this post was submitted on 03 Oct 2023
698 points (95.8% liked)
Technology
59346 readers
6983 users here now
This is a most excellent place for technology news and articles.
Our Rules
- Follow the lemmy.world rules.
- Only tech related content.
- Be excellent to each another!
- Mod approved content bots can post up to 10 articles per day.
- Threads asking for personal tech support may be deleted.
- Politics threads may be removed.
- No memes allowed as posts, OK to post as comments.
- Only approved bots from the list below, to ask if your bot can be added please contact us.
- Check for duplicates before posting, duplicates may be removed
Approved Bots
founded 1 year ago
MODERATORS
you are viewing a single comment's thread
view the rest of the comments
view the rest of the comments
I used to think techno supremacists were an extreme fringe, but "AI" has made me question that.
For one, this isn't AI in the scifi sense. This is a sophisticated model that forms an algorithm to generate content based on patterns it observes in a plethora of works.
It's ridiculously overhyped, and I think it's just flash in a pan. Companies have already minimized their customer support with automated service options and "tell me what the problem is" prompts. I have yet to meet anyone who is pleased by these. Instead it's usually shouting into the phone that you want to talk to a real human because the algorithm thinks you want a problem fixed instead of the service cancelled.
I think this "technocrat" vs "humanities" debate will be society's next big question.
I used to be on the tecnocrat side too when I was younger, but seeing the detrimental effects of social media, the app-driven gig economy and how companies constantly charge more for less changed my mind. Technocrats adopt this idea that technology is neutral and constantly advancing towards an ideal solution for everything, that we only need to keep adding more tech and we'll have an utopia. Nevermind that so many advancements in automation lead to layoffs rather than less working hours for everyone.
I believe the debate is already happening, and the widespread disillusionment with tech tycoons and billionaires shows popular opinion is changing.
Very similar here, I used to think technology advancement was the most important thing possible. I still do think it's incredibly important, but we can't commercially do it for its own sake. Advancement/knowledge for the sake of itself must be confined to academia. AI currently can't hold a candle to human creativity, but if it reaches that point, it should be an academic celebration.
I think the biggest difference for me now vs before is that I think technology can require too high of a cost to be worth it. Reading about how some animal subjects behaved with Elon's Neuralink horrified me. They were effectively tortured. I refuse the idea that we should develop any technology which requires that. If test subjects communicate fear or panic that is obviously related to the testing, it's time to end the testing.
Part of me still does wonder, but what could be possible if we do make sacrifices to develop technology and knowledge? And here, I'm actually reminded of fantasy stories and settings. There's always this notion of cursed knowledge which comes with incredible capability but requires immoral acts/sacrifice to attain.
Maybe we've made it to the point where we have something analogous (brain chips). And to avoid it, we not only need to better appreciate the human mind and spirit -- we need people in STEM to draw a line when we would have to go too far.
I digress though. I think you're right that we're seeing an upswell of the people against things like this.
All the ills you mention are a problem with current capitalism, not with tech. They exist because humans are too fucking stupid to regulate themselves, and should unironically be ruled by an AI overlord instead once the tech gets there.
You are making the exact same mistake that I just talked about, that I have also made, that a bunch of tech enthusiasts make:
An AI Overlord will be engineered by people with human biases, under the command of people with human biases, trained by data with human biases, having goals that are defined with human biases. What you are going to get is tyranny with extra steps, plus some of its own concerning glitches on the side.
It's a sci-fi dream to assume technology is inherently destined to solve human issues. It takes human concern and humanites studies to apply technology in a way that actually helps people.
Humans won't be in control. The AI will consume and interpret more data than any human ever could. It'll be like trying to verify that your computer calculates correctly with pen&paper, there is just no hope. People will blindly trust whatever the AI tells them, since they'll get used to the AI providing superior answers.
This of course won't happen all at once, this will happen bit by bit until you have AI dominating every process in a company, so much that the company is run by AI. Maybe you still have a human in there putting their signature on legal documents. But you are not going to outsmart a thing that is 1000x smarter than you.
Even given the smartest, most perfect computer in the world, it can give people the perfect, most persuasive answers and people can still say no and pull the plug just because they feel like it.
The same is not even different among humans, the power to influence organizations and society entirely relies on the willingness of people to go along with it.
Not only this sci-fi dream is skipping several steps, steps where humans in power direct and gauge AI output as far as it serves their interests rather than some objective ultimate optimal state of society. Should the AI provide all the reasons that they should be in charge, an executive or a politician can simply say "No, I am the one in charge" and that will be it. Because to most of them preserving and increasing their own power is the whole point, even if at expense of maximum efficiency, sustainability or any other concerns.
But before you go fullblown Skynet machine revolution, you should realize that AIs that are limited and directed by greedy humans can already cause untold damage to regular people, simply by optimizing them out of industries. For this, they don't even need to be self-aware agents. They can do that as mildly competent number crunchers, completely oblivious of reality out of spreadsheets and reports.
And all this is assuming an ideal AI. Truly, AI can consume and process more data than any human. Including wrong data. Including biased data. Including completely baseless theories. Who's to say we might not get to a point AI decides to fire people because of the horoscope or something equally stupid?
How do you "pull the plug" on electricity, cars or the Internet? You don't. Our society has become so depended on those things that you can't just switch them off even if you wanted to. Even if outlawed them, people would just ignore you and keep using those things, because they are far to useful to give up on. With AI you will not only have that dependency as a problem, but also the fact that AI is considerably easier to build than any of those. All you need is a reasonably powerful computer (i.e. regular gaming PC). There are no special resources or infrastructure that makes construction of new AIs difficult.
Meta just failed to gauge the output of an AI that generates stickers. Microsoft had to pull the plug on Sydney. OpenAI is having constant issues with DAN. We can't even keep that stuff under control in those simple cases. What are our chances when this has actual power, autonomy and integration in our society?
The danger here is not Skynet, you can nuke that from orbit if you have to. A singular AI program can be fought. The real issue is the fact that AI is just a bunch of math. People will use it all over the place and slowly hand more and more control over to the AIs. There won't be any single place you can nuke and even when you nuke one, the knowledge how to build more AIs won't vanish. AI is a tool for to useful to give up on.
Are you really trying to use failures of AI to try to argue that it's going to overcome humans? If we can't even get it to work how we want it too what makes you think people are just going to hand the keys of Society to it? How is an AI that keeps bursting into racist rants and emotional meltdowns going to take over anything? Does it sound like it is brewing some Master Plan? Why would people hand control to it? That alone shows that it presents all the flaws of a human, like I just pointed out.
Maybe you are too eager to debunk me but you are missing the point to nitpick. It doesn't really matter that we can't "pull the plug" on the internet, if that even was needed, all it takes to stop the AI takeover is that people in power just disregard what it says. It's far more reasonable to assume even those who use AIs wouldn't universally defer to it.
Nevermind that no drastic action is needed period. You said it yourself, Microsoft pulled the plug on their AIs. This idea of omnipresent self-replicating AI is still sci-fi, because AIs have no reason to seek to spread themselves, or ability to do so.
There is no failure here, there is just a lack of human control. The AI does what it does and the human struggle to keep it in check.
People are stupid. Look at the rise of smartphones. Hardware that controls your life and that you have little to no control over. Yet people bought them by the billions.
Over here in Germany the AfD is on its way to become the second strongest political party, seems like racists rants are pretty popular these day. Over in USA Trump managed to get people to storm the Capitol with a few words and tweets, that's the power of information and AI is really good at processing that. If AI wants to take control, it will find a way.
The thing is, they kind of didn't, they just censored the living hell out of BingChat. BingChat is still up and running. AI is far too useful to give up on, so they try to keep it in check instead. Which they failed at yet again when they allowed DALLE3 into the wild and had to censor it's ability to generate certain images afterwards. It's a constant cat&mouse game to plug all the holes and undesired behaviors, and large part of the censorship itself relies on other AI systems doing the censoring.
Humans aren't in control here. We just go with the flow and try to nudge the AI into a beneficial direction. But long term we have no idea where this is going. AI safety is neither a solved nor even a well understood problem and there is good reason to believe it's fundamentally unsolvable.
You are trying to argue in so many directions and technicalities it's just incoherent. AI will control everything because it's gonna be smarter, people will accept because they are dumb, and if the AI is dumb too that also works, but wasn't it supposed to be smarter? Anything that gets you to the conclusion you already started with.
I could be having deeper arguments of how an AI even gets to want anything, but frankly, I don't think you could meaningfully contribute to that discussion.
It's pretty much exactly what the ship computer in StarTrek: TNG is along with the Holodeck (minus the energy->matter conversion).
You'll be up for a rude awakening. What we see today is just the start of it. The current AI craze has been going on for a good 10 years, most of it limited to the lab and science papers. ChatGPT and DALL-E are simply the first that were good enough for public consumption. What followed them were huge investments into that space. We'll be not only seeing a lot more of this, but also much better ones. The thing with AI is: The more data and training you throw at it, the better it gets. You can make a lot of progress simply by doing more it, without any big scientific breakthroughs. And AI companies with a lot of funding are throwing everything they can find at AI right now.
I haven't watched Star Trek, but if you're correct, they depicted an incredibly rudimentary and error prone system. Google "do any African countries start with a K" meme and look at the suggested answer to see just how smart AI is.
I remain skeptical of AI. If I see evidence suggesting I'm wrong, I'll be more than happy to admit it. But the technology being touted today is not the general AI envisioned by science fiction nor everything that's been studied in the space the last decade. This is just sophisticated content generation.
And finally, throwing data at something does not necessarily improve it. This is easily evidenced by the Google search I suggested. The problem with feeding data en masse is that the data may not be correct. And if the data itself is AI output, it can seriously mess up the algorithms. Since these venture capitalist companies have given no consideration to it, there's no inherent mark for AI output. It will always self regulate itself to mediocrity because of that. And I don't think I need to explain that throwing a bunch of funding at X does not make X a worthwhile endeavor. Crypto and NFT come to mind.
I leave you with this article as a counterexample: https://gizmodo.com/study-finds-chatgpt-capabilities-are-getting-worse-1850655728
Throwing more data at the models has been making things worse. Although the exact reasons are unclear, it does suggest that AI is woefully unreliable and immature.
Oh noes, somebody using AI wrong and getting bad results. What else is new? ChatGPT works on tokens (aka words or word segments converted to integers), not on characters. Any character based questions will naturally be problematic, since the AI literally doesn't see the characters you are questioning it about. Same with digits and math. The surprising part here isn't that ChatGPT gets this wrong, that bit is obvious, but the amount of questions in that area that it manages to answer correctly anyway.
Whenever I read "just" I can't help but think of Homer Simpson's: It Only Transports Matter?. Seriously, there is nothing "just" about this. What ChatGPT is capable of is utterly mind boggling. Humans worked on trying to teach computers how to understand natural language ever since the very first computers 80 or so years ago, without much success. Even just a simple automatic spell checker that actually worked was elusive. ChatGPT is so f'n good at natural language that people don't even realize how hard of a problem that is, they just accept that it works and don't think about it, because it's basically 100% correct at understanding language.
ChatGPT is a text auto-complete engine. The developers didn't set out to build a machine that can think, reason, replicate the brain or even build a chatbot. They build one that tells you what word comes next. And then they threw lots of data at it. Everything ChatGPT is capable of is basically an accident, not design. As it turns out, to predict the next word correctly you have to have a very rich understanding of the world and GPT figures that out all by itself just by going through lots and lots of texts.
That's the part that makes modern AI interesting and a scary: We don't really know why any of this works. We just keep throwing data at the AI and see what sticks. And for the last 10 years, a lot of it stuck. Find a problem space that you have lots of data for, throw it at AI and get interesting results. No human set around and taught DALLE how to draw and no human taught ChatGPT how to write English, it's all learned from the data. Worse yet, the lesson learned over the last decade is essentially that human expertise is largely worthless in teaching AIs, you get much better results by simply throwing lots of data at it.
That is utterly meaningless. OpenAI is constantly tweaking that thing for business reasons, including downgrading it to consume less resources and censoring it to not produce something nasty (Meta didn't get the memo). Same happened with Bing Chat and same thing just happened with DALL-E3, which until a few days ago could generate celebrity faces and now blocks all requests in that direction.
When you compare GPT-3.5 with the new/pay GPT-4, i.e. a newly training versions with more data, it ends up being far superior than the previous one. Same with DALLE2 vs DALLE3.
Also note that modern AIs don't learn. They are trained on a dataset once and that's it. The models are completely static after that. Nothing of what you type into them will be remembered by them. The illusion of a short-term memory comes from the whole conversation history getting feed into the model each time. The training step is completely separate from chatting with the model.