Look, I know how deep learning works. I know it doesn’t literally copy the images from the training dataset. But the entire point of supervised learning is to burn information about the training data into the weights and biases of a neural network in such a way that it generalizes over some domain, and can correlate the desired inputs with the desired outputs. Just because you’re using stochastic methods to indirectly reproduce the training data (of course, in a way that’s invisible to humans because of the nature of deep neural networks), doesn’t suddenly erase the fact that the only substance an AI has to draw from is the training data itself.
I think it’s really oversimplifying how humans make art to say that it’s just going over references and creating something new from it. As humans, we are influenced by the work we’ve seen, but because of our unique experience we inject something completely new into any art we make, no matter how derivative. An AI is incapable of doing the same (except for some random noise), because literally all it’s capable of doing is composing together information that has been baked into its weights and biases. It’s not like when you ask a generative AI to make something for you, it will decide to get funky with it. All it’s doing is drawing from the information that has been baked into it.
Just like how ChatGPT doesn’t actually understand what it’s saying because it’s only capable of predicting statistical relationships between words one word at a time, and has no model of meaning, only of how words go together in the training data, AI that generates images doesn’t actually know what it’s making or why. That is totally different from humans who make a piece of art step by step and do so very deliberately.
Edit: I recommend you watch this video by an astrophysicist who works with machine learning regularly, she makes my point a lot better than I can. https://youtu.be/EUrOxh_0leE
How would you classify those “experiences” people have that influence their art or work other than data? Honest question.
And very interesting video. I still don’t 100% align with this perspective, cause I feel it tries to give something extra to the brain than materiality. While I’m no material reductionist, I don’t think our human creativity is “special” or “metaphysical”. It’s our brain, and it’s physical. It can be physically replicated.
I think AI will have a “soul” or consciousness because I think everything already has it. It’s just our human biology that allows this consciousness to be self-experiential and experience other things, such as thoughts and ideas and feelings. A rock doesn’t have those, but it has a “soul” or consciousness. But I feel I digressed a lot lol
Also to make it clear, I don’t think AI exists already. I think these models and developments we have are part of AI though.
I don’t disagree that experiences are data. The major distinction I’m making is that the human creative process uses more than just data, we have intention, aesthetics, we make mistakes, change our minds, iterate, etc. For a generative AI, the “creative process” is tokenizing a string, running the tokens through an attention matrix, plugging that into a thousand different matrices that then go into a post processing layer and they spit out an image. At no point does it look at what it’s doing and evaluate how it’s gonna fit into the final picture.
As for the rest of your reasoning, I neither agree nor disagree, I think we just don’t have the same definition of consciousness.
I feel your description of what a generative AI does is pretty reductive. The middle part of “plugging the ‘token’ through thousands of different matrices” is not at all well understood. We don’t know how the AI generates the images or text. It can’t explain itself.
And we have ample research showing these models have internal models of the world and can have “thoughts”.
In any case, what would you say consciousness is? This is a more interesting question to me tbh.
Well I don’t see the problem, AI can’t explain itself but it’s nothing more than matrix multiplication with a nonlinearity. Maybe you use a Fourier transform and a kernel instead of scalar weights for a convolutional neural network, maybe it has state instead of being purely feed forward, but at the core of it all you’re doing is multiplying matrices and applying a nonlinearity. I don’t know what you mean that we don’t know how it generates images and text. It’s literally just doing the thing it was programmed to do?
What research? I’d like to see some evidence that these models “think,” given that the way every LLM I know of works is by generating a single word at a time. When you ask a GPT how to bake bread, and the first word it outputs is “Surely!” it has no clue what explanation it’ll start giving you. In fact, whether or not it chooses the exact word “Surely!” as the start of the response has a cascading response on the rest of the output. Then, as I had said earlier, LLMs don’t see anything more than the statistical correlations between words. No LLM knows what gravity is, but when you ask it why things fall down it has enough physics textbooks in its training data that it can parrot the answer from there.
One of the ways I really broke down the idea that GPTs have any model of thought is playing this game. If AI had any actual model of meaning, it would understand security and it would understand not to just tell the player the password. Instead, it will literally blurt it out if you do as much as ask it for words that rhyme. You don’t even need to mention “password,” the way GPT works means that if it detects a lot of weight on a certain word in its previous prompt (which naturally would’ve emphasized the password), it’s almost guaranteed to bring it up again. I know it’s not exactly a hard proof, but it is fun.
As for your last question you’re out of luck because I’m actually just a Catholic lol, not a lot more to say than I believe that there is a metaphysical nature to human experience connecting us to a soul. But that’s a completely unscientific belief to be honest, and it’s not a point I can argue because it’s not based on evidence.
Look, I know how deep learning works. I know it doesn’t literally copy the images from the training dataset. But the entire point of supervised learning is to burn information about the training data into the weights and biases of a neural network in such a way that it generalizes over some domain, and can correlate the desired inputs with the desired outputs. Just because you’re using stochastic methods to indirectly reproduce the training data (of course, in a way that’s invisible to humans because of the nature of deep neural networks), doesn’t suddenly erase the fact that the only substance an AI has to draw from is the training data itself.
I think it’s really oversimplifying how humans make art to say that it’s just going over references and creating something new from it. As humans, we are influenced by the work we’ve seen, but because of our unique experience we inject something completely new into any art we make, no matter how derivative. An AI is incapable of doing the same (except for some random noise), because literally all it’s capable of doing is composing together information that has been baked into its weights and biases. It’s not like when you ask a generative AI to make something for you, it will decide to get funky with it. All it’s doing is drawing from the information that has been baked into it.
Just like how ChatGPT doesn’t actually understand what it’s saying because it’s only capable of predicting statistical relationships between words one word at a time, and has no model of meaning, only of how words go together in the training data, AI that generates images doesn’t actually know what it’s making or why. That is totally different from humans who make a piece of art step by step and do so very deliberately.
Edit: I recommend you watch this video by an astrophysicist who works with machine learning regularly, she makes my point a lot better than I can. https://youtu.be/EUrOxh_0leE
How would you classify those “experiences” people have that influence their art or work other than data? Honest question.
And very interesting video. I still don’t 100% align with this perspective, cause I feel it tries to give something extra to the brain than materiality. While I’m no material reductionist, I don’t think our human creativity is “special” or “metaphysical”. It’s our brain, and it’s physical. It can be physically replicated.
I think AI will have a “soul” or consciousness because I think everything already has it. It’s just our human biology that allows this consciousness to be self-experiential and experience other things, such as thoughts and ideas and feelings. A rock doesn’t have those, but it has a “soul” or consciousness. But I feel I digressed a lot lol
Also to make it clear, I don’t think AI exists already. I think these models and developments we have are part of AI though.
I don’t disagree that experiences are data. The major distinction I’m making is that the human creative process uses more than just data, we have intention, aesthetics, we make mistakes, change our minds, iterate, etc. For a generative AI, the “creative process” is tokenizing a string, running the tokens through an attention matrix, plugging that into a thousand different matrices that then go into a post processing layer and they spit out an image. At no point does it look at what it’s doing and evaluate how it’s gonna fit into the final picture.
As for the rest of your reasoning, I neither agree nor disagree, I think we just don’t have the same definition of consciousness.
I feel your description of what a generative AI does is pretty reductive. The middle part of “plugging the ‘token’ through thousands of different matrices” is not at all well understood. We don’t know how the AI generates the images or text. It can’t explain itself.
And we have ample research showing these models have internal models of the world and can have “thoughts”.
In any case, what would you say consciousness is? This is a more interesting question to me tbh.
Well I don’t see the problem, AI can’t explain itself but it’s nothing more than matrix multiplication with a nonlinearity. Maybe you use a Fourier transform and a kernel instead of scalar weights for a convolutional neural network, maybe it has state instead of being purely feed forward, but at the core of it all you’re doing is multiplying matrices and applying a nonlinearity. I don’t know what you mean that we don’t know how it generates images and text. It’s literally just doing the thing it was programmed to do?
What research? I’d like to see some evidence that these models “think,” given that the way every LLM I know of works is by generating a single word at a time. When you ask a GPT how to bake bread, and the first word it outputs is “Surely!” it has no clue what explanation it’ll start giving you. In fact, whether or not it chooses the exact word “Surely!” as the start of the response has a cascading response on the rest of the output. Then, as I had said earlier, LLMs don’t see anything more than the statistical correlations between words. No LLM knows what gravity is, but when you ask it why things fall down it has enough physics textbooks in its training data that it can parrot the answer from there.
One of the ways I really broke down the idea that GPTs have any model of thought is playing this game. If AI had any actual model of meaning, it would understand security and it would understand not to just tell the player the password. Instead, it will literally blurt it out if you do as much as ask it for words that rhyme. You don’t even need to mention “password,” the way GPT works means that if it detects a lot of weight on a certain word in its previous prompt (which naturally would’ve emphasized the password), it’s almost guaranteed to bring it up again. I know it’s not exactly a hard proof, but it is fun.
As for your last question you’re out of luck because I’m actually just a Catholic lol, not a lot more to say than I believe that there is a metaphysical nature to human experience connecting us to a soul. But that’s a completely unscientific belief to be honest, and it’s not a point I can argue because it’s not based on evidence.