Does anyone know of any attempts out there of building more ethical AI models for things like text LLM, image or audio processing?
To clarify, I am talking about an AI model which is trained on a dataset of public domain, creative commons, licensed, or more intentionally fair use sampling methods.
Personally, I believe that the way the modern AI industry has built their giant models is incredibly exploitative, unethical, toxic to society and culture, and arguably not even consistent with the letter or spirit of the existing rules by which we have all been asked to play.
As a tool, machine learning may not be fundamentally bad, but who, if anyone, is out there trying to do it in better, more legitimate way?
I feel no one answered the question and I see in the comments you are not super concerned about the capabilities listed. Anyway this is what intrigues me https://pleias.ai/about
I don’t really think IP needs to be protected and private. The problems with AI are its implementation under capitalism above all else, not the form of ownership of the data it was trained on.
I don’t really think IP needs to be protected and private.
For the most part, I agree with this.
For certain it has strategic purposes, we do not live in a communist society globally or anything. However, I don’t find IP itself to be a virtue.
Does anyone know of any attempts out there of building more ethical AI models
There’s Apertus. There’s some concerns about its data pointed out on the repo still, but since you’re just asking about attempts and more ethical, I think it still fits perfectly.
That this post is 1/3 disliked really just shows how this crowd feels about this tectonic shift in technology. You even ask a simple question like, “more ethical attempts?” and if you dare add the two characters A and I, some folks are just immediately mad. So silly that people are discouraged from any sort of nuance on the topic.
yeah its kinda funny because I hope to see local llm ethically trained get wrapped into os distros to provide a more star trek like computer experience. I certainly see problems with ai/llm but I recognize it can have uses as well. I really don’t get the extremes on this tech. I will say though that like early on there were folks that did not like guis and to be fair the guis were really crappy initially.
nah, fuck AI
EDIT - I didn’t want to add any context but here we are. tech moguls have made it clear that the goal of AI is to eliminate workers. they’re saying the quiet part out loud that there really isn’t any upside for your average person on the street. you’ll lose your job and that’s good. no, of course no UBI you lazy lout, just stop being poor.
second, the science has shown that it’s making people stupider, as their critical thinking skills and even simple abilities like reading dwindle the more they use it.
I say “fuck AI” with full awareness of the nuance. I’m forced to use it daily for work and I hate it. an ethically trained model does not fix all of these other problems. its primary purpose is a tool of oppression.
Should use LLM instead of the marketing term
In lieu of what term? LLM is a subset of Transformers which itself is a subset of ML models. So you should really only use that term if you mean that narrow subset. That would leave off diffusion models which itself has maybe the lion’s share of ethical concerns regarding stolen art.
Instead of AI, because its definitely not intelligence as you yourself just explained
But that would be imprecise in this context, since the poster is referring to at least both LLM’s and diffusion models. AI doesn’t have intelligence; that’s what the “A” means. And it is a suitable term for the broad set of technologies that you and I both know it means and has been in use for decades predating the LLM craze.
You know, that’s an interesting point and I’m not totally sure how to read into people’s dislikes…
I’m probably amongst the most vocally anti-AI people I know. I fully understand that basically all of the biggest AI models today have been built on the exploitation of other people’s (be it writers, artists, musicians, programmers, etc.) stolen intellectual labor. And that’s not just my gut feeling either, it’s been well documented at this point by higher ups at the tech companies themselves. I can’t stand the AI industry, and I would argue it’s pretty much morally indefensible as things are right now. The AI bubble and the way the technology is being used, I think, is truly terrible for society and culture.
But it’s that exact understanding–that the AI industry has been built on a foundation of theft–that makes me wonder if anyone out there is trying to do this in a less brazenly unethical way. I can’t help but be curious what all of this stuff might look like if people took a more fair approach to building datasets and training models from them. Is it even possible?
To me, part of criticizing the way things are is considering the alternatives so that we can make an informed decision on the way things should be. I’d love to know if someone out there is doing AI “the right way” and leading by example when it comes to ethical datasets.
I agree completely. What happened since backpropagation, especially with transformers, is amazing. But I have no trouble coupling that with the view that mega corporations pushing it so hard is a continuation of the wealthy exploiting the poor, harmful to the environment, and built on a throne of lies.
Not only that, but the train has left the station. I would love to find the best path forward whatever that is. Part of me feels like that is with some criminal accountability for all the theft at least, certainly heavy regulation, but maybe also completely nationalizing the technology so it can make some decisions based on public good rather than strictly profit.
I haven’t done too much research into it, but there is the Swiss Apertus AI which is part of the Swiss AI initiative. It claims to be “fully open” and “responsible at scale”. They’ve partnered with Proton/Lumo for opt-in training. Maybe worth looking into?
a self hosted one? a large scale one like the well known one are highly polluting and socially destructive.
I can’t say that I know of one off my head, but I do know people publicly release data training sets from public sources so you could technically train one. Smaller models wouldn’t require gigantic data centers to train. Not sure if that public training data counts information that is publicly available on the Internet but something we’d still consider “stealing” from people: forum communications, online art galleries, etc…
Yeah, sadly I think the word “public” is often stretched to mean “anything I can get access to regardless of ownership or license”.
I’m really mainly interested in any efforts that are trying to do things in a more legitimate way with regard to consent and licensing.
Like, I suppose the question I’m trying to answer is, if one were try train a model on public domain or at least fair use, how useful would it be and what could it do?
Supposedly, Adobe’s image generation models were trained exclusively on images that are either public domain or which Adobe specifically owns legal rights to.
That’s not fully ethical, probably, because a lot of the people who made those images and signed the rights to Adobe probably weren’t aware of the ramifications of doing so, and may not have agreed for them to be used in AI training if they were given an informed choice in the matter.
But if their claims about their training data are true, then at least you can rest assured that it technically isn’t violating anybody’s legal copyright.
it’s not even clear if the current approach can reach actual artifical intelligence from a machine learning approach. this is a randomized regurgitation machine, it hasn’t met the markers of an actual AI, it’s just that the profit motive demands it must be ready and it must be the right path because they need that money.
this is a randomized regurgitation machine, it hasn’t met the markers of an actual AI,
Eh … yes and no. Depends what markers you’re looking for.
But at the very least, it’s pretty good at pretending to be actual AI sometimes. And a chatbot passing the Turing Test has gone from an occasional fluke under very specific conditions to being something so commonplace that you’ll often deal with it in your everyday life.
Fair. I guess I’m curious if any people who exist outside of the profit-driven investment hype “AI” bubble are trying to see what can be done with more legitimate sources of data.
It’s possible that nobody is even trying it…
Hey so my research was on AI. Your specifically talking about LLMs which is a very very small subset of AI. There’s a metric TON of public LLMs that are specialty made. The more recent one was a person who made a tiny one just for one off bash scripts based on his own massive 20+ years of dot files. Hugging face has some but others are just all over the internet. The training sets ate usually public sources. Creative commons or data specifically for that purpose.
Most AI work is happening at university and open source levels.
I don’t really mind if it’s LLMs or other “AI” models for things like image or audio processing. My question is really about whether there are any noteworthy models that have been trained off an more ethical dataset than the big ones that have been trained on massive troves of scraped data.
The more recent one was a person who made a tiny one just for one off bash scripts based on his own massive 20+ years of dot files.
That’s interesting.
The training sets ate usually public sources.
Does “public sources” mean public domain, or just like visible on the internet?
In research (as in papers) you have to use curated results or peer reviewed datasets. So its a grab bag. But some are ONLY available in acadamia. Its a strange world. The stuff AIs are doing are novel but there is many more nitch stuff for very specific reasons.
For example my research was on using a variant on the Probabilistic Latent Semantic Analysis for specific applications (cancers/medical applications). It was trained on datasets and proven to work. It was 90percent on other peoples work and was more a proof of concept. And there was work for the multi-model version at the time.
Datasets were all over, some public domain, some very specific medical data with permission.
