A speaker argues that AI models can be trained on private subsets of data without disclosing their training data, creating ethical concerns because observers only see the final outputs and cannot verify provenance. The speaker likens model learning to musical influence (e.g., Elvis → Beatles → Pink Floyd) to illustrate ambiguity in originality and copying.
So if right now I decide to train my artificial intelligence model on a subset of data, you can only see the final product. You will not see what kind of data I trained on. I don't have to share it. I don't have to say anything about it. Right. It's unethical, but you would never know because all you do is you take a bunch of images, you feed it into a neural network, and then it finds patterns in those images, and it basically creates something new out of this data. So you can say it's like a musician. So, for example, the Beatles were inspired by Elvis. This and Pink Floyd were inspired by the Beatles. So who copied from who? You know, it's the chicken and the egg.
AI News: Dots, GPT-6.1 Sol, Sonnet 5.5, Gemini 4, and everything you need to know
