Amazing talk, best host I have ever heard, so interested in the topic and allowing the guest to talk.❤ the setup is so mint, and the slight angle of the cameras (table)is really refreshing
Prof Fry. Thank you for hosting these and asking excellent excellent questions and pushing these experts to really tell us how they are thinking about these ideas and problems. Kudos to you for keeping these intellectually stimulating instead of letting them turn into a giant ad for google.
I Tried to have a conversation with an arts Professors at uni on AI and arts and was frustrating (by how I handle it, I guess). This video for sure will be my friend to carry on my attempts. I Think that watched something but invite teacher from Uppsala University, would be great.
The Ultimate Turing test: Take a professional string quartet who never made any recording of Beethoven Op 127. The quartet’s 1 violinist, violist and cellist play the slow movement of Op 127 with their eyes blindfolded with five second violinists chosen from some other professional string quartets. Second violinists would play with their eyes open. The quartet members would also play the same with AI, which would produce sound from the second violinist’s position in any technical way and monitor the other players in any way, with microphones, cameras. etc. If the players / listeners did not distinguish AI from the real second violinists in the blind test, the test would be passed.
Question: If I take a childs crayons and replace them with an AI that makes their art for them- have I given them something, or have I taken something away? By giving them access to a technology that interposes it's own structure and interpretation on their creativity have I not taken something vital and alive and reduced it to a moribund statistical mediocrity? We all know the answer to this- but there's just too much money in play for that answer to really matter. All the rest is self serving humbug. The idea that superior text and languge handling ability will result in enhanced control of image generating AI is incorrect because it is based on the idea that words can be used to accurately and precisely define images. If that were true I could send only a written description of my face to a portrait painter and from that text alone they could paint my likeness- is this in fact the case? No- of coure not. Visual artists do not engage with their works via the crude medium of written langauge- they operate on the canvas or the clay or the screen with tools that directly implement their artistic intent - this process is non linguistic in it's very essence. Reading a text descrption of the Mona Lisa is not seeing the painting- this should be obvious- just as it should be obvious that no matter how sophisiticated an AI's understanding of written langauge may be this will not enable a text prompt to replace the subtle and non linguistic relationship between the visual artist and their tools. For this reason image generators are doomed to forever fall short, because they fill that gap between the words of the prompt and the intent of the user with their own very special brand of dross- derivative statisical padding that is already becoming a visual cliche- we already talk about the 'AI Art' look and as more and more of this 'art' is made it's presence will become more and more apparent. Machines cannot be creative because creativity begins with intent. The perverse irony of the LLM is that it has superhuman language capability combined with a total absence of intent- they will never create anything truly novel or inspiring because there is nothing they desire to communicate- It's almost too symetrically ironic that machines so capable of fashioning words have nothing to say. Also can we please stop pushing the bizzare notion that a technology defined by autonomus action can at the same time be a perfect vehicle for self expression? Surely it should be obvious that AI is the very antithesis of unique personal expression. We don't need AI for that, we can pick up a paintbrush or a musical instrument. The entire value of AI is precisely that it is NOT under our complete control- and if that is true then whatever we create using AI is at best a collaboration and at worst a parody. But even that best interpretation is flawed because if everyone is filtering their unique creativity through the same black box then the outcome is not more diversity but convergence- there is no such thing as an unbiased AI- so if everyone is using the same handful of biased AI's as a conduit through which they seek to express their own unique creativity can we not see the lnevitable consequnce? Yes that's right- everyone's works start to exhibit the same biases that are inherent in the machines they are using to create that work- the result is not enhanced diversity of expression but the opposite- a subtle yet pervasive sameness creeps into the mix- we end up with a monculture in which the very expression of ideas themselves is 'democratised' via the dead things to which we have ceded our own vitality. For the avoidence of doubt i do not think AI will do anything to enhance the creativity of our culture- on the contrary I think it's likely limpact will be to solidify and perpetuate the past upon which it has been trained, and as a result it will exert a malign and deadening influence on the future.
Given the potential risks you've outlined, do you think it's possible to create an AI system that is truly creative and innovative, or is the very nature of AI fundamentally at odds with the concept of creativity?
@@laviefu0630 I think we need to distinguish between an AI that is capable of true self expression and what we have today. I don't think anyone can say that in the future genuinely creative artificial intelligence could not be created- that is an AI that actualy had it's own unique view of the world that it might then express in a creative way. But this is very different to the idea being presented in the video, which is that current generative AI can somehow act as an unbiased conduit for human artistic expression- as if the AI were a totaly transparent meduim through which the intent of the Artist is faithfully transmitted. Not only is this idea not true-due to the inevitable biases that exist in any AI- but it's also a self contradiction. AI has value precisely because it is NOT completely under the control of it's user so- by definition- it cannot be a way for that user to express his own ideas in a pure and unfiltered way- everything he makes using the AI is inevitably going to be distorted and shaped by the AI itself. And in a scenario in which a small number of models dominate the landscape this means that everyone's work will share the same biases and thus will tend to converge toward a style of output that is in part dictated by the limitations and biases of the models themselves. This is why it's possible to talk about a 'Midjourney' image as distinct from a 'Dall-e image', each of these AI's exhibit biases and continuities in their outputs that make them identifiable. Add to this the UI problem where we are trying to use words to define images and another form of converegance arises- because in any given domain the lexicon avialable is going to be limited. For example If I want to create an image of a dragon I am going to need to include that word in my prompt- and same applies to anyone else who wants to generate a picture of a dragon. So we end up in a scenario in which not only are we using the same AI-with all it's inherent biases- we are also constructing prompts by drawing from a limited number of appropriate words and phrases that are relevant to the image we are trying to make. As a result current generative AI will not create divergence and increased artistic variety but convergence and increased homogeneity.
Think of it this way- your daughter creates two images- one she draws with her own hand using crayons and paper- the other she made by typing some words into a text box that a machine interpreted to produce an image. If you had to destroy one of these images and could only save one- which one would you value the most? Which would you choose to keep? Or another example- a man at his daughter's birthday party reads a moving poem about his love for her and everyone else at the party is pleased. Then they discover that the man did not actually write the poem himself but entered a prompt into Chat GPT, which then produced the poem for him. Now everyone is unhappy, feeling that the emotions the man expressed in the poem were perhaps contrived and not genuine. Why? The answer is agency- we value emotional and creative expressions because they seem to emanate from the individual themselves- they seem to reveal something personal and genuine about them. By outsourcing our creativity to machines we lose this authenticity, this sense of connection. AI has one defining quality that no other technology hitherto has had- Autonomy. Unlike all preceeding technologies AI has the ability to act on it's own to some degree. That is why it has value- it can write a poem or make an image by itself with minimal input from it's operator. But to claim that a technology defined by it's autonomy can also be a meduim of unfiltered self expression and authenticity is a contradiction, because the thing that makes that technology valuble and useful is the degree to which it is not under our direct control but controls itself. So the very best that can be said of a work of Art created using AI is that it's the machines interpretation of what it understood by what we typed into it's interface- and that AI itself is a biased and limited interpreter of our intent. The result is an echo of an echo of an echo of the thing we intended to create when we typed those words into the machine. You know this- which is why you would keep the crayon drawing and throw away the AI prompted garbage.
AI art is in an uncanny valley, where the AI can sometimes draw/write/sing extremely well but it has very little understanding of what it's making. Without understanding, there can't be any meaning behind the art at all, and the AI will make tons of ridiculous mistakes. Then once AGI arrives, the art will have tons of meaning, but we'll have such a huge quantity of AGI art that we won't even know what to do with it all.
But, as far as I see it, the meaning is with the person experiencing the art. A simple example: Many enjoy the beauty of a sunset, and although it might not be regarded as art per se, it can still evoke deep emotions. AI images that evoke similar experiences will also have an emotional meaning in the eye of the beholder - the AI does not need to understand them for it to be successful.
Here's my take on all this hyper focus on AI's creative abilities (Sora, Suno, Flux, etc): It's all a giant distraction that will prove to be utter fluff once the TRUE capabilities are unleashed. People either 1) have no clue or 2) don't *want* to have a clue re: what is really coming- how unthinkable the changes will be. Ask yourself- why SO much focus on a field that come on, typically gets a societal sneer: creativity, artists, etc. (I am one too, but I still get how the world works). The answer is it's all just shiny, pretty distractions which will be *less* than meaningless once true AGI, and certainly ASI is here. In short- we have NO IDEA WHAT IS COMING- only that we will very likely not recognize this current world- within 5-10 years MAX.
"Art... is a deeply human endeavor for humans by humans." So what would we think of art created by an extraterrestrial species? Obviously we couldn't react to it as the alien species did, but neither can we react to Gilgamesh exactly as Mesopotamians of 4000 years ago did or the Lascaux paintings as people did 17,000 years ago. Still, the alien art would be interesting, surely? And we would recognize it as art. So sticking "human" into the definition of art is a dead end; it denies the likely outcome that we will see nonhuman (AGI) art created within the next 50-100 years.
Nice talk... But can any of your AI's write a 60k word chick lit novel, that-- without human editing-- a woman reader (1) couldn't put down, and (2) once finished, could not stop sniffling? It's a simple Turing test that no AI today can pass, nor, imho, could pass in the next 25 years, if not 50 or more...
Mmmm, sorry not convinced. The smell of gold drives good people to bad things. Until this truly amazing technology is more "open" and free of the race to market mentality, then I fear for its future.
It's not so much that as that these engineers can have the best of intentions and all the passion in the world for their craft, but they're not the ones making the decisions about how the technology is used. It will be the C-suite that turn a potential utopia into a hellscape.
Amazing talk, best host I have ever heard, so interested in the topic and allowing the guest to talk.❤ the setup is so mint, and the slight angle of the cameras (table)is really refreshing
Whybis that wngle refreshing just curious? And what does the setup being mint mean? Thanks for your interesting comment.
Prof Fry. Thank you for hosting these and asking excellent excellent questions and pushing these experts to really tell us how they are thinking about these ideas and problems.
Kudos to you for keeping these intellectually stimulating instead of letting them turn into a giant ad for google.
This was a pretty remarkable talk. Doug Eck is one of the most thoughtful people working in AI right now.
I Tried to have a conversation with an arts Professors at uni on AI and arts and was frustrating (by how I handle it, I guess). This video for sure will be my friend to carry on my attempts. I Think that watched something but invite teacher from Uppsala University, would be great.
The Ai host voice can be used for Ai voice assistant. Her voice is unique in tone and sound. 😎💯💪🏾👍🏾
It would be Exceedingly great to interview sundai puchai!
whatsinmy AI fixes this. AI: Your New Creative Muse?
Incredible speaker and host. WOW
The “prompt whisper” Douglas mentions is like a writer for the J. Peterman catalog! From Seinfeld!
Jaron Lanier as a suggestion for a guest. Thank you.
Magenta was a sneak peak as a jamming partner. Are you aware of any gen AI projects that function as a live AI jamming partner?
Interesting!
The Ultimate Turing test:
Take a professional string quartet who never made any recording of Beethoven Op 127.
The quartet’s 1 violinist, violist and cellist play the slow movement of Op 127 with their eyes blindfolded with five second violinists chosen from some other professional string quartets. Second violinists would play with their eyes open. The quartet members would also play the same with AI, which would produce sound from the second violinist’s position in any technical way and monitor the other players in any way, with microphones, cameras. etc.
If the players / listeners did not distinguish AI from the real second violinists in the blind test, the test would be passed.
Cool!
Did anyone notice how crowded the balloon baskets are? Not sure if a balloon would host a party like this?! 😶🌫
Question: If I take a childs crayons and replace them with an AI that makes their art for them- have I given them something, or have I taken something away? By giving them access to a technology that interposes it's own structure and interpretation on their creativity have I not taken something vital and alive and reduced it to a moribund statistical mediocrity?
We all know the answer to this- but there's just too much money in play for that answer to really matter. All the rest is self serving humbug.
The idea that superior text and languge handling ability will result in enhanced control of image generating AI is incorrect because it is based on the idea that words can be used to accurately and precisely define images. If that were true I could send only a written description of my face to a portrait painter and from that text alone they could paint my likeness- is this in fact the case?
No- of coure not. Visual artists do not engage with their works via the crude medium of written langauge- they operate on the canvas or the clay or the screen with tools that directly implement their artistic intent - this process is non linguistic in it's very essence. Reading a text descrption of the Mona Lisa is not seeing the painting- this should be obvious- just as it should be obvious that no matter how sophisiticated an AI's understanding of written langauge may be this will not enable a text prompt to replace the subtle and non linguistic relationship between the visual artist and their tools.
For this reason image generators are doomed to forever fall short, because they fill that gap between the words of the prompt and the intent of the user with their own very special brand of dross- derivative statisical padding that is already becoming a visual cliche- we already talk about the 'AI Art' look and as more and more of this 'art' is made it's presence will become more and more apparent.
Machines cannot be creative because creativity begins with intent. The perverse irony of the LLM is that it has superhuman language capability combined with a total absence of intent- they will never create anything truly novel or inspiring because there is nothing they desire to communicate- It's almost too symetrically ironic that machines so capable of fashioning words have nothing to say.
Also can we please stop pushing the bizzare notion that a technology defined by autonomus action can at the same time be a perfect vehicle for self expression? Surely it should be obvious that AI is the very antithesis of unique personal expression. We don't need AI for that, we can pick up a paintbrush or a musical instrument. The entire value of AI is precisely that it is NOT under our complete control- and if that is true then whatever we create using AI is at best a collaboration and at worst a parody.
But even that best interpretation is flawed because if everyone is filtering their unique creativity through the same black box then the outcome is not more diversity but convergence- there is no such thing as an unbiased AI- so if everyone is using the same handful of biased AI's as a conduit through which they seek to express their own unique creativity can we not see the lnevitable consequnce?
Yes that's right- everyone's works start to exhibit the same biases that are inherent in the machines they are using to create that work- the result is not enhanced diversity of expression but the opposite- a subtle yet pervasive sameness creeps into the mix- we end up with a monculture in which the very expression of ideas themselves is 'democratised' via the dead things to which we have ceded our own vitality.
For the avoidence of doubt i do not think AI will do anything to enhance the creativity of our culture- on the contrary I think it's likely limpact will be to solidify and perpetuate the past upon which it has been trained, and as a result it will exert a malign and deadening influence on the future.
Given the potential risks you've outlined, do you think it's possible to create an AI system that is truly creative and innovative, or is the very nature of AI fundamentally at odds with the concept of creativity?
@@laviefu0630 I think we need to distinguish between an AI that is capable of true self expression and what we have today. I don't think anyone can say that in the future genuinely creative artificial intelligence could not be created- that is an AI that actualy had it's own unique view of the world that it might then express in a creative way.
But this is very different to the idea being presented in the video, which is that current generative AI can somehow act as an unbiased conduit for human artistic expression- as if the AI were a totaly transparent meduim through which the intent of the Artist is faithfully transmitted. Not only is this idea not true-due to the inevitable biases that exist in any AI- but it's also a self contradiction.
AI has value precisely because it is NOT completely under the control of it's user so- by definition- it cannot be a way for that user to express his own ideas in a pure and unfiltered way- everything he makes using the AI is inevitably going to be distorted and shaped by the AI itself. And in a scenario in which a small number of models dominate the landscape this means that everyone's work will share the same biases and thus will tend to converge toward a style of output that is in part dictated by the limitations and biases of the models themselves. This is why it's possible to talk about a 'Midjourney' image as distinct from a 'Dall-e image', each of these AI's exhibit biases and continuities in their outputs that make them identifiable.
Add to this the UI problem where we are trying to use words to define images and another form of converegance arises- because in any given domain the lexicon avialable is going to be limited. For example If I want to create an image of a dragon I am going to need to include that word in my prompt- and same applies to anyone else who wants to generate a picture of a dragon. So we end up in a scenario in which not only are we using the same AI-with all it's inherent biases- we are also constructing prompts by drawing from a limited number of appropriate words and phrases that are relevant to the image we are trying to make.
As a result current generative AI will not create divergence and increased artistic variety but convergence and increased homogeneity.
My daughter enjoys using an app on my phone to create digital images, is the app sapping away her creativity or empowering her?
Think of it this way- your daughter creates two images- one she draws with her own hand using crayons and paper- the other she made by typing some words into a text box that a machine interpreted to produce an image. If you had to destroy one of these images and could only save one- which one would you value the most? Which would you choose to keep?
Or another example- a man at his daughter's birthday party reads a moving poem about his love for her and everyone else at the party is pleased. Then they discover that the man did not actually write the poem himself but entered a prompt into Chat GPT, which then produced the poem for him. Now everyone is unhappy, feeling that the emotions the man expressed in the poem were perhaps contrived and not genuine. Why?
The answer is agency- we value emotional and creative expressions because they seem to emanate from the individual themselves- they seem to reveal something personal and genuine about them. By outsourcing our creativity to machines we lose this authenticity, this sense of connection.
AI has one defining quality that no other technology hitherto has had- Autonomy. Unlike all preceeding technologies AI has the ability to act on it's own to some degree. That is why it has value- it can write a poem or make an image by itself with minimal input from it's operator.
But to claim that a technology defined by it's autonomy can also be a meduim of unfiltered self expression and authenticity is a contradiction, because the thing that makes that technology valuble and useful is the degree to which it is not under our direct control but controls itself. So the very best that can be said of a work of Art created using AI is that it's the machines interpretation of what it understood by what we typed into it's interface- and that AI itself is a biased and limited interpreter of our intent.
The result is an echo of an echo of an echo of the thing we intended to create when we typed those words into the machine.
You know this- which is why you would keep the crayon drawing and throw away the AI prompted garbage.
Great!
the timecodes are out of sync
AI art is in an uncanny valley, where the AI can sometimes draw/write/sing extremely well but it has very little understanding of what it's making.
Without understanding, there can't be any meaning behind the art at all, and the AI will make tons of ridiculous mistakes.
Then once AGI arrives, the art will have tons of meaning, but we'll have such a huge quantity of AGI art that we won't even know what to do with it all.
But, as far as I see it, the meaning is with the person experiencing the art. A simple example: Many enjoy the beauty of a sunset, and although it might not be regarded as art per se, it can still evoke deep emotions. AI images that evoke similar experiences will also have an emotional meaning in the eye of the beholder - the AI does not need to understand them for it to be successful.
It's pretty amazing what AI can do when you give it a theme, for example; ua-cam.com/video/rzU9Vg3R2og/v-deo.html
Here's my take on all this hyper focus on AI's creative abilities (Sora, Suno, Flux, etc):
It's all a giant distraction that will prove to be utter fluff once the TRUE capabilities are unleashed. People either 1) have no clue or 2) don't *want* to have a clue re: what is really coming- how unthinkable the changes will be.
Ask yourself- why SO much focus on a field that come on, typically gets a societal sneer: creativity, artists, etc. (I am one too, but I still get how the world works). The answer is it's all just shiny, pretty distractions which will be *less* than meaningless once true AGI, and certainly ASI is here.
In short- we have NO IDEA WHAT IS COMING- only that we will very likely not recognize this current world- within 5-10 years MAX.
Fry's are Banned in France for their Super Creepy Style. Like '60's Giant Polka Dot Bikinis.
The way this guy talked about the player piano made me mad. Can't explain why.
"Art... is a deeply human endeavor for humans by humans."
So what would we think of art created by an extraterrestrial species? Obviously we couldn't react to it as the alien species did, but neither can we react to Gilgamesh exactly as Mesopotamians of 4000 years ago did or the Lascaux paintings as people did 17,000 years ago. Still, the alien art would be interesting, surely? And we would recognize it as art. So sticking "human" into the definition of art is a dead end; it denies the likely outcome that we will see nonhuman (AGI) art created within the next 50-100 years.
anthropocentrism cope
How am I supposed to not feel outraged and depressed some stupid AI is more creative and talented than I am?
Nice talk... But can any of your AI's write a 60k word chick lit novel, that-- without human editing-- a woman reader (1) couldn't put down, and (2) once finished, could not stop sniffling? It's a simple Turing test that no AI today can pass, nor, imho, could pass in the next 25 years, if not 50 or more...
Mmmm, sorry not convinced. The smell of gold drives good people to bad things. Until this truly amazing technology is more "open" and free of the race to market mentality, then I fear for its future.
It's not so much that as that these engineers can have the best of intentions and all the passion in the world for their craft, but they're not the ones making the decisions about how the technology is used. It will be the C-suite that turn a potential utopia into a hellscape.
@@raggedcritical True dat.👍
Big brother Google telling people they can't use google media tools for politico express. 🤐 This is why Elon Musk is my hero.