Well, I've been thinking about the AI and how it's going about putting together the art. I'm not convinced that it understands things like mise-en-scène or even the fundamentals of composition and colour, but it can translate text into a searchable visual database like Watson does. That you need to prod it into taking a certain "style" indicates that the AI isn't to be trusted with coming up with the "best" composition, although you can say the same for human artists: most of us stay in our comfort zones. You can choose "no style" which means... I don't know exactly. It's not random, but it also doesn't seem to allow Wombo its own initiative.
The Wombo AI is canny about selecting visual elements, but then the DIGIC or similar processor in your camera does the same thing: it combs a database until it finds something that resembles the target image and then tweaks the processor to add some moxie to the final product, sort of like hiring Carrie Fisher or Quentin Tarantino to punch up your movie script. I guess Wombo has a stock set of visual filters it applies to its pictures to make them look more painterly, that's what I see when running the AI through multiple iterations. Again, not that much different from a human.
We can laugh at or comment on the surreal aspects of the picture, but the AI is learning with every interaction. I remember the matching game that we played twenty years ago to help Google™ become smarter, and then when it was released it made Ask Jeeves™ look like a simpleton by comparison. With cloud computing, Google™ has become even more accurate.
AI has a harder time dealing with imagery than it does with audio or text. But it is learning. Google™ Image Search is very accurate now, to the point where it's easy to spot copyright fraud with a simple mouse click. I believe Wombo is using similar if not the same technology to power its AI. In twenty years, it will be making its own art on its own initiative.
What would the purpose be of AI-created art? Obviously, there's promoters who believe it could be sold. That bugs me a bit, but then again I have to compete with Banksy so it's safe to say the entire Arts Industry bugs me a bit. On the other hand, if the AI is an enslaved intelligence, or even if it is allowed to roam free, or most especially if it becomes our Robot Protector, it will want to be expressive and I don't see why that expression should be hindered, unless it takes up the entire planet to create its homage to Sylvia Cradling Grogu. I'm not quite ready to live out the remainder of my life as Grogu Ear Hair 342,147 in a planetary-sized diorama.
Enjoy clicking on the I'm Not A Robot button while you can, fellow squishy humans.
1 comment:
Well, I've been thinking about the AI and how it's going about putting together the art. I'm not convinced that it understands things like mise-en-scène or even the fundamentals of composition and colour, but it can translate text into a searchable visual database like Watson does. That you need to prod it into taking a certain "style" indicates that the AI isn't to be trusted with coming up with the "best" composition, although you can say the same for human artists: most of us stay in our comfort zones. You can choose "no style" which means... I don't know exactly. It's not random, but it also doesn't seem to allow Wombo its own initiative.
The Wombo AI is canny about selecting visual elements, but then the DIGIC or similar processor in your camera does the same thing: it combs a database until it finds something that resembles the target image and then tweaks the processor to add some moxie to the final product, sort of like hiring Carrie Fisher or Quentin Tarantino to punch up your movie script. I guess Wombo has a stock set of visual filters it applies to its pictures to make them look more painterly, that's what I see when running the AI through multiple iterations. Again, not that much different from a human.
We can laugh at or comment on the surreal aspects of the picture, but the AI is learning with every interaction. I remember the matching game that we played twenty years ago to help Google™ become smarter, and then when it was released it made Ask Jeeves™ look like a simpleton by comparison. With cloud computing, Google™ has become even more accurate.
AI has a harder time dealing with imagery than it does with audio or text. But it is learning. Google™ Image Search is very accurate now, to the point where it's easy to spot copyright fraud with a simple mouse click. I believe Wombo is using similar if not the same technology to power its AI. In twenty years, it will be making its own art on its own initiative.
What would the purpose be of AI-created art? Obviously, there's promoters who believe it could be sold. That bugs me a bit, but then again I have to compete with Banksy so it's safe to say the entire Arts Industry bugs me a bit. On the other hand, if the AI is an enslaved intelligence, or even if it is allowed to roam free, or most especially if it becomes our Robot Protector, it will want to be expressive and I don't see why that expression should be hindered, unless it takes up the entire planet to create its homage to Sylvia Cradling Grogu. I'm not quite ready to live out the remainder of my life as Grogu Ear Hair 342,147 in a planetary-sized diorama.
Enjoy clicking on the I'm Not A Robot button while you can, fellow squishy humans.
Post a Comment