Hackernoon has an interesting quick look article on the “debate” of AI data training sets and copyright infringement. Is it Fair Use or is it infringement? The Fair Use threshold, IMO, is not met by AI companies and the mounting evidence that AI companies new and used pirated materials does not fall in their favor.
A choice quotation from this article for me is: “Although an algorithm’s output may technically be transformative, it can’t create something original. This is a by-product of its training. It learns to anticipate where a word or pixel should go, creating the illusion that it can write a poem or draw a picture. While a generative model without a training dataset can’t produce anything, even babies can paint on a canvas. This distinction could determine whether AI violates fair use.”
Gen AI doesn’t create, it pieces things together. Without the data sets it can’t do a thing. AGI isn’t happening anytime soon. Why? Because no thinking is happening in current “AI” models. Calling it AI is maybe the leap, is it just machine learning accelerated?
PS Feature Gen AI image “machine learning” brought to you by “shaky numbers and text”. If there was actual Artificial Intelligence existing in Gen AI, it would know when text or numbers didn’t look right and fix it, but it can’t.
