Sure, for reasonable copyright terms. Currently, if you create something when you’re young and live long, a 150 year long copyright term is reasonably possible. (Life + 70 years)
Much as I appreciate someone’s rights to their work, things should enter the public domain in something more like 10 to 20 years. Even then, copyright protections are too strong when in force. You published something so people would use it, your ability to limit how should be quite restricted to protecting you from folks selling it as their own. I am also in favor of forced standard licensing terms.
Like say after five years there should be a standard streaming licensing fee for films and shows such that anyone can broadcast/stream/sell copies for a flat rate.
We also have to consider the cost of enforcement. We can’t be soaking up millions and billions of taxpayer dollars to protect copyrights or field complaints that aim to protect mutated copies of said works…just like you don’t send a swat team to enforce parking tickets, we have to consider what is at loss for the New York Times or other copyright holders before clogging up the courts.
There’s a reason lawyers are so quick to file a suit and it’s because it cost nothing to sic the dogs of the American justice system on others.
Kim Dotcom’s adventure calmed things down for the last wave of digital ip theft. Once that happens with one or two ai copyright disbelievers the rest will calm down.
But is it really a problem if the AI is transmitting the information in its own words? And even if that is considered illegal, doesn’t it significantly diminish said crime?
AI doesnt transmit information in its own words. It has no “own” no “self”. It does what it was programmed to do, just like any other type of software. Turns out that some people using ai have made it ingest content without permission so they can resell it for profit. That should not be permitted. My property is not yours to take unless you agree to my terms. I did not give you permission to download my data, art, code or text, to ingest in a token database and then resell it in any shape or form derived or not. No ifs no buts. If you want it you have to pay for it or respect terms. The bulk of ai companies respect that. A handful of sociopaths dont. They are the issue.
I wouldn't call ANYONE disrespecting terms and conditions they may have agreed to a sociopath. Not everything in a contract is enforceable just because it's written there, whether or not both parties signed it. And unless it's spewing out copyrighted materials "verbatim" there is an argument to be made that the LLM learned to talk from an open source and inserted knowledge from a copyrighted one.
However this turns out for private AI, I hope at the very least it can be considered fair use. Monetized LLMs can be forced to pay up or follow terms but individuals should be able to pool together and create open source models. I'm not saying I have the exact legal arguments for why this would work but LLMs in their current forms need to exist.
I absolutely agree that LLMs should exist. Torrents still exist and have their purpose. Criminals always argued that their crime is not really a crime and found all kinds of arguments in favour of it. Similarity people developing ai that doesnt respect people’s property use all sorts of wild arguments in their favour - ai learns like a human, it benefits society, other countries will use it against us, and so on. That doesnt mean we should give into their demands to destroy society and people’s lives so they can have a competitive advantage over honest people. The fact that they want to steal, destroy entire industries they take from, and demolish norms so they can make their software appear intelligent, makes them sociopaths.
A significant portion of the training set for most image generation tools is stuff made in the last 10-20 years harvested from the internet, if not the last 5 years. We're not talking about 150 years of copyright protection here, we're talking about the time frames you suggest. Artists want to protect their own work and their livelihood, and AI is being trained on the work they're actively putting out right now. You would have to shorten copyright duration to something like 5 years to come remotely close to making modern image generation models possible without violating artist copyrights.
Text is different and much less difficult since its history as a medium is much longer - if you excluded the last 10-20 years of prose from your LLM it would probably still be very good at writing. But excluding the last 20 years of digital illustration and photography would be limiting yourself to a much lower-fidelity training set.
Much as I appreciate someone’s rights to their work, things should enter the public domain in something more like 10 to 20 years. Even then, copyright protections are too strong when in force. You published something so people would use it, your ability to limit how should be quite restricted to protecting you from folks selling it as their own. I am also in favor of forced standard licensing terms.
Like say after five years there should be a standard streaming licensing fee for films and shows such that anyone can broadcast/stream/sell copies for a flat rate.