It was an absolutely dog water, mealy-mouthed response. I’d have rather they just straight up said they’re on board with AI (because they clearly are based on this and previous behavior) and won’t be addressing any concerns because that would have been more honest at least. 95% of the community wanted disclosure of some kind, and rather than even TRY to define what that “line” is, or god forbid maybe even gather a consensus, they just told us to go swivel.
3 Likes
I think also that VRChat needs to add an AI label on Steam because from what they’ve stated in previous announcements about this, they use AI themselves. Not disclosing it is against the Steam TOS. I’m not talking about user-made world stuff, I mean the client itself as well as possibly any worlds they’ve made (cough Spookality cough). I dunno, maybe this is a topic for a new discussion post. I feel like I’m watching VRChat fall apart in real time, honestly.
7 Likes
Nothing against the first post on this thread specifically or to anyone, but i believe we should all take some time to chill out for a second.
I see a lot of eletricity in the air and lots of people going out in out to rant.
Nothing is changing overnight and everything is out to change. Lets all be civilized
2 Likes
ai actually triggers my Uncanny Valley quesiness, like I aint even lying its something that triggers me when something looks horribly wrong and disgusting, alot of adult cartoons give me the same feeling in my stomache because some of them have gross designs as just a norm. It should absolutely be something i can CHOOSE to turn off if I can. I have to literally watch Smiling Friends with audio only because sadly as much as I love the show the designs sometimes do that to me.
2 Likes
Yea, let’s just embrace the slop. Who needs user created content when you can just have everything be created by AI?
We should not create, creating is for machines.
Humans are made to consume. Just let us all feed on the slop. No effort involved. Just let the robots make everything for us.
1 Like
So you want people to harass others with the spam label.
I was thinking, can the community do better to understand the needs of the community?
I’ve personally thought VRChat may need to survey more questions, to understand the community’s problems and needs with discovery deeper than at surface level. Prompting good survey questions is hard, as is defining the scope who should answer a survey.
For example, the 95% respondents wanting some sort of AI label didn’t explain to me why the respondents wanted it. And it didn’t explain to me why 33% wanted to avoid AI-assisted worlds entirely, or why 37% were fine with AI-assisted worlds with sufficient human authorship.
It may sideline if VRChat has a problem with AI content being promoted, or a problem with low-quality, low-effort content being promoted in discovery.
I wish VRChat would survey the community with more questions. I understand it’s not a good idea to skip straight to sensitive questions, but there are some I’d like answers to:
- Are “low-quality content” (low-effort) and “AI-made content” the same problem to you in VRChat? (Mostly the same / Related but different / Different / Not sure)
- If a creator used AI somewhere in their workflow, does your comfort level improve or decrease by disclosure of AI or tools used (without filters)?
- If VRChat added a like/dislike system, how often would you use it?
- What would concern you about a likes/dislikes system, if anything?
- Would an option to report spam-like (low-effort) content in general be a sufficient solution?
None of these questions measure what counts as quality or high-effort, which may require additional questions.
Depending on who you ask (world creators or world visitors, AI users vs. not, and creators harmed or not harmed by disclosures), the answers may be different, especially regarding AI or tool disclosure.
This would be an entirely new topic for another discussion to seek answers.
2 Likes
As someone from SL (who has been in VRC before many times, don’t go shitting in my cheerios just yet) let me provide a perspective on something important - I think we also have to address the elephant in the room of how this should be implemented.
As the announcement says, We overwhelmingly agree that AI labelling should be a thing. As someone whose strictness on AI has loosened, but still looks down upon those who put out slop or have in majority given up authorship to the AI - My perspective on what that label would look like, definitely doesn’t find itself anywhere near the firmly Anti-AI camp. But even then I strongly believe at the core - transparency is paramount.
The reality is there were over 3,000 different combinations of enforcement and disclosure policies brought forth in the survey as acceptable, and I can tell you right now due to how polarized both sides are, that the amount of viable policies out of that 3,000 is very, very small and still retain exploitable methods to engage in harassment - something that has repeatedly pushed me away from the Anti-AI crowd’s most ardent members because of their rabid fervor.
I think this is the best we can realistically ask for, and that building upon it is best. As much as hiding dislikes as an idea is controversial, as long as it is solely an impact on discoverability, I think the system will still reward genuine creative works and the works of those who actually put effort forth - AI or no.
1 Like
Comparing AI to child pornography, is to be quite blunt, despicable. One’s a morally and ethically controversial technology at the very least, and the other’s an illegal medium of photography and videography that contributes actively to the trafficking and abuse of minors.
I said this to a friend - begrudgingly:
AI is here. The genie cannot go back into the bottle. You are welcome to not like it. But it is here to stay, invariably, even if tomorrow, the AI bubble burst happens.
That being said, having reviewed it - a system like this would be an excellent way to push for transparency on authorship at a glance.
The comparison doesn’t matter. The point is that boiling it down to a preference is overlooking the broader issue.
AI being used for creative things, especially VRChat worlds, changes the entire social dynamic of everything.
Maybe someone makes something that passes as human created. What happens when the creator doesn’t disclose that at all? What happens when people reach out to that creator for opportunities, but through communication and working together, people learn that they’re actually incapable and that all of the work they did was through the AI?
I don’t think people really understand how much AI hijacks the creative process. It doesn’t matter what the output is and if you like it or not. What matters is that you put in the effort and that you actually learned along the way. What matters is that through the process of creating the thing, you struggled enough to actually change the outcome based on what you think and what you do. This is the entire filtering process that makes worlds what they are.
It removes a significant part of the human from the process. It changes the creative culture from people who actually have extreme passion behind their work, to an entire platform full of slop.
And sure, something can be high quality, but if it doesn’t have a soul, it’s still slop.
It doesn’t matter about the material quality of the world. It matters about the human intent and human effort put behind it.
The entire point of this platform is, “VRChat”. “Chat” is a social feature. What is the point of a platform where all of the communities are surface level and not actually deeply inspired by art and creative people?
What is the point of a platform where all of the output of its users is bottom of the barrel slop that can just be used to turn a revenue, instead of tight-knit communities that actually care about each other and the art that they produce?
I don’t care if people use AI. I use AI. I am actively trying to use AI less because I’m recognizing how much it has hijacked my creative ability and ability to pursue hard problems. What matters to me is that a human spent the majority of the time working on the project and actually putting their heart and soul into it.
Maybe there is a future where people who use AI can still do that. But what’s evident now is that the vast majority of AI output is low quality, or otherwise uninspired and commodified slop that would only serve to flood the platform experiences that could be monetized. It commodifies art in a way that disrespects the entire Creator Community of this platform that made it what it is today.
VRChat should never forget that the users who created all these incredible worlds are the ones who actually got them here. If it wasn’t for them, they wouldn’t have the $95 million of investor funding.
2 Likes
I think there’s a fundamental misunderstanding of those us who really don’t like AI, what is meant by that, and why we want to filter it out. So, I’d like to try and clear it up, as hopefully it’ll be more apparent why the recent announcement is really tone deaf. Is this post opinionated? Absolutely, but I’d at least like to think it’s educated given my experience working in IT, dealing with companies who are heavily pushing for AI from my metaphorical vantage point at a B2B level having to deal with a vendor that’s all-in on AI (read: Oracle), and doing art as a hobby.
First off, nobody is corralling against AI machine translation to do stuff like bridge the English- and Japanese-speaking audiences for people who aren’t bilingual. When “AI” is talked about in controversies these days, what’s meant by that is generative AI specifically as opposed to more traditional forms of AI like computer vision or whatnot- those ones aren’t a bane to anyone’s well-being either, nor does it reward bad behavior.
Secondly, I often see people reduce artists’ opposition to AI to simple matters like not wanting to compete with AI’s speed and opinions on “soul” or whatnot. It’s way more complex than that, but it boils down to 2 things: incentives and rewards. What drives someone to put in all that time and manual effort to learn all the skills required to do their craft when there’s much faster ways? Well, first and foremost, a degree of passion and care about the specific field of creation and an opinion on the what way to do it is the best. It requires enough care that the time and energy spent over time for the high degree of control to get it the way they think is best is worth it, even if they can’t achieve it right away. The rewards? You get something that appeals to your specific tastes, which are specific to you as an individual. This is how we get variety- things that are new and interesting beyond shallow trend-chasing made for a quick buck or clout on social media. Things their creators believed in even if they’re not the current popular thing. And how do we keep getting to see these things? Why do they keep getting shared with us? They don’t have to be shared. It’s because of the rewards of recognition and/or payment for those efforts to share these new things with everyone.
This leads us to the first major problem with generative AI- it provides very similar rewards without requiring any of the effort or care, encouraging people to churn out a flood of AI-generated content at an extremely fast pace (see: DeviantArt) without providing an understanding of what’s being done- and even worse, while also taking a lot of control away from the user. In fact, it actively discourages people from understanding and refining any kind of craft. I think the inherent lack of understanding and care is very apparent in vibecoders or people who otherwise use LLMs to “help them learn how to code.” I earned a degree in programming. That’s my field of work. I can say with confidence that Claude, Codex, and every other AI programming thing write poor-performing spaghetti code- and for new users, they can’t really see why, just that it appears to work. This teaches bad practices. The excuses for bad code I’ve read from vibecoders are also patently horrible and make it very apparent they don’t understand what they’re doing. Similarly, a person who uses diffusion models to generate images doesn’t learn what makes an image composition or character design appealing and doesn’t have the wisdom to improve upon it- nor do they have incentive to. It’s always “well it works” and “it saves me so much time” rather than any kind of concern for the quality or merit of the output- that is, generative AI rewards people for not caring, with caring being a major component of making things new and better. Heavy AI users simply don’t have the drive traditional creators do. The instant gratification provided by these generative AI products discourages being inquisitive because they can just get something “good enough” without having to put in much time or effort. I can hear people say, “well, these models think and learn like people do!” No. The idea they “think” is marketing babble from the companies that produce them, which in their own Terms of Service can’t guarantee any results other than entertainment. The way in which LLMs “think” is concurrent runs of the same model with a different seed to check for consistency (not correctness), and diffusion models are just an averaging out of the properties of their training data. They don’t have taste or preferences like people do which drive them to make something new, just regurgitate weighted averages of their input.
And with the onslaught of AI content flooding all corners of the Internet, what incentives do people have to share their hand-made creations with others when the flood of slop will make it hard to find and/or buy and as such get buried? Human-made artwork can still be posted, right? Nothing’s stopping that- but why post something for everyone to see or to sell when it won’t be found? The result is there’s less new and original things being posted, when they are posted they’re difficult to find, and the creative atmosphere becomes stagnant and shallow. At the end of the day, these AI models require new input and eventually they’ll run out and stagnate if they have no new training data, and the artists that unwillingly provided it no longer share their work with the public because of them, nor will new people have incentive to become artists or programmers (worst case scenario).
“The point of the system is what it does” can be applied very accurately to generative AI as well. Disincentivizing actual legitimate creative expression and individuality is by design. All of the lead people behind major AI companies follow a very nasty set of ideologies which involve subjugating people to be docile to enable their blatant disregard for the human condition to chase what even they acknowledge is probably an impossible sci-fi fantasy, with AI being the means to that end by conditioning people to not think for themselves so they’re easier to manipulate (more info in the linked paper, authored in part by people formerly involved with AI). Really, it’s not just an attack on artists, but also our own agency as people if you really need more reason to not want to interact with it beyond just hand-crafting vs. mass-production. And to address the most common thing I see about stopping this process, which is usually along the lines of
AI is here. The genie cannot go back into the bottle. You are welcome to not like it. But it is here to stay, invariably, even if tomorrow, the AI bubble burst happens.
All of the infrastructure and technology behind NFTs and other previous tech investment scams still exists and is “out of the bottle,” yet isn’t nearly as relevant or as much of a threat to ordinary people because the investors who fell for the scams realized what was happening again. Generative AI is no different- an elaborate unprofitable tech investment scam of empty promises that was supposed to be The Future just like the blockchain. The companies pushing AI heavily know their target audience for funding is tech-illiterate John Retirement who lived off of inheritance his whole life and can’t open a web browser without his grandson doing it: easily persuaded by fancy smart-sounding technobabble as he can’t verify or reason about any of the claims these AI companies make until their lies finally bite him and he’s losing money. We’re seeing the same cycle as with crypto: it was starting to be apparent it wasn’t working out, so the investors tried to use fiduciary duty to force the tech into everything to create artificial demand to try to boost return on investment; they basically demanded companies use the products and services for crypto creating an artificially large market for it. They are now going past that stage with AI and starting to realize they’re losing money. The AI craze has only lasted as long as it has because it’s flashier and gives a better illusion of utility than an unregulated stock market like crypto was presented as, the illusion of which was shut down by simply regulating it. As out-of-the-bottle it may be, at the end of the day, people can stop interacting with the genie at any time- especially devastating to the genie if the people providing all its housing and dining stop doing so.
So, yes, people have very valid reasons to just not want to interact with AI content (or keep it to an absolute minimum) and filter it out, and there are very good reasons AI users get a bad reputation. Is it grounds for harassment? No. Is relying on people to self-declare reliable? Certainly not. But does that invalidate the underlying logic behind the desire? Also no.
2 Likes
I’ll say even as a vehement AI hater I respect and applaud your observations and the steps you’re taking to reduce your AI reliance and I don’t shun you for having used it in the first place. It gives me hope for other people making similar observations too.
The problem is how to effectively filter AI-generated content. You always say AI should be restricted or banned, and I completely agree. But I think the authorities have actually already listed the biggest problems. How can we build an effective AI-generated-content filter? And how can we prevent people from being wrongly harmed by it? Once the filter becomes ineffective because of AI’s rapid iteration, there will be no small number of people who get wrongly harmed. Making it hard to distinguish AI from human-made content has always been one of the directions of AI’s evolution, and this is completely the opposite of pornographic content. Pornographic content needs to manifest as pornographic, and its identification has relatively clear long-term standards, so people can always identify the vast majority of pornographic content at a glance.
But AI generation is moving toward becoming more and more human-like, and the AI features of asset creation are decreasing every day with training. And most importantly, if you think AI is unethical—and I agree with that—then you should go demand accountability from the companies that train AI. Instead of expecting a game platform to spend no small cost to build an effective filter and continuously study AI-generated features to update the filter to prevent it from quickly becoming obsolete, just to satisfy your moral demands, just to restrict AI-generated content that is far less harmful than real crimes such as child pornography and terrorism. And although I also hate AI—it has sharply increased the competitive pressure on my work—and wish I could destroy all AI in the world, do you think it is possible to rely on pressuring a game platform whose technical strength is not particularly strong to restrict AI? From painting to mathematics, there are many industries that would love to flay AI alive. Have they not wanted to unite to issue an AI restriction policy and detection standards to protect their own livelihoods? They have tried, but it is hard to do. To be honest, it would be much more practical for you to go protest against companies like OpenAI and other companies that train these AIs.
For the record, I understand VRC’s viewpoint on why they can’t make a filter- and I agree with it! The technical debt to make something like that in the way people want is absolutely massive and infeasible. I just think the reasoning behind the desire for a filter to begin with is oversimplified and too easily dismissed for something less than it is, when it’s actually very complex and multifaceted. I read a lot of other posts on the subject before making my wall of text, for the record. The social implications of a filter would be devastating and a lot of resources would need to go into making such a system, as you mentioned, and as we had both mentioned, leaving it up to people to tag or identify it themselves is unreliable (and worse, would cause a lot of fighting).
However, it feels like VRC is going out of its way to promote and legitimize generative AI despite it spitting in the face of the hand-done work of creators which make VRC an attractive application to begin with, which is what feels tone deaf to me. Use of AI by the VRC devs is also really concerning given the security nightmares it produces (there’s already problems with avatar stealer bots as it is).
2 Likes
AI is indeed bad, but you have to admit that its productivity boost is simply too high. For platforms, such a high-productivity tool leaves only two options: become an opponent and resolutely oppose the vast majority of AI, or embrace AI’s productivity and keep following up. Obviously, VRChat is, after all, just a game and a small platform, so it does not have the ability to resolutely oppose AI. For VRChat, and even for many other companies, the only option to avoid being eliminated in the future is to embrace AI. There is simply no middle option. I believe the root of all evil lies with the AI training companies: it is they who created AI in an unethical way, which has led to so many difficult-to-handle problems. In essence, the VRChat platform is also a victim.
1 Like
Even this only-two-options choice is the same for most of us. No matter how much we hate and oppose AI, because our individual power is too small, we can only choose to embrace AI when faced with pressure. Cursing AI as unethical while being forced to use it—that is the reality.
Isn’t that exactly what those wailing mathematicians have been doing lately? A subject they’ve studied for over a decade as their livelihood, completely cracked open and finished off by AI in just over ten hours. In that situation, those mathematicians come out and issue some feeble condemnations and protests—but in the end, don’t they still have to obediently start using AI to assist their research?
bruh.
1 Like
I can kinda see your your point about AI having different uses, and I agree that there is a distinction between using AI as an assistant and relying on it to generate practically everything for you. However, I think this distinction is exactly why VRChat needs clearer rules and transparency. My concern isn’t necessarily that someone opens an AI tool once or twice. It’s about the growing tendency to use AI as a replacement for actually learning and developing creative skills.
For example, there’s a considerable difference between someone using AI to help them understand a Unity script, explain why something isn’t working, or learn how to implement a feature, and someone simply prompting AI to write an entire system for them, copying the result, and having absolutely no understanding of how it works.
The first person is using a tool to develop their knowledge. The second is outsourcing the learning process altogether. And I think this is becoming an increasingly common attitude. From what I’ve personally observed, many people nowadays seem to want everything immediately. They don’t want to spend hours learning Unity, experimenting with Blender, understanding programming, troubleshooting errors, or developing their own techniques. They want the finished product as quickly as possible so they can move on to something else.But what’s the value of creating something when you haven’t actually developed the skills to understand or maintain what you’ve created?I also agree that AI-generated content shouldn’t necessarily be treated as one single category. That’s precisely why I believe VRChat should introduce clearer disclosure and labelling systems. Worlds should have visible AI tags that distinguish between AI-generated assets, AI-assisted development, and worlds that are predominantly or entirely AI-generated. This would allow users to make informed decisions about the content they’re visiting, especially if VRChat ever decides to permit AI-generated worlds in its Creator Economy program. I’d also argue for some form of creator disclosure regarding AI-generated code. There should be a distinction between developers who use AI to learn and improve their programming skills and those who rely on it to produce code they don’t understand. AI shouldn’t become a substitute for understanding the systems you’re introducing into your creations. Regarding your suggestion about flagging spam worlds, I particularly like the idea of making them less visible instead of immediately deleting them. However, I would also like to see proper categories and community reporting mechanisms, so users can understand why a world has been flagged. Ultimately, I don’t think VRChat should encourage a culture where the amount of content produced matters more than the effort, creativity, and knowledge behind it.
AI can assist creativity, but it shouldn’t replace the creative process itself.