Mastodawn

Cassandra is only carbon now

Yesterday, had an argument with an AI booster. I'm not going to link, both because I don't want to platform that and because I don't want anyone to go harass them. But what I thought was very interesting was that I asked point-blank if there was any degree to which ethical problems with LLMs could make them not want to use AI — they told me no, there was not, and implied that they evaluated AI purely on the basis of its efficacy.

Cassandra is only carbon now Feb 22

I don't have time nor the inclination to argue that point with them further when it comes to AI. But I do think there's a broader point that is worth critical examination, especially as tech continues to build out surveillance, age verification, automated filtering and censoring, and other tools that do immense damage when used by authoritarians.

We *cannot* afford to evaluate tech purely based on whether it "works" or not.

Cassandra is only carbon now Feb 22

AI doesn't work¹, so it's easy to forget that larger point, I suspect? That *even if* AI did work (and again, it doesn't), it still would need to be critically examined from an ethical perspective.

Failing to do so is how we have massive surveillance networks today.

___
¹Here again, referring to the wave of current hype products. Boosters love wearing the ML shit that does work as a shield against criticism.

I'm not sure I'd say "AI doesn't work" anymore. It definitely doesn't "work" to the degree that the loudest boosters will claim it does. But like, I do think it's recently crossed a threshold where it can be a useful tool in the right hands.

Which I personally find very annoying since I too have moral qualms about the broader AI industry. E.g. the point about surveillance you're making I think is an important one.

Kevin Granade Feb 22

@xgranade it's exhausting as fuck when people that both agree with and disagree with your main point (don't use or promote LLMs) object when the rationales you muster in support are not precisely the same ones they care about.
IMO this is where people get the idea that the "purity culture" angle presents a wedge issue they can use to attack anti-AI people.

Cassandra is only carbon now Feb 22

@kevingranade I never want to put my AI Luddism on a pedestal and make it immune to critique... this is, for example, why I said my closed-mindedness on the subject is both temporary and a reasoned response to bad-faith DDoS attacks on discourse.

To that extent, I'm glad for critique from "my" side. But the purity culture discourse (with a few important exceptions) isn't that, it's a wedge.

Cassandra is only carbon now Feb 22

@kevingranade I find that progressive moments, as a consequence of the laudable and correct willingness to self-criticize, tend to be vulnerable to wedge attacks. Fuck, as a trans person I *am* a wedge, or at least the right-wing has turned me into a wedge used to weaken opposition to violent and cruel immigration policies.

We need to get a lot better at distinguishing wedges from critiques.

Mario Munoz Feb 23

@xgranade This has also been fascinating to me lately.

My latest blog post is specifically musing on what that calculus looks like for each person.

I always presumed that a major factor is in the belief that harms are not as bad as reported (which calls into question what sources we are viewing as authoritative) or that the current and/or future benefits to humanity are worth it (current harms are collateral damage in lieu of progress).

But for a person to be unable to imagine any scenario that would change their mind seems crazy to me.

I often ask myself what it would take to change my mind on this issue, and while I think most of those scenarios are highly improbable.. I can still imagine them.

Mario Munoz Feb 23

@xgranade hmm, now that I think about it, maybe I should articulate that in a future blog post.

@xgranade
In a meeting today, I recently raised the analogy to debates about torture. The debate gets muddy when people mix together questions of efficacy with questions of ethics and morality. Sometimes this confusion is deliberate but I suspect most of the time, it's just a lack of clarity on priorities. Whatever action we are considering, the first question should always be whether it's an ethical or moral action. The efficacy question should always be secondary.

tuban_muzuru Feb 23

The stuff which does work - is in its infancy, anyway.

tuban_muzuru Feb 23

How would you define "work" in this context? By this I mean what claims are being made by the hype.

💀 𝓕airchild 💀Feb 23

@xgranade And especially given that it works reliably 'in this niche, but not this one'. Reliability and repeatability is all over the map.

Blake C. Stacey Feb 22

@xgranade had a very similar conversation the other day too; only was able to somewhat shift my interlocutor’s position after pointing out they don’t own this “tool”, and they are at the mercy of fash/oligarch class who can (and will) start extracting rent at any moment.

Dominic 🇪🇺 🏳️‍🌈 🇺🇦Feb 23

@wbftw @xgranade Yes! Even if we stick to one domain where “AI” works today (yes yes, FSVO), namely coding assistants, what is the future of non-commercial software if it just becomes normalised that being a programmer means paying rent of tens or hundreds of dollars a month in tokens? And that price rising once everyone is locked in, until the LLM operators can make a profit?

leastaction Feb 23

@xgranade Its the government's job to regulate AI on an ethical basis, on behalf of all of us, because individually we don't have the power to do that, and the government represents the people in a democracy, and it does have the power. The fact that the government is not doing this at all shows to what extent we do not have a democracy. Not even close.

@leastaction @xgranade it just shows that most people (represented by politicians) don’t care. They care much more about economic growth and their wealth then ethics. Also people in general are extremely lazy, which is why it is so tempting to use it and most are even willing to outsource their thinking to it.

@leastaction @xgranade by that I didn’t mean that politicians are not pushing their own agenda that can go against what their voters want. The only thing I claim is that the biggest flaw in any political system is ultimately people, so more democracy wouldn’t mean less push of AI.

leastaction Feb 23

@neongod @xgranade A lot depends on whether we see government regulation as an imposition by a powerful authority, or as a protection. In a democracy, government works *for* us, as per Lincoln's ddefinition. In a democracy ethics is a way we all protect each other.

We also have to be careful not to blame the victim, I think.

💀 𝓕airchild 💀Feb 23

@xgranade
Sociopaths gonna socio... path.

The Tattooed Nonna 🧙‍♀️Feb 23

Chaddicus Feb 23

@xgranade yikes...there are a lot of things that are effective at efficiency accomplishing a goal that are certainly not ethical. I understand that it's easier to rationalize away ethical concerns when it's abstracted through several layers of stubbornness, doubt, etc., but I would have a very difficult time trusting or even conversing with someone who so utterly rejects ethics as a consideration. That recent satirical post someone made about the efficient orphan smashing machine comes to mind.

Andrew Radev Feb 23

@xgranade The way that I personally interpret cases like this is a sort of "just world" belief. If it was truly bad, surely it would not be allowed? If there was a real problem, there would be some kind of higher power that stops it.

This also aligns with conversations where I point out that this stuff is heavily subsidized and the person says "well, it's free/cheap now", with no further elaboration. The implication is: "I will use it because I can. If it was bad to use, it would not have been usable."

If you believe that the status quo is good and just, then you don't need to consider anything outside of your immediate gratification. The consequences (to society or to your own brain) are someone else's problem. Once the rockets go up...

C. R. Collins 🌳 🐸Feb 23

I am so sick of the general capitalist culture's habit of evaluating everything on whether it's "profitable" first & foremost & whether it's good, decent, healthy, or moral later, or never. It's a sick way to run a society, full stop. I will never willingly use their error prone, unethical, environmentally disasterous slop machine nor purchase a work from anyone who has. I'm fine being a Luddite on this or whatever else they want to call me. I don't respect them enough to care.

@xgranade @jwz eech!

Avi Rappoport (avirr)Feb 23

@xgranade The willful Ignorance of externalities is so depressing

William B Peckham Feb 23

@xgranade AI (currently Automated Incompetence) can certainly work if it's configured and fed correctly. Fed tons of poorly curated stolen data from all over the internet, which is 90% garbage, it mostly spews garbage. (GIGO applies) the concept of such an LLM being able to properly feed a front end that would then magically give correct answers was flawed from the beginning. It is that unethical back-end that is both the ethical problem and the technical problem with what they're pushing. There are solutions for that in the long run, but in the short run, the only correct step is to stop using LLMs, or never start.

@xgranade I almost fell out of my chair when I checked out an AI training course and one of the takeaways was it is liable to hallucinate so double check all its output instead of blindly relying on it.

That advice of course makes sense, but then how exactly am I saving time by using this thing?