Technology

A lot of these are the distractions imo, they take problems with how AI is currently developed and deployed and treat them as inherent properties of the technology itself. A lot of these criticisms have been repeated so often they've become accepted wisdom in certain circles, despite not standing up particularly well to scrutiny. You've also raised quite a few of them in a relatively short space, without much supporting argument, so addressing them properly is inevitably going to take a much longer response. Apologies in advance but bear with me if you care about the topic at all.

"Fundamentally unreliable" is an increasingly outdated way of viewing LLMs given how much they've improved in fairly short order. There's no denying the fact they can be wrong, but reliability is entirely task-dependent. AI might be poor at specific factual verification while being extremely useful for programming, accessibility, education, brainstorming or countless other applications. It's a tool, and how useful it is depends enormously on what you're using it for and how well you understand its limitations. Dismissing an entire technology because it performs poorly at certain tasks is an arbitrarily narrow way of evaluating it.

"Rooted in plagiarism" confuses training on existing material with reproducing it. There are legitimate copyright and consent issues, but that doesn't make the technology itself plagiarism. Likewise, there's a big difference between pointing out unethical practices in the AI industry and claiming that the technology fundamentally depends on them. You've demonstrated neither why those practices are unavoidable nor why they couldn't be addressed. If you want to point out a company that's done something unethical, entirely valid, that's not the same thing as saying the technology cannot exist ethically, that's nonsense.

Re artists, fully agree there are legitimate concerns about compensation, consent and displacement, but I don't think the existence of AI generated art inherently devalues human creativity. Much of the damage here comes down to how the technology is commercialised and regulated, which ties into my comment above.

The environmental costs are real, but you've not attempted to weigh them against the benefits, so you could be saying that about any polluting technology. You've used one to write this to me. You probably used one to get to work this morning. There might be one keeping your house the correct temperature. New technologies are very often relatively inefficient in their early stages, I doubt you'll struggle to imagine prior examples of that, and LLMs have already become significantly more computationally efficient in the few years since their mainstream introduction, as has the hardware. That obviously doesn't negate their environmental impact, particularly as demand continues to grow, but treating today's resource requirements as an inherent, permanent characteristic of the technology is shortsighted.

That's a regulation issue. Ireland put in some rules last year that any new data centre had to also install enough solar to cover it's entire demand and connect it to the grid. More of the world needs to be doing shit like that. If they want to profit, they need to earn it.

The cognitive debt research you mentioned is preliminary and task specific, and evidence of AI reinforcing delusions in vulnerable individuals doesn't mean that the technology generally causes psychological harm, and it's hardly the first technology capable of that. Books can do that, tv, social media, not to mention 4chan. That's another regulation issue, safeguarding. Similarly, AI reinforcing people's existing biases is a legitimate concern, but that's not really unique to AI. Algorithms have been doing that for years, and AI can equally be used to challenge people's assumptions, depending on how it's designed and used.

The corporate greed, labour displacement, reckless investment and the concentration of wealth are serious concerns, and the core part of why I think the political and economic context surrounding AI matters so much, and that leads me into to the thing that drives me to argue any of these points. The chances of the technology going away are essentially zero, there is absolutely no point in talking about that. It's a fantasy. You'd need a solar flare or a global revolution, most of us cant even revolt against being reigned by overt paedophiles. There is far more value in talking about how to mitigate the issues around it, because that is actually plausible.

I just think you're making a big leap from identifying problems with AI, some perfectly legitimate, to concluding that the technology itself is fundamentally bad. And calling it a "technological dead end" seems especially strange given how much it's improved in just a few years. I don't see how you've arrived at that conclusion, or why you'd be confident these problems are impossible to address.

It's worth comparing AI to previous 'technological revolutions'. They brought enormous benefits alongside serious social disruption, exploitation and unintended consequences. How those benefits and harms were distributed depended massively on the political and economic systems surrounding them. There's no reason to look at AI any differently.

I appreciate that we're probably not going to agree on this, but you put some effort into your response, so I thought it deserved a proper one in return.