We haven't gotten to the part of the trial for Anthropic yet where we determine whether they actually broke the law when they downloaded from pirate sites. Copy

We haven't gotten to the part of the trial for Anthropic yet where we determine whether they actually broke the law when they downloaded from pirate sites. Copyright has multiple exceptions. And on the topic at hand here (training on YouTube videos to understand space and relationships in it), I don't think even Google would want to make the case that it's a violation of copyright. That's the thing about copyright; it's a whole category of law more based in utility than morality. One of the reasons AI is such a fight right now is that nobody was opposing it as an academic project when it was generating, for example, tools that could go from an image to describing the image, or from an image to recognizing the likely artistic style and helping somebody find the original artist. But with just a few tweaks those tools became devices for generating novel images, and now people are upset. Intent matters. And again, you are drawing equivalence between harvesting data from openly accessible sources online and hiding a server in a closet with unauthorized physical access to a network. Swartz's prosecution wasn't accusing him of copyright violation; it was accusing him of compromising a network. A far more serious charge; if the researchers in the story here had collected those YouTube videos by wiretapping the fiber optics between two of Google's data centers I suspect they would have concerns.