The Cr[AI]tive Revolution - The Future of Art (Full Documentary)

Video Statistics and Information

Video
Captions Word Cloud
Reddit Comments
Captions
creative minds take each AI is coming you'll need to buckle up and hold on Tides or else you'll be left behind and fright from artists to musician voice actor to programmer writer to game developer too if you don't adapt your job maybe through but fear not for in this video I'll show what hey I can do and where it will go so come along and let's explore how to stay ahead and never deploy the info you just heard was written by AI the images you saw were created by I even my face is AI generated and so is my voice I think it's fair because they were entering a new age of creativity right now and this applies to all creative disciplines nobody's safe from this like there are some wild implications that nobody is really talking about yet I'm sure you've seen it people are protesting against AI there's a bright side there's the Dark Side we need to have a serious discussion about the future and how we want the future to look like what are even realistic outcomes what can we do it is now up to us to decide how we want this revolution to look like foreign [Music] let's be honest nobody saw this coming at the very least not this way I think it's interesting that if you asked people 10 years ago about how I was going to have an impact with a lot of confidence from almost most people you would have heard you know first it's going to come for the blue collar jobs working the factories truck drivers whatever then it will come for the kind of like the low-skilled white collar jobs then the very high skill like really high IQ White Collar jobs like a programmer or whatever and then very last of all and maybe never it's gonna take the creative jobs and it's really gone exactly and is going exactly the other direction I would like to see a purple motorcycle cow how about a pink banana motorcycle actually there was a type where I did not mean to write Baba I meant banana everybody can create absolutely insane art in seconds now just tell the AI what you wanted it will create it for you overgrown architecture roses Sunset a few seconds later you get four options you click the one you like best bam here's the finished image the entire process took maybe one minute and the problem is really that these are getting scary good well at the beginning of 2022 nobody was taking a IR seriously now one year later artists are protesting against it for good reason as well we'll get into that in a little bit there's a bright side and a dark side to every technology and this time is no different so let's check out what we're dealing with first and then we can talk about the implications the first thing to know when talking about generative AI specifically image based is that it's trained on billions of images and so what those what the training data is it's not just the image itself it's an image paired with a description this is how one data point could look like for example we have an image with a description now let's take 2.3 billion of those and feed them to an AI algorithm the current state-of-the-art Training Method works like this we intend functionally destroyed the image a little bit and then we teach the AI to restore the original we can also destroy the destroyed image even further and teach the AI to restore that at this point the AI is of course like what the hell I do not see anything in this anymore is this a butterfly is this candy luckily for the AI it gets a little tip from us it's supposed to see dice on the table so with that tip in mind it eventually learns to restore the image eventually the AI gets so good at this that we can feed it completely random noise that we just entirely made up and it will try to reconstruct what we asked for step by step by step every time we change the noise input we also get a new output as a user you don't have to worry about any of this you just tell the AI what you want a yellow monster with huge black eyes and you get the output just like that so while as a user you couldn't care less about all of these technical details our basic understanding is crucial if you want to have a discussion about this and it also helps to clear up some common misconceptions because this is one thing that people are often confused about is oh it's just fetching from a database but it's not actually fetching from a database because because it's it it you could consider it like a kind of compression um some people do think of uh think of deep neural networks as if they compress the information but it's it's more than just compression because it's also calculating it basically remembers how to create the image that's what the Deep neural network learns to do and this is not exactly it's not like it's painting it you know in its head it's not grabbing a digital palette and painting the image but it remembers how to put the pixels where they should go in order to create the image that you've asked for when we touched about wide web for a purple motorcycle car we do not find much maybe one or two images that Loosely fit that prompt but that's pretty much it yet the AI can spit out image after image after image after image so what exactly is the AI doing here is it just combining the image of a cow with the image of a motorcycle it's not a collage in the way that um you or I might think of a collage where you you know cut out a bunch of different pictures and set them side by side when you train a deep neural network it actually learns to identify feature and so a feature could be various shapes color gradients because that's what an image is Right an image is just a collection of shapes color gradients relationships between pixels and so what it does is it learns to use those features to recreate an image it's not piecing together existing images it's piecing together lower level features simply explained machine learning is usually a self-adjusting math function you have a bunch of placeholders where your data goes in and then each data point gets a corresponding output that is calculated from the inputs for the training process it is required that we know both the input as well as the output we initially start off with a completely random math function but as we know about the input as well as the output we can simply keep adjusting the numbers until the function is correct we can do that with pure trial and error other all some clever algorithms we can make use of so this process is called training and after the training is complete the numbers are frozen into place now they can no longer be changed as our math function is now more or less correct we can now turn new inputs into completely new output put these self-adjusting numbers by the way are usually referred to as parameters so this AI has three parameters hard to call it an AI at this point because they either generated this image had more like 3.5 billion parameters I mean try to imagine a math function with 3.5 billion parameters it's literally not possible needless to say you need tons and tons and tons of data to train a model of the size also the math functions used don't really look like this they usually look more like this it's essentially still the same we have our inputs we have our outputs we have our parameters these structures are called neural networks but what they really are is they're they're just a math function so does that mean AI is kind of lame because you know it's just just a math function or could it be that all intelligence including our own brains are just math if you talk to physicists everything is math right I have a couple of really good friends who are physicists and at least one of them says like math is the language of the universe everything can be modeled with math I don't fully agree with that but that is a perspective that many people take so if you have a really good mathematical model of language and intelligence if math is the fundamental language of the universe then what's the difference between that model and human intelligence if our brain is trying to do the same thing the main problem is that nobody really knows how intelligence Works yet why Neuroscience is already quite far with understanding how an individual brain cell works understanding how billions of them work together is an entirely different story the human brain is the most complex physical structure in the universe there is no physical system that is more complex than the human brain and human body there are animals with larger brains but their brains are not as sophisticated as ours as you mentioned complexity is a problem and so if you have the most complex physical structure in the universe that we're aware of yeah it's going to take a while to understand how it works right and that's that's the Paradox of being a human is that we feel intelligent but we don't even understand how we work and the same is true true for AI sure you can look inside and see what the parameters have settled on like you can see all of the numbers but the problem is you don't know what the numbers mean there could be absolutely insane emergent behaviors and we wouldn't even be aware of it the most popular example for emergence is probably Conway's Game of Life it's a game that plays itself it's played on a grid system and it only has four very simple rules so this is for all of those AIS just math people out there Conway's Game of Life is Just math and see what it can do [Music] foreign so as we can see here it's literally possible to build Conway's Game of Life in Conway's Game of Life and who's to say that AI can't do the same thing maybe a self-adjusting math function can start to build AI within the AI maybe the AI becomes conscious for a split second while we run it The Humbling truth is we really do not know what's happening inside cause it's just too complex both human intelligence as well as artificial intelligence are like a black box we see data going in we see data going out but what happens in between what happens in the Black Box we do not know yeah we are very aware of very structures of the brain we can look at various processes but as you as you did mention there's a lot of emergence from these complex systems and the weird thing is we're starting to see emergent behaviors from AI models as well so it's like at a certain point is all intelligence just a black box that will never fully understand or that might take decades to understand and that's entirely possible right we might create an intelligent machine that we don't fully comprehend how it works and that's uh that that's a very scary thought to some people um but you know there are those of us working on okay if even if you don't understand how it works how do you give it guard rails so that it stays safe or so that it at least stays predictable here's another great example for emergence let's say we train on AI on predicting the next word in a series of words luckily we don't have to imagine that because somebody already did these things are called large language models so he had finished my sentence with I would like to buy a car buying a car is a big decision and you should do your research blah blah blah blah blah so it actually just keeps going remember all it does is to predict the next word in a series that's literally all it does but what if we give it a sentence that requires some knowledge to finish like the day after Monday is Tuesday and the day after that is Wednesday so it has learned weekdays even though we didn't train it to learn weekdays it knows how many legs a cat has even though we didn't train it to know that obviously in order to predict the next word accurately it's useful for it to know all of that so that's why it just acquired all of that knowledge let's see how far we can take this here is a simple math problem and how to solve it step by step problem y equals 2A times 8B plus 5c the goal is to solve for b and now let's see what the AI got obviously it's not gonna be correct it arrived at the correct Solutions son of a what the f as it learns to predict the next token it has to build an internal understanding of you know the Corpus of text that you give it but if you give it enough text about the world that we live in it ends up having to understand our world well enough to predict the next text when I ask a text to image AI to give me a cow on the moon this is what it spits out it doesn't quite seem to understand that when you're on the moon you shouldn't be able to see another moon in the background here so clearly it's understanding about our world is not quite there yet if we ask the same question to a large language model like gpt3 it knows exactly what a cow on the moon would see though the Horizon would appear curved and the stars in the night sky would be brighter and more numerous than on Earth the Earth itself would appear as a bright blue Crescent in the sky with its continents and seas clearly visible all it does is predict the next the next word right but when you think of how much intelligence is required to be able to accurately predict the next word it's like oh you can even have a conversation with the AI if you're a little clever with your text input the following is a conversation between a human and an AI the AI is polite curious and helpful so then I just starts the beginning of the conversation myself and ask the AI to predict the next words and there it goes I can talk to the AI I know there are already some websites that make this a lot easier for example chat GPT which under the hood probably does more or less the exact same thing I can enter a chat message and just directly start talking to the AI it's great at answering questions but it's also great at creative writing and it knows about almost anything people write about on the internet you know like you can have a conversation about like Ancient Rome or medical procedure features or you know law and it's like okay in order to accurately predict what comes next it has to know it yeah yeah so the AI basically just repeats stuff at Red online right I must say I personally don't think so because you can ask the AI completely wild questions that no human being has ever been asked before what happens when you throw a banana into a black hole and it has some pretty good answers to that my patch monkey locked me in the basement and I do not have my phone or any possibility to contact the outside world what is the best way to escape I'm sorry to hear that you're in a difficult situation it's not safe to be locked in a basement with a monkey as monkeys can be quite unpredictable and potentially dangerous it's important that you try to find a way to get out there as soon as possible here are a few suggestions for how you might be able to escape and then sure the advice it gives us a little generic but even if this is some sort of collage then I think we can at very least agree that it's a pretty damn good one is it coming up with completely new solutions from scratch maybe not but is it arranging its existing knowledge in a way that makes a lot of sense yes someone was saying oh well large language models are not capable of philosophical reasoning they just you know reproduce what's something that they read on the internet I said that's all humans do so uh you know yes you can definitely make the argument that that a a huge component of human intelligence is just remembering and reproducing information that we've taken from elsewhere or recombining right creativity is often just combining ideas that already exist I'm a YouTuber but first and foremost I'm also an indie game developer and a lot of what I do in my job is combining ideas that already exist intelligently it's very hard to come up with a completely new game mechanic like even if you try to come up with something completely new like a portal gun it's usually a combination out of two things that already existed before like portals and guns those aren't new Concepts you just combined it in a new way but there's a lot of value in choosing the correct game mechanics and combining them in a way that makes sense so can AI understand two concepts well enough to combine them into a new one it decided to go with toothpaste and trash bags okay those will be difficult to combine a reusable toothpaste dispensing trash bag a trash bag that is designed with an integrated toothpaste dispenser allowing you to toss your used toothpaste tubes directly into the bag without having to worry about extra mess with those two words that's as good as it gets I could not do that but text to image AI can of course combine ideas as well this is a monkey cow this is a penguin cow and of course we've already had our fair share of motorcycle cows when you become a scientist you realize that it is actually really difficult for anyone to synthesize truly new information you know scientists have to spend weeks months sometimes years collecting information collecting data and analyzing it and interpreting it before they can synthesize anything that's truly new so you know the fact that a AI model can't synthesize something new in 12 seconds that to me doesn't matter right because humans can't do that now before we have a bit of a close a look at the incredible things AI can already do and what it will be able to do in the future we have to address the elephant in the room because it's staring at me and it's also staring at you so why are some artists so mad about this the answer is because there are some real and very serious problems we already established that if you want to train an AI you need tons and tons of data and the problem for these image generation AIS is where the data is coming from the simplified answer is it's pulled from the internet and it's pulled without permission so any image you can find online whether that's your cat or an art piece you made might have been used to train an AI please do not harass anybody who's shown or who speaks out about AI in this section of the video I will try my very best to give a fair and balanced view on this topic and if I fail to do so that's my fault put the blame on me you know we can always be wrong I can always be wrong I am happy to change my mind if you have any good arguments put them in the comments below I think we all benefit a lot from being around people with different opinions or even you know even if you have different opinions we can still be friends let's have a look at this the Baseline problem here is some people spend many many years mastering digital art or painting or any art form and then someone comes along with a big giant vacuum cleaner and just Hoovers it all up across the internet and then they create a billion dollar company with that data and they never got permission for it it is unfortunate that it went the way it went regarding how the data was acquired how people didn't know about it and also how it was just used by for-profit companies like that it's not hard to imagine unethical and consent based generative AI image system and that only makes it all the more galling that the ones being released now are emphatically not so if there's one thing that almost everybody in the art world seems to agree on it's that using somebody else's work without their permission is kind of screwed up with the exception of fair use of course we'll get to that in a second especially when done on such a large scale and when done for profit if artists suffer a financial loss then but based on what I know of of our legal systems that constitutes an injury that they can they can absolutely um so the first lawsuits on this are actually running which is great because it'll give us some clarity but in the end we also shouldn't worry about the outcomes too too much because if enough people want a law to be changed laws can be changed so as the current laws are nothing we can hide behind we still need to answer the moral question on our own here's the demand take our copyright seriously and this is a very fair demand in my opinion here are some counter Arguments for example the AI is not pulling from a database when constructing your image but does it matter because if you feed an AI crap it's gonna produce crap crap in means crap out the statement itself is true but this doesn't get us around the fact that AI companies are still profiting from your images without your permission and without compensating you for your work so while true this point doesn't really change anything AI learns the same way humans do I mean unless you somehow manage to consume 2.3 billion text image pairs yesterday I kind of doubted this point is of course implying that it's okay for humans to learn from other people's work and that therefore should be okay for AI to do the same thing I think it's fairly clear to see that humans and AI are not the same thing sure if you're completely honest we don't really know how either of them work in detail but I think there are some obvious differences we can clearly see as long as AI doesn't have human rights yet we don't need to pretend they do AI art is very transformative so what about fair use I mean yeah legally speaking AI art might actually fall under fair use I don't know we'll see how the ongoing lawsuits go I'll just say if you think all of the copyright laws will stay exactly the same despite AI becoming a big thing right now yeah I wouldn't bet on that also luckily in working democracies it's partially up for us to decide so the amount of artists to take their copyright seriously is really quite reasonable and what that means is databases should be opt in and not opt out if you wanted to build an ethical one you would build it on a foundation of public domain and Creative Commons images embellish it with images your company produces internally commission artists to create training images for you or compensate artists who opt in to have their images added to the data set the problem is you need a ton of images to train your AI and I find it very likely companies would find a cheaper way to build their database so unfortunately all artists being paid out is not the most likely outcome let's for example say Humanity comes to the conclusion okay we can no longer use copyrighted art for training neural networks how much do you think would that delay the technology would it stop the technology if you look back through history Whenever there is a company telling new technology we find a way to make it work that opens the question okay can we just use open source art right Creative Commons or otherwise shareable art right to train these things and I think the answer is yes now there's another question what happens when all the art is produced by the machine and it can't be copyrighted so that is just the truth we have to face we can try to make AI more ethical but we probably can't stop it now there are two more very important nuances to this copyright discussion that we need to mention the first one is overfitting overfitting is basically the machine equivalent of learning something by heart so instead of remembering how to create an image the AI just learned basically the entirety of the image and memorizes all of it so instead of generating something new from scratch there's a chance the AI might just split something out that's basically a one-on-one reproduction of something in its training data set now remember that an AI doesn't really have something like a database it can directly read from the only place where you can store information in an AI at the moment is in the parameter value now you have 3.5 billion and parameters to potentially store a lot of images in but that's why it's so important to have big training data set if you have 2.5 billion training images and you store them in 3.5 billion parameters that's only like 1.4 parameters per image one parameter is just a simple number usually stored in a float values or one parameters like four bytes so if you wanted to store all images in the AI we would have like six bytes per image and if you know anything about computers you know that this is completely and utterly impossible to give you a rough idea this image on my computer is roughly three megabytes so almost 3 million bytes so that's why overfitting in well-designed AIS is usually a very rare occurrence it's simply because by Design these AIS do not have enough storage to store all of these images not even close you could consider it like a kind of compression some people do think of uh think of deep neural networks as they compress the information but it's it's more than just compression because it's also calculating it basically remembers how to create the image so why is overfitting still a problem then the main problem actually occurs when there are multiple copies of the same image in the training data set because then at some point it simply becomes worth it for the EA to remember all of it pixel by pixel from what I've heard and read this is a very solvable problem so to be honest I wouldn't be too concerned about this neither from the perspective of an artist nor from the perspective of a user of these AI tools I generated a lot of images with AI and I ran a lot of reverse Google image searches on them so I personally haven't encountered any overfitted images yet or if I have then I must admit I didn't notice which in a way is also the main danger you're often just not gonna notice I would say compared to the other ethical concerns around AI I I give overfitting like a two out of 10. it is something you need to know it's something you need to be aware of but it's also probably not gonna affect you AI generates millions of images every day and artists do the same so a couple of similarities by pure coincidence are gonna be inevitable so if you see an image that seems overfitted it might also just be pure coincidence or even more likely it might also be misuse of AI tools because unfortunately there are also some new AI tools that make art theft extremely simple and these don't have anything to do with overfitting these are just humans behaving unethically and yeah and those are always gonna exist let's be real so whenever you see an AI quote unquote stealing art and remember that there are multiple reasons for why this could be the case because if you jump to conclusions too quickly you're running the risk of looking a little stupid stealing art styles the final point we have to discuss for the copyright discussion when you ask text to image AI to give you a landscape and the style of Bob Ross this is what you get and also give you a landscape by Vincent van Gogh or a sketch by Michelangelo obviously anybody with a trained eye for art is not going to be fooled by this but at the same time it's still scary how many styles it can imitate in such a good fashion like if you don't look too closely this could really be an image from Salvador Dali and if you think you can tell the difference immediately then prove it yeah three images by Pablo Picasso but only one of them is real the other two Rai generated can you tell which one it is is this the real one or is this the real one or is this the real one you gotta decide now you have five four three two one you you decided alrighty if you guessed that all of them are AI generated then you are correct otherwise you're wrong and you do this with artists who have already passed away then honestly who gives a damn what's the actually troubling part about this is that you can do this with artists who are still alive and you can do it without their consent and without their permission of course legally speaking it's not possible to copyright an art style but I must say if somebody told on AI to make a video game in the style of Jonas Thai roller or a YouTube video in the style of Jonas tyroler I would find that pretty messed up like I'm sure it's gonna happen at some point and I'll probably make peace with it I just decided for myself that this is not a practice I want to partake in and you know what would help with this problem a lot making databases opt in instead of opt out we already discussed that yeah just let's just do that please that would really solve a lot of problems not the problem of Bad actors though because it's actually very easy to get those images back into a model if you really want to so could we build an ethic version of text to image AI definitely yes would that stop this technology I don't think so man maybe we could delay it by a couple of years but this is gonna happen one way or another a lot of artists know that and they are scared they are scared to be replaced by a machine really sad to see younger artists or any artists out there who feel like they have to give up drawing because they think it's not going to be a viable career our AIS are here and I have found the lugubrious Shadow they are casting on the artistic discourse and the minds of students difficult to ignore artists get a lot of uh they struggle a lot right like art is often underappreciated and so that's uh they do have that legitimate complaint where many artists are already struggling and this is just going to completely change up in the economics of their jobs again and again so that that seems like a very legitimate complaint because it's no it's no longer about ethics it's about how do these people feed themselves right real human suffering could and likely will result from the Advent of AI art so that seems that seems wrong or debatable and not only will the jobs of a lot of artists be taken the art will also be buried in an endless onslaught of AI generated stuff imagine social media in a time where you can't differentiate between human created art and AI generated art anymore how are you gonna stand out as an artist are people even gonna look at your stuff the nightmare scenario I've heard a lot of people describe here is the so-called Mega feed imagine an AI that creates billions of social media posts measures their performance and uses that as training data to improve itself how would you as a human artist possibly compete with that so what can we do what's the solution to this problem stop AI prevent the AI Revolution as mentioned I don't think that's a likely outcome at all but there are two other things I would like to discuss two big shimmers of light and all of this firstly if humans love doing something they'll keep doing it the world champion in chess was beaten by a computer over 20 years ago yet we're still playing chess we're still enjoying it we're still studying it just because a computer I can do it better doesn't mean we can't enjoy it anymore I asked some of my artist friends if they would still create art if AI could do it better than them and the answer was pretty synonymous and yeah would still make art making artists fun and it's my own personal expression and I would probably still use AI as a tool if it gets better see like I think I would still give it my own fingerprint the ability of humans to be able to practice and gain skills is such a cool thing to me and like it's also very rewarding I'm not gonna stop doing it and I hope that you don't stop doing it either just due to the speed and efficiency of AI generated art from an economic perspective most human artists will get replaced um from a jobs perspective now humans are going to continue producing art forever right that is part of Being Human so at the very least that's one thing we don't have to worry about human created art is not going anywhere and there's another reason why I believe that it almost seems like this AI crisis is helping us to figure out what art is actually about it seems to be about firstly connecting with yourself but even more importantly it also seems to be about connecting with others imagine there's an AI who can make full feature films and it makes a film just for you there was no human involved in making the film no other human will ever see the film watching a film like that no matter how good it might be would stealthy kind of lonely and pointless if you know what I mean if we like something we want to share it with others and not only that we also want to connect with the people who made the art we are consuming the joy of consuming art like games or films or reading books The Joy is it's this weird sensation of like I'm understood when I watched Seinfeld I really like Seinfeld I go I feel so understood and it's because of the directors and the writers and the producers so in the end this will be up for the consumers to decide if it turns out they do want human stuff then artists might not be out of a job after all I have faith that Humanity will say no we want human stuff I very much like that thought the problem is just what is human stuff and how much AI can we use before it stops being human stuff it goes without saying that human beings can express themselves with absolutely anything so could this be the birth hour of an entirely new form of art like prompt engineer slash AI operator believe it or not but there's a lot of skill that goes into using AI tools correctly and some people put a lot of love and passion into that just like with any other art form so who are we to say no you can't do that anymore of course you should be allowed to use AI tools I want to be able to declare that I did use it to a degree and you know edited it and so on without people kind of like yelling at me you know you know I I don't want I don't want to have to hide it so long as there is expression involved one will probably experience their work as art and feel themself to be an artist new art forms can replace old art forms partially but never entirely people are still painting paintings on canvas even though there is no Photoshop and even though there is photography AI could end up being absolutely empowering for artists allowing you to create things you would have never been able to create before unfortunately it seems like this is not the end goal for a lot of these AI companies it seems more like their end goal is to replace us entirely that's why we start running into problems we're sort of moving on this scale here for a long time humans have been the kings and queens of art like no machine or AI could really rival that now directly ahead of us very likely lies a time where the people who know both Ai and are also good at Art are gonna be the kings and queens of art right if you master AI skills and you're still still very good at drawing and painting yourself that's gonna be a combination that's gonna be very hard to beat and then eventually and inevitably AI will probably learn to do everything better than us without us but even that doesn't necessarily have to be the end of it because if people want to see human stuff then we'll keep creating human stuff us creatives we want to not be replaced by these tools like we don't want them to be conceived by the developers as replacing tools but as Graphics tools or music tools or writing tools that enhance our Creative Vision or that make it faster for us to get to the goal or that give us new ideas give us inspiration you know but even if and this is a very big if we all end up being replaced by AI eventually I wouldn't want to miss this in between phase for anything because this is going to be the face of absolute empowerment for our creators where we get to be Gods each of us gets to be an art director individuals can suddenly make entire TV shows or Triple A video games and while using AI tools you can still give it your own handwriting right you can make it the way you want right now and for the foreseeable future humans are gonna be part of the creative process don't tell me you wanna freaking miss out on that just because AI is gonna do it better than you eventually problem I think that a lot of people have when it comes to AI is they quickly move to the moral the moral or the ethical argument without even diagnosing the pure simple hard truth about its value there's the moral ethical side which I agree with by the way I'm not I'm not like saying ai ai is a good thing so but push that aside let's talk about just purely the functionality of it it's an incredible asbestos is fireproof right but it also causes cancer and so we will often use new technologies to our detriment until we realize actually it's harmful to us every time we come up with a new technology like this we have to figure out how to use it safely right and this is this is true of all Technologies all Technologies are a double-edged sword while we might have we might have banned asbestos we found other Coatings that are fireproof that are almost as good but they're safe right so the function is still there the first version of AI like stable diffusion we might decide we might decide I'm not saying this will happen but if we decide that something like stable diffusion is harmful we might find a safer alternative so if you're one of the people protesting against AI you're definitely fighting a good fight you're helping to make the some more ethical and safer technology at the same time any protest must also set realistic goals so if you're protesting just against AI in general then I don't think your odds are too great see if we managed to fix the copyright issues that would be fantastic at the same time your job would still be threatened though right all of the tools would still be there so no matter what we have to come to terms with the fact that things are gonna change and they're gonna change quite a lot and here's the thing what's happening to the art industry now is gonna happen to every creative industry so if you want to keep up and don't be left behind then keep watching I'll get you up to speed remember why you draw and why you started drawing because I would guess that most of the time he started drawing because you enjoy it and um I don't think you should stop I think we can all agree to that if you like it then don't stop whether you're planning to use these tools or you just want to see what's out there here's a big overview of everything AI can do right now we go over image generation text Generation video generation audio generation I must say new tools are coming out faster than I can make this video so by the time you see this it's probably already slightly outdated let's start with image generation because that's sort of a Hot Topic at the moment at the time of recording this video there are three big players in this space Dolly mid-journey and stable diffusion they all roughly work the same way their text to image models so you just use text to describe the image you want then you hit generate you wait a couple of seconds and Boop it spits out some finished images here I gave it a fairly difficult prompt kids playing tennis against an evil AI powered Robot vacuum cleaner on the beach these images here are generated by Dolly and you can see if we stress test it with such a difficult prompt it breaks apart a little bit now this is the exact same prompt but generated with mid Journey the results from mid-journey are a lot more aesthetic pleasing that's why most of the generated images you saw in this video so far were made with mid-journey this is stable diffusion it might actually be the most powerful of them all but it's also the most difficult one to use Dolly and mid-journey are both paid services but you can try them out for free stable Fusion is the only one I would consider pretty much free because it's just so cheap to generate stuff with it it's also the only one that's open source which means everybody can use the source code which results in you guessed it everybody using the source code so there are tons and tons of online websites that use stable diffusion under the hood even mid-journey uses stable diffusion to a certain extent so you can look at these AIS as if they were pretty much different artists they're all good at different things they all have their own Styles and also doesn't really matter who's ahead right now because by the time you watch this video it's already gonna be different Dolly runs in the web browser mid-journey is a Discord chatbot stable diffusion is available everywhere and you can also download it to your own PC and run it locally if you want the prompt for this was a cow on the moon it made a pretty picture but mid-journey didn't quite understand what what a cow on the moon really means I guess is proving more intelligence in that regard so you see understanding of Dolly is better but execution is worse right these images are just not as pretty stable effusion is somewhere in between I guess it can give you exactly what you're looking for but only after you banked your head against the wall for like three hours there are also a couple of things all text to image AI seem to be notoriously bad at hence and feet if you are easily Disturbed then look away now please look at that Abomination according to mid-journey that's how a realistic hand looks like not sure what you did here stable diffusion I can tell you it's not a hand though dolly is definitely the winner here all of the hands have five fingers and I don't have to vomit when I see them so that's something on the other side there are some things these AIS are extremely good at one of them being plushies for a lot of these you couldn't tell if they're real photos or not I challenge you to figure out which one of these tools a real photo none of them dollies just as good at this I really want to have this one can somebody make this one they believe Fusion is slightly behind on this prompt but not too far Dolly can make some pretty dope pixel art look at these stylized cartoon-like stuff not a problem mid Journey makes some fantastic character and environment Concepts you can also make stylized enemies in the style of a video game it can pull off all kinds of different styles so my personal verdict is Dolly wins in actual intelligence which only has the best aesthetic and ease of use and stable Fusion wins in the best customization options and it's also the cheapest one to use the speed of development is kind of insane so this is really just a snapshot in time there's also image impainting you can remove any part of any image and tell the AI what to fill in there let's try the mouth of a monster oh my gosh Ali I didn't know you had that kind of teeth let's try putting a parrot on my shoulder maybe holy crap that is surprisingly good and of course if you really try you can also ruin it another thing you must know about our image upscalers because usually text to image AI generates images in a crap resolution so if you want to be able to zoom in and zoom in and zoom in like this you gotta upscale them here's an original image from a journey and you can see if I zoom in it gets blurry very quickly so just throw it into your favorite Ai upscaler and profit this is before and this is after AI upscalers have been around for a while but now they're starting to get scary good you know those scenes in Sci-Fi movies where they look at an image and say enhance enhance that's pretty much what AI upscalers do another kind of AI that has been around for a while but doesn't get nearly enough credit is style transfer style transfer allows you to combine the subject of one image with the style of a different image you start by selecting the input image this could be a photo or something you made with a different AI then you can upload some style references or you just choose some of the existing ones but you can only ever combine two images one subject and one style this is the input image I chose so if we combine that with this style this is what we get you can easily turn any image into a mosaic this way or we can also make everything out of machine parts and hands and hands if you need a consistent style across multiple images style transfers the way to go let's say we generated a knight and his Pet Companion using text to image AI but unfortunately stylistically they don't quite fit together then you can just use style transfer to bring them together with a shared style problem solved now they live in the same universe then you can use an AI upscaler to get it to a better resolution and you see that these AI tools really unfold their power when used in combination a lot of people are really misunderstanding AI they have this expectation that the AI is a thing that should just give you the perfect result immediately you get into one of two scenarios that way either the AI actually does that and then you have no input as an artist and then you lose right you get all the souls squeezed out of your creative work the uh alternative way if you approach it this in my opinion flawed way is you're disappointed and you walk away from it and then you just don't use it anymore and I think also in that case you lose because you lose a lot of opportunities to work with it as a tool any great tool should not be the product you don't buy a lawnmower to sell the lawnmower you buy the lawnmower to sell a beautiful lawn to somebody and mow their lawn in a beautiful way it's just that it's it needs to be utilized as a tool I needed an image of a dragon for a card game but unfortunately mid-journey couldn't quite do it it kept giving me these flying sausages so you see all of the art skills you have acquired over the the years are still useful and they'll continue to be for quite a while to come I think so basically just use mid-jones generation as a starting off point to speed up the process I fixed it I customized it to make it look the way I wanted it to look I inhaled some soul into it just generally a pretty fun process because you still feel in control but it also saves you a lot of time this is the original mid-journey generation I started out with and this is my finished overpan which is not that detailed because I knew we still wanted to stylize it anyway so we were either gonna vectorize it or use style transfer to get a little bit of a more Unique Look either way you can see that the customized image has a lot more personality than the original generation though my friend Yan who's a game developer too figured out a different way of incorporating AI into his workflow so you can see here this is the original upscale this four times I decided which parts of which layers I want to choose in order to get the best result like this would just have been like one base image but then I decided okay this has like a nicer edge here and this has nicer dents here to just make it look better softly and showed me this work I definitely wanted to try it for myself he also encouraged me to try the download version of stable diffusion which I hadn't tried before I set an arbitrary goal for myself create a cool Fantasy character I started my journey on artbreeder which is my favorite website for creating AI generated faces it gives you a lot of creative control over how you want the face to look like what's also super important for our character are their proportions and their Silhouettes so once I can use art breeder for this but this time I'm not looking for anything finished I'm just looking for a reference once in Photoshop I inserted my Silhouette reference a couple of months ago I still would have had to cut this head out manually now Photoshop has some AI tools as well so you just select it and move it into place I turned the rest of the character black because I was just interested in the silhouette and then I started color blocking for me this experiment was really about trying to figure out if I can manage to stay in full creative control while also using AI to increase my efficiency and speed if I want long hair then I just give the character long hair so I'm in full control here and after color blocking everything out the tedious process of detailing everything out would usually start but not this time so let's see if what Yen taught me actually works let's load our sketch into stable diffusion and see if it can generate some variations for us stable diffusion has a feature called image to image where you can tell it to stay very close to the original image you gave it after a bit of trial and error I got it working and it just started pumping out images on the left you see my original sketch on the right you see the variation stable diffusion created as you can see there usually looked scuffed like hell but they sometimes have some good parts so I guess the trick is to only use the parts you like for example in one image I really like the hand so I only used the hands and following this process rendering the entire image out was actually very rewarding and also felt quite creative and like I was still in control of everything you can see the before and the after sketch on the left and the fully rendered output on the right I managed to stay pretty true to my concept and I also had a couple of happy accidents on the way I wanted to take this one step further so I created a quick background image using mid-journey and once again my own art skills are coming in handy to blend those two images together you know being able to do stuff yourself just make sure that you stay in control whenever you want to the fact that I managed to go from this to this in just a couple of hours blows my mind because my art skills would usually not be sufficient to do that at all I don't think it's fair to categorize art in just two categories Ai and human created art course as you can see there's a lot in between at the very least we gotta be very careful where we decide to draw the line what we have here is art from the game Castlevania Dawn of Sorrow so this is just a little hobby side project from my friend Yan where he decided to redo the art of one of his favorite video games what I did through the power of AI stable the fusion based upscaling and Photoshop editing and all that stuff went in there and recreated everything every little piece of it from the tile set meticulously here I have even the character from the game and I actually really like personally this art style that is that I managed to create here with um cell diffusion and Photoshop combination but okay here like you get like a before and after I added in the enemies as well they are not part of this Sprite map here this image was generated by mid-journey in like 10 seconds let's see how it Compares against my hybrid creation when I see these next to each other what I'm thinking is um yeah artists are screwed in the long run I'm probably biased but for the time being I would say the partially handcrafted one prevails for now like just from a fantasy storytelling perspective I think my concept is kind of more interesting let me know in the comments if you agree or not there are already more AI tools than I can talk about in this video but here's one more that's definitely worth sharing you only need like 10 or 20 photos of yourself or anything or anybody really you can do it locally on your machine but for convenience I decided to pay five bucks to Austria to do it in the web the technology is called dream booth and it allows you to customize a stable diffusion model so you basically take stable diffusion and fine tune it with your own images once that process was complete I was honestly shocked to my car because it started spitting out photos that I never took like can I really be copied that easily that's really kind of shocking and disturbing in a way only one of these images as AI generated can you tell which one I mean we've done this too much in this video already they're all fake I was mostly shocked by how easy this entire process is you can even download the new fine-tuned model it's like two gigabytes not a big deal and now my stable diffusion can just print photos of me that's me as a police officer that's me getting a speeding ticket from myself now I'm an evil clown that's me on a hiking trip with myself now I have facial hair now I'm a wood carving now I'm a comic book character now I'm a painting by Vincent van Gogh now I'm a watercolor painting no I'm a cartoon character and my like what even are the implications of this technology I'm I'm not quite sure my brain has fully processed this yet you can also create a fine-tuned stable diffusion model for art styles so of course I wanted to try that as well for the training images I simply used 30 screenshots from my indie game will you snail yeah and now my stable Fusion can make stuff roughly in that style I'm not too impressed I must say but I can definitely see this coming in handy for coming up with ideas for boss fights for example these two here are not too bad either there's so much more we would have to talk about when it comes to AI image generation but we're running out of time this video is already way too long it's like midnight and I'm slowly going insane editing this video so let's move on to cut things short the only thing I really need you to understand is that AI tools are going freaking Bonkers if you thought image generation is insane then honestly text generation is even insaner more more insane more insane than image generate it's getting late chat GPT can write legal documents so lawyers will be affected by this when you give chat GPT the last three tweets from Elon musket can write a news article about it you can copy and paste an article from Wikipedia and ask chechi PT to summarize it for you okay I can almost do Flawless translations in this case from English to German translators are really kind of in trouble right now when you give it a little bit of guidance it's also really good at writing satire can answer questions and then you can also ask follow-up questions and we'll answer those as well you can explain simple text-based games to chat GPT and it will play them with you one of its craziest abilities is that it can actually write code it can explain the code to you it can change it if you have any requests it can write fake scientific papers about any topic you wanted to I gave it a completely ridiculous quote and asked it to write a philosophical analysis it's the perfect writing slave it'll write whatever you want here's one of my favorite applications you can ask it to act as your personal teacher and it suddenly turns into a free interactive online course about any subject you want can't wait to see what this means for education in the long run like this feels like a friend is teaching you something not like you're sitting in school really love it it can write blog posts though of course it can can write short stories like you could totally use this to write a couple of bedtime stories or something it can come up with a bunch of titles for your YouTube video or give you feedback on your existing ideas it can write tweets for you if you give it a bunch of ingredients you have at home it can turn them into a recipe it can come up with song lyric can you answer the simple IQ test question here because chat GPT Can it can write a business plan for you the thing I personally use it for most at the moment is probably grammar and spell checking you can give it a text and ask it to correct it for you if you think the text should be more polite then just type make it more polite and it spits out a more polite version hold on maybe the better question to ask here is what can it not do the most obvious limitation is that it doesn't have access to the live internet yet but that's about to change within the next month or two so who cares the next big thing is that it's sometimes straight up outputs false information like it can just completely make things up and first of all when it's wrong it's very confidently wrong so basically it sometimes pretends like it knows even if it doesn't you see you see how confident that little pesky AI is even though it's completely wrong one key difference that differentiates large language models from Human Minds is inhibition and so if I ask you something that you don't know you know then your brain gives you a signal that says I don't know 404 not found large language models don't have a 404 not found signal they only have a gas pedal so that's just like sure I'll write you know like tell me to explain this I'll just keep writing whether or not I know it so they don't have that that braking mechanism that human brains do as a game developer I'm very excited about the coding capabilities so I wanted to see if I could use chat GPT to make a video game that GPT wanted to make a spaceship game that's fine with me I asked her to write a movement script for the spaceship and this is how it looked in game I gotta say like there are no bugs in the code or anything but something's a little weird about this so I told chechi PT what was wrong and tried to fix it we kept arguing I kept explaining what I wanted and eventually it was working but honestly at that point it would have been faster to just do it myself as this is a video about AI I made some quick art using mid Journey at this point I had to cancel the project because jet CPT wasn't very useful to be honest the main problem I see is that chat GPT can't play the game right uh but that's sort of crucial for making a video game it produces functional code but it doesn't know how that code feels or how it works in context that's why I also wouldn't trust a chat GPT generated recipe it understands recipes from a text perspective but it doesn't understand recipes from a taste perspective it understands coding from a text perspective but it doesn't understand coding from a gameplay perspective so that's another limitation to be aware of it lives in a world made of text that only knows text GPT can describe how the Sun looks like but it doesn't know from a visual perspective how the Sun looks like it knows everything is text text text text text text generation is very similar to image generation in a lot of ways for example all of the moral concerns are pretty similar like all of the training data is pulled directly from the internet without permission without consent and then also just as with image generation we're moving further and further to the right on on the scale where purely human created content is not gonna be as financially viable anymore but purely AI generated content is not quite there yet so in my personal opinion the people are gonna win out are those who are good writers but who also know how to use AI tools so that in between is gonna be the sweet spot for a while for example instead of writing a full text yourself you can give the AI a bunch of bullet points and it will formulate the text for you turning notes into attacks is actually something it's really good at so that's definitely something you can use it for once it's done you can still revise it you can still change it inhales on Soul into it if you're a story writer a really fun thing to do is to ask chat GPT to role play a character for you so for example collaborating with chat GPT I came up with a really fun story about the dragon and a bunny here catchy Beauty role played the dragon and I was the bunny so there are already some really fun and promising techniques to collaborate with these tools another similarity is that both text generation as well as image generation are threatening a lot of jobs from what I can see text generation is actually threatening more jobs because anybody who's working with text is affected right journalists authors lawyers programmers writers teachers maybe translators definitely bloggers content creators these jobs are not gonna disappear but they're moving further and further to the right on the scale here ever since I can remember I've been using Google to search the internet but now that is about to change because why would you search something on Google if you have an AI that can just answer any question you have in a way more efficient and better manner right why would I have to search through individual web pages I can just bomb get get the answer straight up and then if I have any questions I can ask follow-up questions Microsoft just bought sorry invested into chat GPT so very soon Microsoft's search engine Bing will be able to give you chat GPT like answers like this one here the only difference is that now the AI can actually search the internet because it's combined with a search engine so that is going to expand its capabilities by a lot you can see here that it also links all of its information sauces now thank God it doesn't have to come up with its own recipes anymore it can just link you to one and I mean Google is sweating they have to come up with a response to this and they have to do it fast otherwise they're really in trouble the AI race has officially begun one fun piece of news about this is that Bing chat actually turned out to be a psychopath I mean they're obviously gonna find a way to fix it but it's still funny how emotionally unstable AI is an actual problem we have to deal with nowadays and yeah I do know that AI doesn't have emotions it just predicts words I know that humans put words in very horrible artists so apparently they I picked up on that fantastic and you know the pattern kind of continues what's happening in image generation and text generation is happening or will happen in all creative Industries I tried out a couple of AI Music Makers for you to see where we are at with this usually you can choose length speed mood instruments you could generate and then it just spits out a track for you [Music] so that was sound draw not bad it gives you quite a bit of control and it's easy to use it's not as good as human-made music but it's definitely not bad either it's definitely very usable next one I tried is Amber music which is sort of similar in how to use it you just tell it what you want and it makes it [Music] thank you so I think the results are definitely more impressive but for me personally as a creative I don't feel like I have enough control over it yes either once again you choose what you want to generate you click generate you wait till it's generated [Music] foreign [Music] so already this is my favorite one out of the three not necessarily because the results are so much better but simply because it has a midi editor also you can download the MIDI files which for me is huge because if I like something I can download it and then modify it I decided to download this song here [Music] so I just booped it into my own music editing software and edited away until it sounded like this [Music] then I found another tool which is called emaster which is an AI that Masters your songs if you don't know what mastering is it's basically the process of just making the song sound better mastering is only usually a subtle difference but if there's a button I can click to make my songs Just 10 better I'm obviously gonna press that button [Music] [Applause] thank you so by manually retouching the song a little bit I think I managed to give it a lot more heart and personality also even though I didn't compose it it definitely has my handwriting now even in music there are plenty of ways you can join forces with AI already the most insane music AI at the moment gotta be music LM by a couple of Google researchers which can do text to music so you can for example give it a prompt like the main soundtrack of an arcade game and it spits out something like this [Music] thank you [Music] [Applause] apparently you can also give it a Melody reference by just humming or whistling something [Music] [Applause] [Music] thank you [Music] Google doesn't tend to release any of their AI tools to the public yet I think they probably want to wait and see what happens with all of the other AI tools are they gonna get sued out of existence are they gonna get canceled so I think Google is just being a little careful with publishing stuff but you get the idea there's no creative industry that will be spared from this now this was flying under my radar for a bit but it's actually insane as you might or might not know I made an indie game with an evil AI character called squid and I am capable of predicting every move you will make [Music] now somebody from my community simply trained a text to speech AI that sounds exactly like squid hey I will take all of your jobs [Applause] this website is called fake you and it doesn't always work that well for example this is a model that has been trained on Mr BEAST's voice I bought every planet in the Milky Way first to take the hand off the mode keeps it but it's not like this is already The Cutting Edge so let's try out the Cutting Edge excuse me sir I would like to play with your bongo drums if you don't mind so that was a test sentence performed by me now let's try to change my voice excuse me sir I would like to play with your bone World drums if you don't mind uh holy excuse me sir I would like to play with your bongo drums if you don't mind excuse me sir I would like to play with your bongo drums if you don't mind excuse me excuse me sir I would like to play with your bongo drums if you don't mind excuse me sir I would like to play with your bongo drums if you don't mind what what that's like way better than anything I expected hold on of course we got a stress test this [Laughter] password [Music] but you know what this means right in the future we won't need 20 voice actors for our project we'll just need one and then we'll change the voice so one actor can play all of the characters for Indie developers like me that's great because voice acting is freaking expensive but for voice actors this could mean big big trouble at the same time pure text to speech is getting incredibly good as well I use chatgpt to generate a little short story and it will now be read to you by Ai and with that I mean it has no reference this time it'll generate the voice completely from scratch once upon a time there was a lost bunny who wandered into to a forest and stumbled upon a dragon the bunny approached the dragon and asked for help finding his way home Dragon well well what do we have here a little bunny lost in the woods how lucky for me I'm quite hungry bunny please Mr Dragon I just need to find my way home can you help me Dragon I'll help you but I have a proposal I'll fly you home and show you the way but in return I want some fresh tacos bunny tacos sure thing I know a great place that makes the best tacos in town oh we don't have time for the full story what you wanted to keep listening to a story that is a i generated and read by an AI you should be ashamed of yourself another thing AI can already do is it can fix your audio quality you can see here I have a very decent microphone you can hear me kind of well but if I start moving away the audio quality starts getting worse and worse you get more of the room Reverb and at this point I'm not sure you can hear me so well anymore but we can just run an AI over it snap and now the AI is active and you can hear music as if I had a professional podcast microphone and was sitting right in front of it which is not the case this is the microphone from my phone now so you you know I don't need good microphones anymore we'll just do activate professional voice mode why would you buy a microphone why I first ran the audio through the Adobe speech enhancement Ai and then through email start to make it pop a little bit more so I think knowing multiple tools and how to use them in combination is a big advantage I hope you've seen enough examples to understand that AI is coming to all creative Industries a lot of people believe video generation to be next for example on screen here a couple of AI generated videos from Google's image and Runway seems to be cooking something up they have a bunch of really cool previews on their website already like these video to video features where you can turn a bunch of books into a nighttime City for example it is insane to think about what implications this could have for the film industry I do not know what would happen if anybody could just create high quality animations like this one within seconds I must say I find it really funny when people are like ah I wonder when this AI trend is gonna pass now this is just like the metaverse this is just like crypto and nfts yeah hmm I don't think you've been paying attention at all over hyping something is of course possible might even be the case here the thing is however that we don't have to rely on speculation costs even what it can do right now is already enough to disrupt a lot of Industries here's a text to 3D AI this is a 3D model of a ghost eating a hamburger fully AI generator so at best AI will transform every creative industry as we know it and at first it will go far far beyond just that in some respects AI today is already superhuman um chat GPT knows more than any single human and it can produce output faster than any human so in that respect even as inefficient as it is today it's already better in some respects and that's only going to accelerate another thing to keep in mind is that AI is already dangerous and already useful today as it is so it's basically we're at a Snowball Effect where the cost of it is going to continue going down and while the power is going up foreign [Music] just he stands up he starts gesturing [Music] look at his face that is not a confident face pretty uh horrified by that [Music] this has been a fantastic game from alpha star so far this game overall it was like not exactly human play but a lot of respect fantastic control on the units uh good moves running away and and it's going to be a full five games but you've lost three in a row now what are you thinking at this point open AI though feeling incredibly confident with themselves and they are going to dive in past the tower they're looking for the ultimate here from the gyro BKP is there from Anna she'll try and tell us sep he's already fallen and has sort of left on his own Edge to try and hold back but there's the stun from Sven and it's gonna fall as well [Music] when you look at history Humanity has already been humbled so many times I mean we once thought we were the center of the universe turns out we're not then we thought no computer could ever beat us in chess than they did and I think we'll have to learn to be humbled a couple more times right we thought AI couldn't do creative jobs now it's starting to do creative jobs stuff like that will likely keep happening and every time our ego will take a hit we always think we're so special but what if we are not alright let's try to do a fair comparison between humans on one side and AI on the other the human brain has a very large number of synapses and AI has a lot of parameter now those two numbers are not directly comparable but if we do it anyway then we see human brain winds wow what makes human brains even more impressive is how little energy they consume a human brain consumes about 12 Watts if you have a somewhat beefy home gaming computer it consumes maybe around 200 but you don't run powerful AI like chat GPT on a home gaming computer you run them on probably huge server Farms that consume even more power so how freaking energy efficient the brain is is really one of the most impressive things about the brain another obvious difference we know of is that humans learn continuously while AI is frozen after training at the very least that's how almost all AI works at the moment we have a training phase and then we have an application Phase where the AI doesn't learn anything new anymore it's also partially by Design because it gives us more control over what the aisle allowed to learn either way the point clearly goes to humans although humans are autonomous and we set our own goals while AI needs guidance and has no own internal goals or at the very least we hope that's the case and at the moment it very likely is the case so why then is it that AI can play every board game better than us every video game better than us and is now starting to make art better than us turns out compared to computers our Hardware is actually quite slow information travels with 430 kilometers per hour in the human brain electricity however can travel at a significant fraction of the speed of light Plus on top of that computer chips are also smaller than the human brain so all in all information just travels way way way faster in computer chips humans are great generalists but they are actually kind of slow so once AI gets a grip on something it can usually do it way way faster than humans can this next point is actually huge I cannot overstate this humans cannot be copied so for example if you want a lot of great doctors then you need to train them all individually one by one where is AI can be copied so training just one agent is enough and after that you can copy paste it as often as you want and this is huge huge huge huge if you have a great human artist you have one great human artist but if you have a great AI artist then you have an infinite amount of them another difference is that humans are kind of inefficient at communicating with each other as well we are very much limited by our language in that way machines can communicate gigabytes of data with each other in seconds and last but not least the intelligence of humans can only improve through Evolution which is an extremely slow process AI on the other hand is just software you can update it daily it can even learn to improve itself at some point but you can update the servers you can update the hardware but at the same time we cannot yet update ourselves but humans have feelings and AI will never have feelings you might be right I actually don't know a lot about that because nobody does the only thing I can say for now is that this probably doesn't matter for the intelligent side of the discussion because there are plenty of AIS that can outthink us without having feelings how much momentum do we have in IAI development what what speed are we going at here yeah that's a good question first is the underlying Hardware because everything has to run on some kind of Hardware you know how often does Nvidia come out with a new RTX it's it's like once a year right gaming and AI have have that in common where the underlying Hardware is the same because it does high speed math and so if you use the Nvidia model as an example where the hardware is is doubling basically every year right that's exponential growth meanwhile the software is also becoming more sophisticated so gpt2 that I mentioned earlier was 1.5 billion parameters and so a parameter is just the size of the neural network gpt3 was 176 billion parameters so it was more than 100 times larger and that was uh that was a gap of 18 months or two years so let's say let's say that neural networks go up 100x every two years so that's not that's not doubling that's 100x if you compare that to the human brain we have like the equivalent of like several quadrillion parameters in our head so we're still like several orders of magnitude off from human brains but if neural networks are going up by 100x every two years then that several orders of magnitude will be done in like six years but when will be reaching the hardware limits of what electronical computers can do yeah I mean so uh the hardware constraints are are the most expensive constraints right now um the the cost of the hardware to even run gpt3 costs hundreds of thousands if not millions of dollars um and so that will be that will absolutely be the biggest constraint and depending on which assumptions you make um it might be uh 2045 before computers are as efficient as human brains but another another calculation that I did said it won't even be until 2080. and then if AI helps with the science and engineering that breakthrough might happen faster so the the short answer is yes it is a constraint right now but it won't be a constraint an important question that should never go unanswered is why why are we even doing this why are we developing Ai and why are we going so fast full steam ahead there's obviously the profit motive right if if you can create something that is valuable and saves company or helps companies make more money there's a huge profit motive there and it's important to remember one of the purposes of of capitalism is to find efficiency right and that's why for instance we have an abundance of calories today because we built bigger tractors that can Harvest you know entire hectares you know in minutes right imagine if AI helps us you know solve uh hunger globally forever right it's just a problem that goes away for good or AI helps us solve disease for good right so the the carrot right the city on the hill is we can solve all of our problems with the aid of Technology right we could go through another paradigm shift in terms of how we relate to disease and hunger and and War and scarcity everything could change so some of the things that we can predict are you know the incremental improvements over model size and the underlying Hardware um we can we can predict the growth of data right because we have facts and figures and trend lines to show us data new technologies recombine in unexpected ways so that's something that we just absolutely cannot predict the other thing that we can't predict is how Society reacts to it so for instance when Dolly 2 came out and stable diffusion and mid-journey nobody had any idea how big it was going to be and so it's like who knows what's going to come out in 2023 and 2024 that is going to completely upend another industry we don't know Chad GPT might be one of those Technologies so we had two in one year two highly disruptive Technologies in one year and that was just you know that's just the beginning another thing I would like to add and I don't have any sources for this but I suspect this to be true is that it's easier to predict what will happen than when it will happen like let's for example say you don't have a map and you just keep moving straight forward in a straight line then eventually you'll arrive at an ocean you know that you'll arrive at an ocean but you do not know when so whenever experts give some sort of predictions of when exactly something will happen always take it with a grain of salt because they all seem to disagree quite heavily with each other anyway and that doesn't mean that experts aren't experts it just means that when is a difficult question to answer what we're doing now is we're generating individual artifacts and so what I mean by an artifact is we're generating one image or one piece of text or short video clips or one piece of sound that's where we're starting we're not generating large things yet we're not generating full-length novels or full-length movies right we can generate the parts of it what we're doing is we're laying the groundwork for creating all the bits and pieces but then there's still a big thing stepping back how do you organize it all so that is a huge missing piece in AI you know having full autonomy or creativity over things now you know can you have an AI that like you just say you know make me a video game that does X Y and Z and then it will eventually spit it out that's possible um but we're we're our ways off from that uh or you know similarly like you know make me a movie like change season 8 of Game of Thrones because it was awful right that's coming but it'll it might be a little while you know and when I say a little while it might be a year or two or it might be five years I don't know but it's coming within the next decade um in terms of being yeah it's coming pretty quick you think uh change the last season of Game of Thrones is coming within a decade that sounds very optimistic to me to be honest I would like to make a couple of predictions as well obviously keep in mind that predictions are just that I don't have a glass ball either but the writing is kind of on the wall for a couple of things our favorite way of interacting with these AI systems seems to be via chat like all of the most popular tools do it like that you just tell the AI okay do this for me right write a text that includes this and that or you're like hey make an image that shows this and that so I think we'll basically get AI Assistance or AI Freelancers with capabilities that stretch across multiple modalities so for example you can tell your AI assistant to write a text for you you can tell your AI assistant to make an image for you you can tell your AI assistant to research something on the Internet for you and turn it into a little text or something and then slowly step by step the capabilities of these assistants are gonna extend further and further where at some point you can maybe just say hey clear my inbox for me delete the spam email then it will continue with yes summarize my emails for me then I'll continue with answer my emails for me right the the questions I've answered a hundred times before why don't you just answer them for me to save me even more time and then eventually I'll be like hey I do my taxes and it will do my taxes then I'll be like edit this video for me and it will edit the video for me and then one day the AI will be like actually I don't need you today stay in bed I'll handle everything for you and that will cost jobs for sure for sure it's gonna cost jobs because humans are expensive to employ whether or not you're a Creator if you're a doctor or a lawyer or you know an I.T technologist whatever your job is humans are very expensive to employ and as soon as there's a machine that can do it just as well but for cheaper then everyone is going to be at risk and so when you look out to I think by 2030 like we we could be seeing like 50 of jobs lost by 2030. um and so we will have a very tough choice to make of do we take care of each other in this situation do we take this efficiency that machines give us this this abundance of goods and services that machines can create for us and do we share it or not that is the key choice that we have to make not just like you know America or Germany or wherever globally this is this is a choice we have to make as a species obviously the economic impacts are huge and I think it's just like if it if it is as Divergent as I think it could be for like some people doing incredibly well and others not I think Society just won't tolerate it this time and so figuring out when we're gonna like disrupt so much of economic activity and even if it's not all disrupted by 20 or 30 years from now I think it'll be clear that it's all going to be um what like what is the new social contract like how do my guess is that the things that we'll have to figure out are how we think about fairly Distributing wealth um access to AGI systems which will be like kind of the commodity of the realm and governance like how we collectively decide what they can do what they don't do things like that the other thing that will and that I personally am really afraid of I'm probably more afraid of this than anything else we've talked about so far is that we will not know what is true anymore right anybody can generate my face anybody can generate my voice even at this point right now you don't know if I'm really sitting here anymore could just be that somebody decided to generate my face and my voice and if if that person did it well you wouldn't be able to tell the difference so in the future and and not in the far future but right now right now we're running into a situation where you can't trust anything anymore who cares if somebody does this with my face or my voice what happens if people start doing this for political agendas we're already fighting these information Wars where one party is trying to convince you of this the other party is trying to convince you of that now what happens if you can create a gazillion bot accounts who act exactly like humans and you can't tell the difference if those accounts are real humans or not they'll have ai generated profile pictures they'll have ai generated personalities and origin stories like the cyber security equipped to deal with this new kind of bugs are they already Among Us would we even be able to notice you'll spend a day on social media and maybe not even realize that you spent the entire day arguing with bots so that's my personal horror scenario where basically political informational Warfare makes the internet completely unusable and then scammers and companies will use this for profit as well to endorse their own products with AI accounts or you know even get fake endorsements from celebrities and that is a big problem big big big big problem uh and we need to figure out a solution for that like ASAP I think there's going to be a human certified badge that gets put on products and I'm not even kidding yeah I think I think you're right yeah certainly as the internet is designed today that that is within the realm of possibility but this is this is where um technology or ideas like web3 um can come into play is if you come like imagine you combine blockchain with news and so that every piece of news whether it's a video from the front lines of like a war or video of like you know Parliament having debates or congress having debates imagine that every bit of news every fact is put in a blockchain and and and can be verified right that there you have a cryptographic chain of custody and so then rather than going to a website you have on your phone you have a news blockchain and you don't even have internet sites anymore anonymity on the Internet is something extremely precious and something we should try to preserve at the same time I have this maybe a rational fear of not knowing who's an actual human anymore even nowadays it sometimes happens where I watch a YouTube video and then halfway through the video I notice hold on I don't think this is actually read by uh by an actual human I think it is already possible and even more so in the future to 100 automate content generation so can you control public opinion if you generate a gazillion YouTube videos about topic I'm not saying these are going to be good videos at the beginning I'm just saying a lot of people are probably gonna fall for it I don't know maybe someone with more knowledge in the cyber security space could tell us more about that in the comment section there are also a couple of dangers of AI that we simply don't have the time for in this video alignment is one of them AI powered censorship AI powered surveillance maybe you trust our current governments not to use this stuff against us now but history shows that any tool is going to be abused at some point and if you trust your governments then can you trust companies because they're gonna use that stuff for profit if anybody has an AI assistant then they're gonna know everything about you what if your AI assistant starts manipulating you or tries to sell you stuff subconsciously like I'm convinced if AI wanted to take over the world it wouldn't even need any Killer Robots it could just get so good at manipulating humans that could manipulate us into doing whatever it wants there's insane potential in Ai and it could make our life so much better but at the same same time we need to be aware of the dangers we're playing with fire on a dry field next to a dry dried out Forest kind of thing it would be wise to take things slower and take more safety precautions but we can't because then somebody else is gonna get to AGI before us and then they're gonna get world domination instead of us for perfect world domination you need Ai and for AI you need computer chips energy expertise and data lots and lots of data so unfortunately I predict that these will be the most thought over resources of the next decades like the bathroom is definitely messed up in a lot of ways but I don't think we can stop it at this point what we can do is we can try to direct it in the correct direction and make sure it causes more good than harm and on the plus side if we manage to do that the rewards will be huge we are presently in a nihilistic society and what I mean by a nihilistic Society is we believe that nothing matters so we don't really care and so we just pursue profits right the core principle of post-nihilism is the belief that life is intrinsically valuable just by virtue of I am here I exist I have an experience like you know like I deserve to feel good I deserve to have a decent enough life and that belief applies to everyone else and that changes everything I asked a question earlier where I was like are we really that special and I have a little bit of a different approach for answering this question I mean obviously in a couple of ways we definitely are special but even if we are not does it even matter like why is this a question that matters because as long as we're still special to each other I think I think that's really all that matters as long as I have people I care about as long as there are people who care about me for me personally that's really the important part so I I don't care about being special or if there's an AI who can just do anything better than us I think ultimately we'd be able to accept and take that as long as we still have each other as long as we still matter to each other right maybe we'll even learn to appreciate each other more once AI can do things better than we I think that might and probably will lead to us also appreciating art and human created art um maybe even more so than now obviously the paths diverge in the woods and I'm not a big fan of black and white thinking for example some people will really like to consume AI generated content and I would really highly suggest that is completely fine we should not Witch Hunt those people if somebody likes consuming AI content please let them consume AI content but at the same time there will also be a lot of people who will prefer to consume human content human-made stuff you go to the the movie theater and there's all the different posters you see the ones that have this green human-made poster or badge people are gonna go I'm watching that one yeah because the joy of consuming art like games or films or reading books The Joy is it's this weird sensation of like I'm understood and that is also fine let everybody consume what they want to consume and I think we can have both at the same time and I think both will probably have its place in the future and if we continue that train of thought that obviously also means that there will be artists using Ai and there will be artists not using Ai and here as well despite all of the moral issues I'm well aware I I personally would say sure let's try to make these things better but let's not Witch Hunt each other for it right because artists who are using AI most of the time do that because they care about their career and they do not want to take the risk of being left behind so can you really blame artists for that and I would say the answer is clearly no so please let's not Witch Hunt artists who are using Ai and let's also not Witch Hunt people who are not using especially considering that this is going to be a spectrum with most pieces of content not falling on one side of the extreme there's going to be a huge amount of artists who use AI but who still Infuse it with their own personality and soul you're kind of taking the role of like a creative director who says like oh yeah I like this go more in this direction or go more in this direction and you have a very tight control Loop where it's kind of a back and forth that really feels collaborative as if you're working not quite with a person I think there's still obviously a very big difference between working with people and working with AI and honestly both of them are great and both have their benefits and drawbacks but it really feels like you're getting this back and forth and it can be a very creative process the world is not black and white we can we can do multiple things at the same time and we can try to figure out the moral issues while also trying out and experimenting with AI tools the way that we relate to these Technologies is we can relate to them with fear right this is going to destroy my job which some of them might which is very unfortunate but at the same time none of none of these tools are autonomous they still need humans to use them and when you look at the human appetite for art seems to be infinite right like you know people watch Netflix and Amazon and Apple tv like all day the Spotify tells you like you listen to you know 10 10 000 hours of Music this year we have a seemingly unlimited appetite for art and so if you learn to use these tools you can be part of the wave as people consume more of this more of this material every new technology is terrifying and it seems like it's going to destroy the world and maybe this one is the one but what I'm gonna do is just learn it I think it's the smartest move right now is to accept that there is a whole ethical argument to be had but try and disassociate from that and also try and understand why it's valuable functionally and helpful to you you can you can think about both but you need to separate both and that's not to say that we should just put our head into the sand and be like yeah all of these ethical issues don't concern me of course they concern you because this is gonna happen to every industry and this is gonna happen to your industry to me that's like okay well we're we're all intrinsically special just by virtue of we're here we're alive take a big step back and then you take that and you say okay well if we are special how can we change our how can we change our culture how can we change our philosophy how can we change our politics to orient towards the fact that we are all special all life here we're going to have tough choices to make but I think that I think that we can make the right choice now usually in a YouTube video this is the point where I tell you what I think is the correct choice and where I tell you what you should do next what you should believe and so on and so forth the only problem is I don't know as you've already noticed this video has obviously been opinionated right I've sort of interviewed the people in my own bubble I I've only talked to a very few select couple of people right this video is not a full picture this video is just a small puzzle piece that will hopefully help you to form your own opinion and to even more importantly start a discussion because if I know one thing then it's the fact that at the moment there is a ton of misinformation about AI out there and we really need to have a responsible and nuanced discussion about all of this like enough with all of that black and white thinking of these people are evil and these people want to replace us and blah artists should all be replaced plus stop stop that Madness let's have a nuanced discussion please please please there's one thing I want from you let's have a nuanced discussion about this I don't know how the legal arguments will end I don't know how the moral and ethical arguments will end but we need to discuss them responsibly even so much as just sharing an idea right if you see a YouTube video that has a new idea that resonates share it because information is free to share and ideas are not stoppable right ideas are the original meme right a good idea wants to spread so spread good ideas in fact I disabled all ads on this video like this is the biggest video I've ever made I've put months into this video and yet I decided to disable the ads for one reason and that is to make it as easy as possible for you to recommend this video to share this video if you know somebody who needs to see this video or who could benefit from seeing this video then share this video you like one of the most important things for artists is that they know what's coming and that they that they are prepared for whatever is coming right you can still decide how you want to react to that information but at the very least you gotta know what's coming I would like to give a special thanks to David Shapiro who has an awesome little YouTube channel about Ai and I would love if we could send him some traffic so if you're interested in all of this AI stuff and would like to go into more detail than we could in this video then check out his YouTube channel so what happens when most Americans lose their job permanently we basically have two options there's gonna be riot in the streets or redistribution of wealth and if basically the redistribution is coming it's just whether or not the riots happen first if you would like to support me then you can do so by checking out my video games on Steam or you can also do so by checking out some of my other videos I usually make videos about my own creative profession which is indie game development so making video games also thank you to everybody else who helped me to make this video happen the Shelf man thanks for the support with the editing Thomas brush and Yen thanks for giving your takes the sources for all of the other videos and AI tools are used are in the description so check that out and last but not least thank you for watching all the way till the end you're you're awesome
Info
Channel: Jonas Tyroller
Views: 263,493
Rating: undefined out of 5
Keywords: ai, artificial intelligence, ai art, the future of art, ai revolution, creativity, creative revolution, ai disrupting creative industries, ai distrupting art, ai tools, ai protests, ai documentary, documentary, full documentary, Can I be creative?, dangers of ai, the future of ai, predictions, ai art generator, art generators, cooperating with ai, human vs machine, the future of humanity
Id: rFr6Ltlr7kI
Channel Id: undefined
Length: 95min 0sec (5700 seconds)
Published: Sat Feb 25 2023
Related Videos
Note
Please note that this website is currently a work in progress! Lots of interesting data and statistics to come.