First and foremost, this is not about AI/ML research, only about usage in generating content that you would potentially consume.
I personally won’t mind automated content if/when that reach current human generated content quality. Some of them probably even achievable not in very distant future, such as narrating audiobook (though it is nowhere near human quality right now). Or partially automating music/graphics (using gen AI) which we kind of accepted now. We don’t complain about low effort minimal or AI generated thumbnail or stock photo, we usually do not care about artistic value of these either. But I’m highly skeptical that something of creative or insightful nature could be produced anytime soon and we have already developed good filter of slops in our brain just by dwelling on the 'net.
So what do you guys think?
Edit: Originally I made this question thinking only about quality aspect, but many responses do consider the ethical side as well. Cool :).
We had the derivative work model of many to one intellectual works (such as a DJ playing a collection of musics by other artists) that had a practical credit and compensation mechanism. With gen AI trained on unethically (and often illegally) sourced data we don’t know what produce what and there’s no practical way to credit or compensate the original authors.
So maybe reframe the question by saying if it is used non commercially or via some fair use mechanism, would you still reject content regardless of quality because it is AI generated? Or where is the boundary for that?
That’s a good start, but where do you draw the line? If I use a template, is that AI? What if I am writing a letter based on that template and use a grammar checker to fix the grammar. Is that AI? And then I use the thesaurus to automatically beef up the vocabulary. Is that AI?
In other words, you can’t say LLM and think it’s a clear proposition. LLMs have been around and used for various things for quite a while, and some of those things don’t feel unnatural.
So I’m afraid we still have a definitional problem. And I don’t think it is easy to solve. There are so many interesting edge cases.
Let’s consider an old one. Weather forecasting. Of course the forecasts are in a sense AI models. Or models, if you don’t want to say AI. Doesn’t matter. And then that information can be displayed in a table, automatically, on a website. That’s a script, not really AI, but hey, you could argue the whole system now counts as AI. So then let’s use an LLM to put it in paragraph form, the table is boring. I think Weather.com just did this recently and labeled it “AI forecast”, in fact. But is this really an LLM being used in a new way? Is this actually harmful when it’s essentially the same general process that we’ve had for decades? Of course it’s benign. But it is LLM, technically…