the opportunities to create our visions and have that mean something, are slowly closing. And it's unfortunate to exist in the time where we remember when those doors were wide open, when it'd probably be best to exist when the doors have since been closed fully long ago, or in a time where you wouldn't have to worry about them closing in your lifetime.
I often see people say that opinions like mine are overreactions, and will point to historical technology evolution for examples as to why. Things like the camera, steam engines, automatic letter presses, ect. But there's two huge, *huge* difference between those past examples of growth, and A.I:
1st, all of those things, still required *skilled* human input to function.
2nd, Not only is A.I's endgame to be as easy to use as possible to operate, but A.I can even *use it's self.* The latter being the biggest difference between all other innovations we've seen. Seriously think about that for a moment: this is the first time in history, that **something other than a human**, is being able to work with human interfaces in a coherent manner. THAT'S the difference here.
But, we gotta count our blessings. For now, there's still plenty of time and plenty of reason to learn new things and not worry about it being wasted time.
While I agree that AI might take over certain things in the future, my worry is that people trust the output from it too much and stop relying on critical thinking and become lazy. AI as a tool is good but I feel like we must still be alert to catch any weird output that's generated. Of course, a day will probably come where it's 100% error free but I just can't shake the feeling that I always have to vet through everything which is just about as time consuming as doing the thing myself. Probably just a skill issue on my part, great video as always.
AI's unreliability is one of the main points preventing mass adoption at this point. I think AI companies are aware of that and working hard to resolve it. We'll see how well they do I guess. I also think it is very warranted to be worried about AI for many many reasons. Unreliability or even the exploitation of trust being much higher up on the danger scale than AI making games for sure.
I agree with this 100%. I was working on a software project with a classmate a while ago and he was using github copilot. He kept just hitting enter and generating stuff without looking over it, and then kept wasting time sifting through the code again to figure out what went wrong when it didn't work. Even if AI gets really good, as long as there is error (and since training, weights etc are essentially random there will always be error imo) it will compound itself, especially if you slap another AI onto it to "correct" it. It might even hallucinate a mistake and correct it making it worse, etc.
In my opinion this is similar to AI driving (Tesla)
People just are relying on them to much, sleeping behind the wheel
But there is no fun or enjoyment in that things
It's only really been a couple of years since the huge uptick in the AI trend and general web usage experience has plummeted. Due to all the theft websites are trying to obfuscate data and the hype behind "AI" has meant jumping away from tried and tested data analytics or pumping out low tier articles 24/7. I hate how much censoring this is going to bring about as well. The fact I have to jump gates and break LLMs to be able to get it to just tell me a dirty joke is a bad sign for seeking truth in the future as truths are often an uncomfortable thing.
agree, I think this is happen sooner than later, then humans will be the non trusted source :)
I deeply appreciate you creating these types of videos. No one else has added this much context to such a "game changing" (heh) topic. Great video
I really like this 1h long videos. They are very fun to watch, informative and make you think for a bit. I'm looking forward to see more of this type of videos in the future.
I'll keep doing these if good opportunities for them arise.
man. Just 17m in and already know you're spot on - Many people miss small details when thinking about this. It leads them into a false sense of security - which is normal, humans are scared of sudden change. But the potential for good here is so immense. Extremely excited for the future.
Edit after finishing: Yeah. Spot on man. You put this into words so well - exactly my thought processes too. The future will be wild, and we're along for the ride!
i think everyone has "false sense of security" you just mentioned, but in a much broader sense. what make humans.... humans? we are just brains, a network connected with electric signals. its gonna get simulated anytime soon. we are gonna have a REAL issue when AI threatens our entire identity, not just our jobs. huge accomplishment of evolution perfected through millions of years. are we really sure humans are precious beings, even when there is an Ai functionally exactly same?
@@junyong0716 What you're saying has zero basis in reality and shows you know nothing about any given technology.
Also, you're heavily underestimating the adaptability of humans (something AI will never achieve, atleast not with the current binary technology).
If AI takes over jobs, more jobs will be generated. Most notably the entertainment industry.
If AI takes over identities, more people will eventually learn to recognize false patterns.
I honestly look up to you so much man. You're able to be incredibly silly at times but still facilitate really interesting and important discussions for the community, all whilst making some awesome games.
As an illustrator i feel we have speed run some of these steps already. I'm betting on elements that show concious intention like composition, shape selection, paraeidolia, intentional roughness, etc..AI generated art does have (so far) a certain sheen to it and other common artifacts that give it away.
I can instantly tell most of the time but I'm also artist that's been doing it for many many years. Us artists are much better positioned to take advantage of it though, I can sketch out generally what I want and have multiple variations of clothing ideas or whatever you like and finalize the design. Normal users are writing prompts hoping its close to something they are imagining. One thing I don't see these generators figuring out properly is pixel art and properly animating these.
Also instead of text to speech, you can voice the lines by yourself, and then convert that to the voice you want. That way you can make it sound more expressive.
so basically youre saying they're making will you snail in real life
This vid , much like your others on AI, is insanely good. The experiments & demos to support your predictions is my favorite part. Thanks for sharing!
Your videos about AI are top notch of the quality, please keep doing those longer videos every once in a while and I'm praying to youtube algorithm to bless you for them :)
When I spot a topic that
1) I know enough about
2) That deserves it
3) Is not talked about enough
I will.
IMO AI shouldn't be used to replace artists or devs, but should be used to enhance the tools that artists and devs have to both expand and expedite what is possible to create. Cascadeur is a great example.
Very well made video. By far the most in depth and well produced video i have seen on the subject
Wow, that was an amazing video, haven't seen such a good talk in a while. Thank you for making this Jonas
This is the kind of video I needed to help explain this stuff to people.
I'm no one special, in my own special way. But with a video you've made, I can link people to it so they see these points from, not a stranger, but a beloved game dev's view point.
Been sharing this vid in game dev servers I'm in.
this video is so good... I did not expect this level of depth and insight. Amazing job!
Watched it all. Great video - good job :)
This video is insanely good and insightful. Very well made and entertaining as well. Good job! What a time to be alive.
Good video! Well thought out, wise and optimistic. Kudos 😊
Wow. I was worried about this but thought it would happen in the far future. You explaining how it would be an architecture with different AIs working together seems very plausible now that I think about it. What if AI not only makes games personalized to you, but adjusts those games in real time as you play them if it finds something is un-fun for you?
(This comment was totally definitely written by an actual human being.)
These videos are amazing, i actually just had a talk about this exact topic yesterday with someone, you articulated everything i am thinking but in such a better way. Will definitely send this video to them
Noice.
I really liked the final conclusion. Throughout the video, that was the main thing that I was thinking. That I make games because I enjoy the process.
One more thing that I would like to add; the actual process of doing the every little thing teaches you more about it and allows you to make more interesting things.
I believe AI games and Human made games will be like Factory made pottery and Homemade pottery.
Very nice comparison, I feel the same way. I feel like AI generated games will just feel stale and repetitive, like AI art does today. AI art that people like is usually a mish-mash of famous artstyles but looks so samey imo.
@@HeraldOD The reason it looks samey is people in the west seem to prefer more realistic style, and are often limited to one models like Midjourney or Dall-E. With local version of stable diffusion you can train or download any number of specialized models, and combine them as you like. Setting a simple flat color style model to negative can add detail to any image for example, or you could get one model for character, and mix it with another one for pose. I also prefer more abstract style models since they are more expressive and flexible.
I love that analogy.
@@HeraldOD Yet there is room for both.
@@willhart2188 oh cool!!
It is a very nice video yet again! Kudos to you! :)
Thanks for providing feedback to the first draft. Much appreciated! :)
thank you so much for making these large pieces - truly amazing!!
Exceptional analysis of the situation, very reasonable with many great points. Thank you for making this video.
Speaking as someone in the industry and in education I have 2 predictions.
1- AI is fundamentally very bad at meeting particular technical requirements and always will be. It is difficult to wrestle 2d images into a particular style, just imagine trying to wrestle standardize topology or UVs out of a generation model. I think there's a reason that there's no good 'Auto-UV AI' already, which is something us game developers would actually really like! Even when there is one, game production pipelines are more about standard practices than quick generation of content.
2- In the 1950s, chromed metal was very cool, exclusive, and expensive. Anything chrome looked awesome and was desirable. In the 90s and 2000s, it became possible to make identical chromed plastic that in most use cases is indistinguishable from chromed metal. What this accomplished was making people associate chrome with cheapness in general, so even the 'real thing' now looks cheap. Same with 'piano black' plastic, fake wood flooring, and other imitation plastic materials. I think AI will get better at making certain kinds of games and images, and as it become effortless to flood the market with those kinds of games and images, they will overnight become extremely undesireable and 'cheap' feeling to the point that they will be rejected completely. The big corporations want an infinite money spout where they can put in electricity and raw materials and get profit from the other side without labor. I think fundamentally this doesn't work in the way that you don't go out to Tower Records and buy a CD when spotify is cheap and infinite (and mostly not even AI-generated yet). There will be a tiny amount of money to be made on these cheap generated games for a while, they will sell for pennies each or less, and any profit will be had in massive scale until it dries up, and people will rapidly get bored of the kinds of games that AI is good at making. An example of this in games today is mobile games in general, they are so deeply associated with cheapness and gambling that it's difficult to make money with anything on mobile anymore.
As of now, there's no evidence that AI can do everything that is promised, it is a machine, not magic. It can definitely do some things right now, it can probably do some more things in the future, but just like autonomous cars, there are limits to the entire machine-learning paradigm and what it can and cannot do. I think that means that the kinds of games and content it makes will be relatively narrow and very quickly discarded as it floods the market.
If we think about VR, 3d printing, and any other limitless, 'change the world' technology since after the smartphone, all started out huge like this but eventually settled into an interesting and productive niche (except crypto, lol) that generally did not change 'everything'.
great job covering this, thanks.
Your mindset is trully inovative, Jonas! Very insightfull insights! Really really cool
I agree with your take. I probably think it's nearer than you do.
Delighted to see you rediscover Turing computability (machines processing their own code). It can do that much faster than humans. The recursive loops you're talking about can be done in minutes as game speed doesn't matter to computers, it's actually completely abstract concept to machines as a whole (Turing machines can't boil me an egg).
Where the current focus is on text as an input to training, the next AI input training step will be realtime video. Could be a camera feed, game output, mobile phone camera, etc. Then you're learning full motion video. I don't think that's needed for making games, your process loops should be enough but that will be the end game because the data density of video is much greater than text. Naturally, that won't be used in isolation, it'll be text, audio, etc.
Very well put together video!
Incredible video. thank you
Amazing work as always Jonas. I really appreciate what you're trying to achieve with this video. Helping people to adapt to this rapidly changing world is what we need the most right now in my opinion.
Great vid! Interesting look on the topic
Happy you enjoyed it.
Phenomenal insights and thought process. As always :)
Happy you found it enjoyable.
I'm 16 years old studying computer science and maths with the intent to study computer science in university and get my hands dirty in the gamedev industry. I've been playing around with making games for the past 4 years ish and I can't imagine doing anything else. Before watching this video I was really naïve towards this topic... I knew that AI will most likely become the future, and I've played around with chat gpt, but I had no idea it would accelerate so fast. But besides how scary it may seem, I'm definitely still set on the same future for myself, if anything more excited for the rate the indie gamedev industry is going to accelerate due to these tools. Thank you so much for the really well put together information Jonas!
The goals to action approach does seem very much possible, and till now I wasn't so sure how AI will be automating game development, but now it does look like there's a direction to achieve that and it's here I suppose the new developments will gonna take place.
Plus, the last part about enjoying learning was so true. In fact, I realized that for quite some time I have been too much caught up in thinking of what I'm doing right now so will that gonna be worth it in future. Like there's so many things I start learning and I enjoy doing them, but the moment I start thinking of its viability in future (even before ChatGPT happened), I start to question and abandon certain things. Now it is indeed important to priortize one's effort and time and be realistic, but in doing so I was slowly alienating the joy of learning things that I enjoy. Hearing you say about the learning aspect was so crucial for me, and it'll be something I won't forget.
This video was super awesome and helpful to me, and keep up the great work Jonas!
Best video on the topic. Love you man.
Yeah for real! 100%. First 10 minutes of the video I already felt like it was the best way to explain to people that the discussion on AI can be civilized detailed and doesn’t have to remove opinion and bias as this is a discussion of complexity and there isn’t really just one solution to anything here
Great Video!👍
This video changed my mind in many ways on this topic. Thank you.
My thought I wanted to add is that AI outputs are not always perfect. So when you take the approach of agents and actions, each AI output might have some sort of error, and when you have a lot of AI calls that all rely on the previous outputs, the error will propagate and compound. Even if you use AI to check it's own outputs, who's to say the checker wont hallucinate or have errors itself. I could easily see this being a fatal flaw of AI when making anything really complex.
The ways to fix this in my mind is to improve the quality of AI outputs to make them better, or decrease the number of generations you do to decrease the error propagation. Improve the quality of AI is already happening, but defining what 'better' means is a hard problem that I think might have to be solved on a per action basis. So I see this taking a lot of time and effort to get right. Decreasing the number of generations, however, would also likely reduce the quality of the game it makes.
Another way to potentially solve this would be "agent swarms". So instead of having one agent carry out an action you can just have multiple agents carrying out the same task independently from each other, and then voting on the best solution at the end. This is just one example but there are many ideas for such architectures that decrease the rate of error. If you think about it, humans are quite error prone, too. Yet, with the correct checks and balances and a mixture of internal and external feedback, we manage to be quite reliable. We're not all that reliable on our own if you remove all of our tools and our touch to external feedback, though.
My favorite game dev on youtube, inspiring in many ways. Algo comment
I think the biggest flow to correct will be to some agent feedback or something else agent to know when its good enough to stop improving ! In the video when you asked the AI to simulate human feedback, it was only negatives ones and generate a lot of improving step. It will be difficult to prevent it to fall into the inifite improving like human can do xD.
But it's clear that it will happen !
Haha. Yeah, the fact that they were trained on the internet has some interesting side effects, one of them being that they mirror our mental flaws and biases.
genuinely suprised how AI implementation is way further ahead than I thought, thanks for showcasing all this Jonas, often look at twitter communities where people are just noping out about everything AI related but it's better to be hyperaware of the advancements than to be too late when it eventually does take over jobs/influence your hireability.
Damn, awesome video
Great video
great video once again!
Happy you found it insightful. Thx.
Absolutely solid video. It's really easy to see developers you once looked up to on, like, Twitter or whatever, and see them take antagonistic stances one way or another on this. I think at the end of the day it's a defense mechanism and there's some level of emotional charge. Not all of them are dumb (on either stance). Thank you for making this.
The indie game dev scene did not have to deal with ethical discussions a lot before now. This seems like the first real trial how well we can manage the dispute. I hope well.
@@JonasTyroller That's a good take I didn't consider. Feels like many are going through a lot all at once.
I'm definitely already generating code for my game. Granted I'm a beginner and I'm asking it how to accomplish basic tasks, but it's generally pretty solid regardless
I took me a while to start watching this. I thought Ok lets just watch the first 10 minutes, but ended staying for the full 67 minutes. This video should have way more views. Thanks for the insights Jonas
There's a bigger problem that no one addresses. (Excuse me if you do but at 32 minutes in, it feels relevant and like it won't be addressed.)
Work shouldn't be necessary for life. For a long time in history, automating away jobs was fine with our economic model because we introduced new jobs in their place. We've finally started reaching levels of automation where new jobs aren't needed, but our economic model is still fundamentally based on a stupid idea - that every person must justify their existence by working. We shouldn't be stopping automation because it takes away jobs, we should be stopping that a job is necessary for life.
Related: We overproduce so many things and throw away or destroy what isn't sold to keep prices higher for profit. The biggest example is food. We have enough food, but instead of making sure everyone is fed, we destroy masses of it every day to maintain a requirement of payment for food. With housing, it's the same story except instead of destroying the housing, it stays extant but unavailable.
We keep only thinking within a very simplified version of a very old economic system that isn't working anymore.
very true, and worded beautifully.
Great video.
Refreshing take on AI, instead all the "Ai rant" videos. Keep it up!
Really liking these AI videos, they're a nice bit of variety from the rest of your content
Awesome video, Jonas. I could probably step in to many points, but just regarding audio/voice-acting: take Cyberpunk 2077 that used Keanu Reeves for the main character. In the near future Keanu's voice and voice-acting can be taken, automatically translated in writing AND in speech, all using his exact voice/vocals. There are some very impressive AIs out there already that will question the need of translators (in business, politics etc.) very soon. In my field I totally see online meetings coming up where everybody talks in his/her native language and receives audio with the respective persons voice, but in the native language of the receiving person.
This is all very mind blowing.
Thinking this a bit further where meshes and facial expressions can be linked to what the character's are saying, then this will be possible for every language. And if you make even one step further those language and behaviour models will take into account what is happening in the world. There are mods for Skyrim that make baby steps into this world already. Now imagine what an AAA studio can do with this. Just imagine GTA6 could be capable of implementing this kind of stuff, this is next level immersiveness and it all comes from text-based AI (which is the most advanced model already).
TES6, GTA6, ... just imagine.
Such a beautiful video and amazing take on the subject. I almost completely agree with your point of view and the way you look at life
Thank you for making a video that I can just send to people instead of explaining myself every time someone asks me about this topic haha
Thank you so much for this ambitious video. I can't imagine how long this must have taken to make.
Glad to see I'm not the only one thinking exactly this.
Just thought id chip in about the whole "playtesting AI" thing. My concern is that this AI will have to be very carefully balanced to be both bad enough at a game to mimic a human playing it, and good enough to be able to actually complete the game in a meaningful way. Remember also, AIs generally have drastically shorter thinking times than we do, and any AI designed to play games will likely need a variable delay between what happens on the screen and its response. This process of dumbing down the AI just enough to emulate that of a human player is what I think will be incredibly difficult. Especially since no two humans are the same. You will need potentially hundreds if not thousands of different models with varying characteristics to paint an accurate picture of the game from a human perspective. In fact, it might be best in the meantime to just straight up outsource that process to humans and have AI agents play in tandem to analyse the differences between real human playtesters and emulated ones.
Thanks for the vid and your thoughts I was very heartfelt. I share similar feeling in the subject to you.
Thank you very much. While this video was focused on gamdev, it also gives a really good perspective on the future as a whole. For me this opened up a lot of "wow, this will apply to X as well..." kind of realizations. Your concluding thoughts were very well put together, completely agree that once we get to the point where AI can make games on it's own, so much about our relationship to AI and ourselves must have changed, gamedev is the least thing to worry about. I'm lucky I'm only in this for the fun, so it's less of a worry for me, but I feel for everyone who feels threatened, it's not a nice feeling to have.
Very much appreciate your overview, predictions and thoughts on this, Jonas!
I can't emphasize enough how much I needed this video. I was thinking about this subject for a long time
One of the best discussions, even if it was with yourself, on the future of AI in creativity.
Appreciate it.
1:01:20 really appreciate this, it pretty much exactly what i need to hear right now. Thanks man
This is a well-thought out and nuanced level-headed discussion. I appreciate your commitment to keeping the discussion two-sided and adding disclaimers of uncertainty. That said, I completely agree with you.
I think that was an awesome video, very well made.
Regarding the AI playing it's own games to test them, perhaps players themselves could do so. Could be an evolution of something like the current early access. The players could provide direct feedback, but the AI could also analyze the videos of players playing and statistics about their behaviors (how long they spend doing this and pressing these buttons etc etc) and thus do the recursive loop.
Could even be a sort of an ephemeral job of the future in the transitional stage before AI can play its own games. Some sort of platform where someone logs in and gets fed games in development to try. Perhaps the algorithm of such a platform would suit what games to throw each player depending on what they have played and liked before (like the youtube recommendation algorithm).
Yay, definitely another really great video on AI by you. 💙
I wonder how people would feel if public facing AI devs (like the ones writing devlogs, doing marketing etc) would be somewhat designed like "virtual influencers" but created by AI and not agencies in a way that people have it easier to see their games as more personal. Mainly because they can relate to the AI devs with them portraying certain character traits, including flaws and preferences, possibly signature game art styles / game mechanics / etc. Many people are able to build parasocial relationships to artificial people or are even fictophil. They just need the AI to become a character and (for some or many) that may bridge over to the emotional/personal needs they have for (the making of) a video game. I'm sure there will always be people who only want "human-made" games (just as some people only want hand-drawn artworks or hand-sewn gowns), or people who are okay with either but value it differently just because of "who" made it. But for many it could be very blurred lines as soon as you make it easier for them to see the AI as a "person".
That is an interesting thought you bring up there. If people demand "human made stuff" others will likely find a way to trick them by making their AIs more human like and relatable or on the even darker side, by straight up pretending to be a real human. Have not fully thought that through, but sounds like something that might happen.
I feel like a more human AI would possibly do better on the market even with it being fully transparent about being an AI. Because there is already known fictional characters that work and feel relatable to them.
But you are absolutely right, there will be blatant faking of being a human and lying about it too. Some humans scam people faking to be another human. Sure AI is getting better and better at it too. If they make people believe they are "real" and pair it up with evoking emotions that make people more likely to buy their product they could be scamming people just like humans do. @@JonasTyroller
Im glad you are bringing light to this. As someone who is going to be in the games workforce, it was something that worried me when choosing to take it as a proffession. But i chose the fun i get from it over how stable it could be. Luckily having fun is something AI likely imo wont feel within my life.
My only worry is that when I show my game off to the world, no one will see it. Even though i enjoyed it, it will still feel like a kick in the teeth seeing all the crazy good games AI can make and how puney and rubbish game. Even though it wouldn't be, it would be the fact that a 3 year project I could have generated with a single click.
Really well balanced opinion piece. Thanks for taking the time to put down your thoughts while steering clear from the easy clickbaity "hot takes".
This is a beautiful video. Gute Arbeit.
I think there is a bit of a conflation of what people "like" and what gets attention. Algorithms are built for engagement but a large byproduct of that was the emergence of rage farming which leads to division and fighting. I don't think engagement is necessarily a good proxy for fun/liking something which is seemed to be used synonymously. Fun is pretty obscure and data around that is difficult to collect since it is so subjective. It's like asking an AI to write a good song, good is different for everyone and changes over time.
You have blown my mind! I have never thought about ai in this way before
Yes, one of the reasons I made this video is because I see a lot of people still viewing AI as an "information retrieval system" which it is very clearly not where things are headed right now. Happy I could open your eyes to it.
Incredible video
Great!
Jonas is so mature, intelligent, and well-spoken. Thank you for making this incredibly important video
I was firmly in the "they can't tell what's fun" and "we'll always need human designers" camp, and was one of the comments shown but before even getting to that point in the video my opinion was already changed. You bring up a very good point, I don't think emotions are actually that complicated anymore. Especially bringing up how we already have dichotomous methods of determining what's good and what isn't, in the form of reviews, likes/dislikes. Even in the absence of such data, there are already sentiment analysis models that can turn text like this into positive or negative. I can now see that there are huge changes coming to probably all creative industries.
And just for funsies I fed this comment into a sentiment analyzer and it determined that it is "very negative/serious". I don't know if I would say it's that "negative" but I can definitely get behind "serious"
As a follow-up, I feel like right now we're in the same area as the advent of computers, and the advent of cars.
There was a period where digital art was "not real art", or where cars were "killing the equestrian industry" but at the end of the day we still have traditional artists, and we still have equestrians. There will always be "traditional" game developers.
@@blakehanna9543 I agree.
Wow, Jonas. This is video is a masterpiece. So many considerations. I totally agree with your final words: We shoudn't be afraid. We should just enjoy that wec an make money with our hobby and in future there will be even more helpful tools to support us.
Always amazes me how maturely you take all of this. Especially since you're (apparently) that young. You're smart and pragmatic and positive; cheers man.
Very well made video. Sums up perfectly where we are right now.
Sooooo... Squid will start his gamedev career?
Welcome to OpenAI devday. Today we are excited to announce... Squid... The evil killer AI that will kill you all. Oh yeah... and he will also make video games. Have fun y'all!
Great summary, AI is involving pretty fast and whether we like it or not it will be possible to make full games in the next 2-3 years.
Watched this entire video. Great, even-handed points about the coming AI revolution. Games will rapidly become more impressive over the next few years. And all of these devs recently laid off by the big publishers will now be able to compete with super high quality games with the interference of upper management only interested in profits. Yes there will be a lot garbage to sift through at first but excellence will rise to the top.
I think you have hit the nail on the head. We are heading this way as a whole and we need to be reafy. We cant deny it or ignore it. Be ready or be left behind
What an incredible video! I'm genuinely curious about how you manage to accomplish so much, constantly learning, reading papers, and delving into investigations. It's truly impressive. The topics you cover can be quite stressful for me, especially considering the restrictions in my country where the use of GPT and similar tools is limited. This creates a significant gap in development, leaving us trailing behind the rest of the world.
For individuals like programmers and developers, who are deeply involved in these technologies, the impact is even more pronounced. Your predictions are quite plausible, and it's alarming to think that the upcoming changes may not heavily affect those in developed, affluent countries. I'm aware that many people, myself included, have built a modest livelihood around computer skills, and losing such opportunities would pose a significant challenge to our way of life and even survival.
That storywriting around 26 minutes is good stuff. Sounds like a scifi novel as an audiobook
part of the new SAG agreement is limitations on AI and protection of human writers jobs and work, and I think that is something more industries should think about (and why unions are important).
Unions are important, agreed. I would very much like such important decisions and developments be subject to democratic rule.
That being said, this seems like a very complex issue. There is a cost to pursuing progress, but there are also costs for not pursuing progress. For example I feel like the economic incentives for not loosing your job may be tainting the conversation at hand a bit at the moment. Individual interests vs common interests may be another moral aspect to consider here.
I don't actually have any solid opinions or answers on this, it just seems like a fairly complex matter when I glance at it.
this video is rly good
The stages of grief is a great point. I feel and see that happening already in certain circles and it is really inevitable that however AI changes our work and our lives we will in the end have to reach acceptance. It's like the Invasion of the Body Snatchers. You can run, you can fight, but eventually you're snatched.
I really like this discussion. I wish more people made this exact type of video but for there own professions.