Personally I think the vast majority AI is ethically indefensible purely based on how it was trained. Generative AI represents, at best, the privatization of culture by for-profit corporations, and at worst, one of the most unprecedented instances of exploitation of human labor and creativity in history.

And you don't have to take my word for it because it was recently uncovered in court that Microsoft director of applied science Brent Hecht called it the "largest theft of labor in human history".

Even if we were to totally ignore how wasteful most AI is, how damaging to the climate it is, and how much it threatens to further broaden the gap between the rich and the poor, I already hate AI for what is so obviously the capitalistic strip-mining of all human culture in the name of creating shareholder value.

Pro-AI people love to brush off pretty much all of the ethical concerns with the simple argument that the ends justify the means. I reject that outright, but for the sake of argument I want you all to consider whether that is even a valid point... Do the ends justify the means? We are doing all of this AI shit to what ends?

Who benefits from this supposedly increased "productivity"? Definitely not the working class who have at most a relatively minor stake in the success of big tech businesses. Can anyone seriously claim we are seeing things improve at all, let alone an accelerating rate? Do we really see AI bringing about a utopian society where everything is great? Or are we seeing things enshittify and become more corporate and less affordable week by week? Is any of this shit making your life easier or better?

Now don't get me wrong... it seems to be making some things much easier: scams, disinformation, surveillance, propaganda, fraud, and of course generic commercialized slop. We honestly couldn't ask for a better set of tools to trick people into giving us money. Generative AI has truly disrupted the market for scams, and it seems that it has brought about a new golden age of fraudulence.

local open-weights models, self-hosted setups, and practical developer/productivity tools

Not to be a wet blanket, but I feel this stuff really needs a reality check too.

"Open weights" is nothing more than a freeware binary blob.

I can see how that's at least preferable to using some company's subscription service, but at the same time your local LLM still has a hard dependency on someone else's corporate training dataset (pretty much all the result of stolen labor, by the way), and so it's very much not as "open" or "local" as people like to make it out to be. And from an ethical standpoint, it's not a whole lot different either way.

I think an ethical alternative to today's generative AI could exist--if it was trained on public domain, creative commons and licensed data, and thus not exploiting human labor and culture--but I'm not seeing people invest any time in making that.