Does AI Use Stolen Data?

Where does intellectual property begin and where does it end? Do we really know?

Lately, the intellectual-property lawsuits filed over the data used to train large language models have been dominating the agenda. OpenAI, in particular, is often in the headlines over this.

Of course, taking a news article verbatim and sharing it without attribution, or copying and distributing a book without permission, is plainly copyright infringement. That much is clear. But the matter doesn't end there.

A few examples:

• You bought self-help books, read them, learned from them. Then you started offering consulting with the knowledge you'd gained.
• For years you read newspapers, watched TV programs, were shaped by ideas, grew. One day you became a columnist.
• You improved yourself with publicly available resources on the internet. Then you taught in that field, maybe launched a physical product, made money.

Did you infringe copyright?

You're not copying anything word for word. But the knowledge comes from somewhere. So when AI does this, why does it fall into a different ethical/legal frame? The model doesn't share the content as-is; it learns, represents, turns it into vectors. Is that the real problem? Or does the matter only become more sensitive because the one doing it isn't human? AI companies claim this falls within “fair use.”

Or: if the model developers sat down, read and learned from these sources, rewrote what they'd learned in their own words, and trained an AI model on that new content — would that be ethical?

I'm not on either side of the line. I'm genuinely curious about the legal perspectives here. Where should the line be drawn?

I have a feeling we'll be talking about this for a while.