Sarah Silverman really did a hit job on AI systems on the 11-8 episode of the Daily Show. I feel like it is largely fueled by ignorance of how the mathematics in these systems actually works. These systems do not make “copies/are copycats” like Sarah ignorantly espoused, they train on data and project that into an n-dimensional space to generate something new from its experience, not much different than humans do. They do not memorize the original data and make copies at all.
Most of are you are familiar with 2 dimensions like a piece of paper or 3 dimensions like a cube, machine learning systems learn in n-dimensional space where n can be any number; most of these systems the space is 10,000-1 million dimensions. These systems aren’t simply making a simple copy but extracting the most salient features in text, images, etc into a n-dimensional space to create a new product based on all of its experiences.
This is really no different than how humans create art, they observe lots of styles, learn from it, and try to create new things based on their knowledge based on the many dimensions learned by their observation and experience. Why is it wrong for a computer to learn from art posted online, but it is no issue for a human to learn from art posted online? Do humans have to cite every single painting they ever saw when creating something new? This seems like a double standard honestly.
Also creating AI models is in itself an expression of the artistic process. These systems are created by humans, not machines; they are an extension of human mathematical and scientific creativity. Fire was made by hand for 1000s of years; is it not an extension of human creativity to create a lighter such that you can create a flame at any time; likewise generating AI systems to create art is in itself an extension of human creativity and ingenuity in the same way that creating a lighter to make fire making easier is.
I liked Sarah Silverman for the rest of her segments, but she really showed her ignorance and lack of any technical understanding from a scientific/mathematics perspective on the development of AI.
Its not a copy, its a transformation using linear algebra. Do artists get sued if they simply viewed someone else’s painting years ago and don’t attribute credit?
The issue is 99% of lawyers dont have a mathematics or computer science background to actually what is fundamentally happening, these are not copies, it’s a linear algebra transformation. Only 1 data point exists in the inference matrix for every 10,000 to 100,000 data points in the training data. Is it copying if an artist views 10,000 paintings and that artist doesn’t attribute credit to each of the 10,000 paintings?
You can say that all you want but if it makes something that looks similar to a human artist’s work, they can and will get sued. Just like an artist who makes a song that sounds just like a Marvin Gay song can get sued for copying it. The problem is 99% of computer scientists don’t have a legal background.