E.g. If a large language model doesn't censor inappropriate content (like violent, sexually explicit, hateful, etc. content), or allows users to make the model swear, is this legal?
6
-
1While it is always difficult to prove something is legal, the existance of huggingface which seems to have hundreds of thousands of models fitting your description I think indicates the answer is at the moment yes.– User65535Commented Jun 3 at 16:18
-
1There are currently very few laws regarding AI, because it's very new technology and lawmakers haven't had time to figure out what should be legal and write the laws.– BarmarCommented Jun 3 at 17:03
-
1I can't imagine any jurisdiction that generally allows free speech outlawying AI's that swear. If humans can swear online, why shouldn't a LLM?– BarmarCommented Jun 3 at 17:04
-
@Barmar, actually the usual laws apply to AI -- which can be a serious problem for new technology.– o.m.Commented Jun 3 at 19:28
-
@o.m. Other than copyright, what are the "usual laws" that apply to distributing (as opposed to running) software?– BarmarCommented Jun 3 at 19:33
|
Show 1 more comment
1 Answer
No law in the U.S. requires artificial intelligence large language models (AI LLMs) to filter inappropriate content. The main issues with creating an AI LLM are whether it used copyrighted content in an infringing matter to train it, and whether it otherwise used illegal means to obtain the training data.
All Internet content in China is subject to strong state censorship of content deemed inappropriate.
Other countries may have other policies.
-
Interesting thing would be to create an AI model completely absent from any content filtering (incl. bias of its creators), but having some capability to "think about" the possible biases of all his sources. Commented Jun 4 at 3:00
-
The main issues with creating an AI LLM are whether it used copyrighted content in an infringing matter to train it- which is an interesting way to put it given that generally it's held that training is inherently fair use (although this hasn't been fully litigated yet). Commented Jun 5 at 22:32 -
3@Clockwork-Muse "given that generally it's held that training is inherently fair use" This is not at all a generally held belief and is indeed a hotly debated issue which is the subject of pending litigation. Commented Jun 5 at 22:59
-
@Clockwork-Muse An analysis of the current debate over fair use in the context of AI training can be found in a recent law review article at papers.ssrn.com/sol3/papers.cfm?abstract_id=4834410 Commented Jun 5 at 23:55