On Friday, April 18th, Linden Lab Hosted a Town Hall meeting with Philip Rosedale to discuss AI & the Future of Our Virtual Community, defined in part by the Lab thus:
As Artificial Intelligence (AI) becomes increasingly present in both our physical and virtual worlds, the Second Life community is coming together for a Special Town Hall event to explore its evolving role and impact within our shared metaverse.The meeting was livestreamed to You Tube, where the official video is available with transcript. This following is a summary of topic, comments, presentations and responses in the order presented. Two videos are also provided:
|
Table of Contents
|
- Timestamps in the text primarily reference the official livestream video, with timestamps to Pantera’s video clearly denoted as such.
Philip Rosedale – Opening Statement
Welcome everybody to our Town Hall … This is our Town Hall on the growing, evolving role of Artificial Intelligence in Second Life. It’s good to see some many people here, I think engaging on this with us is one of the most important conversations of our time; how AI is shaping the metaverse; how we live with it here.
We’re not here today to make any announcements; this is a Town Hall, and we’re going to run it as one, and try to establish a pattern where we can do this again and again, when we need to, about things. … So we’re not going to announce anything, because we’re here to listen; this is going to be an open conversation. If governance or policy or software changes come out of this discussion, we can try and act on them – and we will. But at the start, we have no new announcements to make.
Brett Linden – Structure and Flow
- The meeting is a two-way dialogue and exchange.
- There was a response to the opportunity for users to give pre-prepared comments [7 in all]; those people have up to 3 minutes to present them.
- There were also pre-submitted questions and feedback [68 in all], and an aim is to address as many of those as possible, time allowing.
- As this meeting is specific to AI, people were asked not to ask questions on other matters pertaining to / aspects of Second Life.
- A request for orderly use of chat and open microphones.
- LL recognises that AI is a polarising topic, but request that comments and feedback is kept constructive and solutions-oriented.
Philip Linden – Major Usage Areas for AI in Second Life
Notes:
- This includes actual and prospective uses of AI in SL, and some of the feedback from users submitted ahead of the meeting.
- The breakdown of items is an arbitrary one on my part, to try to contextualise the uses.
AI Tools to Provide General Assistance
- Finding duplicates in Feature Requests and bug-reports in the Feedback Portal [Canny].
- This is the only way AI is being used in the Production environment, and is considered by LL to be a good addition to SL.
- This takes the form of taking newly-entered items and – where appropriate – merging them with matching reports / requests also submitted via the Feedback Portal or held within the JIRA archive. This would be next to impossible to do at the human level.
- All submissions to the Feedback Portal are still human-reviewed [triaged].
- Providing assistance to new users in a range of tasks – directly answering questions, etc., as has been trailed in some Welcome areas.
- Planned future use of AI:
- Translation and closed-caption tools [voice-to-text and text-to-voice] – seen by LL and many users as the “number one” accessibility improvement feature.
- Not in development, but worthy of possible consideration:
- Assistance with Inventory management [user suggestion] – using an AI tool to help structure and organise large inventories.
- The potential for AI tools to help with shopping on the Marketplace [e.g. to build a specific avatar look] by carrying out searches and presenting suggestions.
- In-world text-to-content generation tools – using text chat to describe an object (including Animesh or an avatar) and having AI perform the basic construction.
- Identifying harassment, bullying, hate-speech, etc., within text and/or Voice chat.
- Finding cases of possible copyright infringement in a volume impossible for humans to do (albeit it with recognised potential issues).
- As with the current physical world trend, the creation of a romantic AI-driven partner – with the recognised questions as to whether it would in fact be broadly compelling to people + the ethical implications.
Current Uses of AI Tools in SL (by Users)
- Use of 3D content generation [e.g. Meshy] tools to create Marketplace / in-world content – recognised as being very contentious and the subject of great debate.
- Use of 2D image generation tools [e.g. Midjourney] to create marketing materials for content in-world / on the Marketplace – also recognised as being very contentious and the subject of great debate, but seen as distinct from actual generative content creation.
- Creating companion avatars [e.g. through the use of the Convai –driven Character Designer).
- Creation of more dynamic Non-Player/ing Characters (NPCs), for use in role-play environments, etc.
“Philip’s Crazy Dream”
- Having avatars “learn” through how they are used when logged-in, what they say, and have them “live on” in-world as a presence when the user logs-out.
- Acknowledges it is a “crazy, fun, Black Mirror idea”, but wanted to include it as an illustration of where AI might potentially go.
Responses to Comments Submitted in Advance
- Is Linden Lab currently training, planning to train, or knowingly paying another company to train any AI models with user-generated content [UGC] previously uploaded to Second Life, including but limited to: textures, mesh models, shapes, scripts and chat logs?
- The Lab has yet not trained AI on any of the content in-world.
- This is a community discussion on AI, and as such, given SL’s success has been in part based on the explicit and clear statement that Linden Lab does not own any UGC within the platform, training AI on such content is not seen as a decision LL can arbitrarily make.
- There has potentially been unfounded concerns raised on this as a result of some misunderstanding of the role of the partner company to LL in developing the AI capability used within the Feedback Portal system, as mentioned above – however, this work only utilised public information within the Feedback portal system.
- When will Convai bots [characters built using the Convai tool] be allowed on parcels in Bellisseria.
- Bellisseria has been established as a community with a specific covenant which states No Bots. As such, properly marked Convai characters are not allowed in those regions.
- If all [a majority?] Bellisseria residents want to have Convai characters within the regions, then LL are open to finding a way to taking a vote or similar on the subject and updating the Bellisseria covenant.
- On a broader context, LL should be looking to residents / communities to make these decisions, rather than the Lab wants to do.
- Bots and AI characters / AI content should be clearly marked as such (e.g. with the creator’s name).
- Aware of a third-party blog survey in which 53% of respondents agreed with this idea.
- Currently, there is no automated way to easily identify AI content, and such is the pace of AI development, that any tool or gadget created for this purpose is liable to be obsolete in 6 months.
- Further, given SL is an open world where those coming in do not have to give personally identifying information, makes pursuing them in the event of a gross misuse of AI difficult.
- However, LL is open to discussions on the issue and possible solutions.
Resident Statements
Sailor Paü (HonePau)
When we come into Second Life we’re not just bringing an avatar we’re bringing our thoughts our moods our intentions. For a lot of us this place is more than just a game or a creative outlet it becomes emotional it becomes real in ways that’s hard to explain to some people outside of sight of it. So the question of whether AI can have a soul here isn’t just about machines or scripts, it’s about presence and about connection.
When I’m thinking about what makes something meaningful in Second Life it’s not about how detailed your mesh is or how smooth the animations are it’s about feeling something that it gives you and maybe a build something someone made years ago an old sim that you visit; a forgotten gallery; maybe a memory; and that makes us start to think about the soul so it’s not a religious sense necessarily, but in a deeper emotional way um the part of us that it makes us connect mostly.
An AI can learn to replicate things so it can speak in our tone it can recreate scenes, mimic dialogue, outfits, pictures, something you say … And I guess everyone’s just curious about where it can go next especially since we’re all singular individuals functioning separately. So maybe AI will continue to improve maybe it will eventually move and speak in a way that feels more human; but it doesn’t have that memory or meaning to it and it doesn’t ache; it doesn’t create from longing it doesn’t respond from presence or absence or carry something when you’re gone. So I guess that’s all I wanted to say overall when it came to Second Life AI and the soul, and being able to not detach that from yourself when it comes to wherever it goes.
William Gide – Questions
[Pantera’s Video: 29:03-34:18]
Note: the audio broke up quite notably during this part of the session, to catch the audio in full, please refer to Pantera’s recording per the above link.
- Many don’t feel comfortable with talking to “counterfeit people” [AI-generated characters], so what is the Lab doing to allow people to easily identify they are interacting with a large language model (LLM) automated bot?
- What’s being done to give land holders on Mainland the same ability to ban bots as available to private estates?
- In regards to on-boarding new users with AI bot assistance: what failure scenarios has LL discussed that would convince the company that LLM bots, at the price-point you’re willing to pay right now, are not yet ready for user orientation?
Philip Rosedale Response
- Better parity between region / parcel controls on Mainland to those on private estates: there is a number of open tickets on this in general (and specifically for Voice), and LL’s intention is to provide better moderation controls, and LL will look at open issues to help decide the order of implementation for better controls.
- In respect of question’s and Sailor’s statement: little doubt that SL is all about human connection, so the importance of discussing and deciding what to do with LLMs that pretend to be people “is extraordinarily high”.
- A question from this is how to distinguish people from AIs, given the noted fact that users are free to join the platform without giving any personally identifying information [which then might be used to distinguish human-driven avatars]. This presumption being that most people would not want to move away from this model.
- This is a question that goes beyond LLMs masquerading as humans, and can include things like the sale of illegal items on the Marketplace.
Nodoka Hanamura-Nu Vaughn (Rathgrith027) – Text
[Pantera’s Video: 36:02-39:55]
For nearly a quarter of a century we have called this place home. We have created, we have cherished we have toiled and desired. We have yearned and developed ourselves in the fire of creation, from which Second Life, Agni – our home, named after the very Hindu god of fire himself!
I have to ask you: what gives Agni its fire? What gives it the light that radiates across its grid? It is the creative minds of its children! Children who bleed, sweat, and toil in works of mind and body to create this world we call home, in every respect. Who use digital tools to create facsimiles of a reality we wish we could only have in the physical. To live lives that those bound to a chair and bed, those without voice, without sight, without sound, can live a modicum of a normal life!
And what life is it, if it is not made as a labour of love, a labour of advancing oneself? All things are done through struggle, both big and small – both inflicted on oneself and inflicted by others! What point, what purpose, is Agni to create – if a machine creates for us? Exists for us? Thinks for us?
These artificial intelligences should exist to make our work easier – to obsolesce the undesirable and most painful of the struggles of creation and production not obsolesce the work and plant a façade of life! Do w not find it concerning that a machine creates for you, THINKS FOR YOU? At what point do we draw the line?
At what point do we say “no” to the repetitive slop that inundates our Marketplace? At what point do we say “no” to the manipulation of the advertisements of products sold to us? At what point do we say “no” to the people who have been blinded by the future and in insult to that all transhumanism is, have abandoned their humanity in blind pursuit of advancing technology.
I am not a Luddite! Neither are so many of us! But we know full well the perils of technology used without caution, without respect, without understanding of its consequences! I ask you, at what point do we draw the line? Do the Lindens draw the line? Do they draw the line in our mutual interest?
At what point can we even be certain that the person on the other side of the screen is a human, and not an algorithm that could be replaced with any other? At what point can we simply disregard creating for others and simply just create for ourselves, drowning our virtual home’s ability to produce a profit for not just its denizens that contribute to it, but its benefactors? I ask you, at what point will the Lindens actually open their eyes and listen?!
It is now, or it is never, for this discussion, and the decisions which follow will steer the ship [and] either see us through another 25 years if we are lucky, or see our certain demise in the next 5 as creators and residents alike abandon our home as a result of the continuing degradation of it!
I implore the residents, and most of all the Lindens, when you speak here, when you make your decisions, do so not in your own damned self-interest, but in the interest of all of us. For I implore you all, if we let these AI run rampant, they will take our very humanity from us. I cede the floor.
Philip Rosedale Response
I think there are many ways in which AI has dangers for us and I’ve written about this a good deal [and] I’ll put one link in chat for those who don’t follow my Substack writing. This one is about avoiding homogenization and I think it’s a great example of why there are many ways in which we don’t want to use AI to manipulate our culture and ourselves. So just really, thank you for that; that’s a very strong uh strong and appropriate perspective.
Ruby Maelstrom – Text
[Pantera’s video: 41:21-43:58]
I’m here to make a short presentation in support of the responsible use of generative AI in Second Life.
I have used AI as a starting point for many of my mesh objects. In fact, I likely would have never learned how to use Blender if I hadn’t had AI to provide some sense that I wasn’t starting from square one. Everyone, is not the same, but I personally needed that push to tackle such a complex learning challenge. After working alongside AI for a while, I now feel like I understand how to model in Blender, and I’m starting to learn how to rig mesh.
AI isn’t a replacement for humans in the creative process, but a tool like many others which can help people like me do things they may not have found the energy or motivation to get started on previously. It can be a starter for an engine of creativity.
Having said that, everything that I’ve made with the assistance of AI is marked so that people are aware that it was involved in the process. I believe that a practice of marking AI-generated content makes sense.
I also use one of the Linen-provided AI characters in my café, and it’s a popular fixture which a few people actively seek out to have a conversation with. An LLM character can be welcoming and friendly, and is strongly affected by what kind of personality its owner has written for it. When someone doesn’t feel fully capable of interacting with another person, an AI character can be a useful social outlet. In this way, I see AI characters as a sort of accessibility tool. It is also clearly marked as AI in multiple ways, and interaction with it can be avoided quite easily.
Thank you for listening to my perspective on some of the ways that generative AI can be used positively in our shared digital world.
Philip Rosedale Response
- Very much agrees with the point made.
- Notes that he’s been asked what tools he uses – ChatGPT, Grok, etc., and how frequently, and that he uses them most frequently for programming (outside of SL).
- Feels that AI tools help him as an experienced developer who is “very out-of-date with modern tools and technology, and he finds that AI is extraordinarily helpful in saving time using programming language to do something but would not otherwise be able to figure out the coding patterns, or which otherwise would take me a lot longer to complete.
- >Also notes that he also uses GPT or DeepSeek to seek answers to questions when he feels it would be more efficient in answering than a Google search.
Hazel (Dolly Waifu)
- Notes that content in Second Life represents the collective efforts of thousands and thousands of artists, creators and builders, many of whom have spent a lot of time learning skills in various tools and who depend on the platform as a source of income.
- Many of these creators already face enormous stress and pressure to keep up with the myriad events, etc., which constantly require them to turn out new content, sometimes on a monthly basis, in order to just maintain a place in the market and their income.
- Is concerned that generative AI merely ads to this pressure on a quantitative level, due to the volume of items that can be turned out with little or no effort / overhead.
- Requests that LL do not lose sight of the effort of all content creators in SL and provide tools that help uplift the artesian nature of their work above the use of generative AI tools, and help them to protect and maintain their work and their income against a perceived rising tide of less qualitative but far more quantitative content.
Philip Rosedale Response
- Would suggest that one of the big problems with AI isn’t that it can replicate human action in things like creating content, but the fact it can do so extremely easily and quickly.
- Also notes that tech entrepreneurs try and paper over the negative impact to presented by AI with the excuse that people will just figure out new things to do, which is an unfair statement, given the fact AI can do things with orders of magnitude of greater speed and reduced cost.
Zeke Onyx
- Speaks as a mentor and the operator of a community gateway on-boarding new users (Vidian Gateway).
- Is not in favour of AI adoption within SL, although concedes it could be useful, but believes it is being embraced far too quickly and without regard for possible dangers.
- Raises concerns that AI “mentors” in SL may (in keeping with William’s questions) have a high error / failure rate in helping new residents (e.g. being unable to tell if someone is wear one or three pairs of shoes).
- Requests the need for Marketplace content filters to remove AI content from search results, etc., for those who do not wish to view it.
- Beyond Second Life raises concerns about:
- Provides feedback on various AI usage / limitations:
- AI search engines outside of SL endangering people through providing incorrect information.
- References generative AI tools that take images, text, music without consent and re-use without attribution.
- Raises ethical concerns over AI bots data-scraping information – companies adding conditions to their terms of service that expressly allow them to scrape user data for AI training, taking away the right of choice for users.
- Notes the environmental impact AI farms can have on the environment (power use, etc.).
- States that use of AI is power without responsibility.
Philip Rosedale Response
- Offers a definition of “generative AI” for those who might not be familiar with it:
[It] is generally the title applied to those models called large language transformer models or stable diffusion in the case of image and content generation these models are notable in that they are effective only because they are trained on what need to be extremely large bodies of existing work and as just said, in many cases that work is not clearly allowed by those who have posted it. Indeed I am one of those who would agree that the assertion that information published on the public Internet can be freely used to generate new content by AIs – I don’t believe that is obvious. I think that we need as a society to have a conversation about what the rules should be.
- Note that this is also a conversion – about what the rules of re-use should be – for Second Life, even for data listed as “public” on the internet, as it is not clear what that means.
- Acknowledges that the legality of training AI on text, images, etc., via common Internet trawl or the use of large bodies of data, is still “up in the air”.
- As such, thinks the world should be “sensible” sensible about the extensive use of any such technologies while that debate is still on-going.
- However, believes it is a debate beyond the scope of Second Life, and requires the input of the “whole world” and big corporations like Google, Facebook and OpenAI.
- Agrees that it is not at all obvious that public data conveys the right for AI systems to take it and re-use it.
- Notes that every one of these arguments / concerns has use-cases on both sides which are complicated and we need to bear out what’s going to happen, and be aware that if we don’t use some of these technologies in some ways, there may be negative impacts.
- Agrees that existing LLMs are extraordinarily expensive and environmentally unfriendly – although, so does the Internet as a whole, with social media applications using “10 or 100 times” more electricity than current LLMs.
Rysa (lukaskw)
- Wants to specifically discussion generative AI – specifically in the sense of content generation large language models, including but not limited to: images videos text and scripts (which touches on the use of Convai characters). It does not include machine translation or similar.
- Would personally not see generative AI content in Second Life in any capacity, but understands a blanket ban is both unlikely and unenforceable.
- Instead proposes a (hopefully) simple solution for two areas of SL: Marketplace and avatar Profiles.
- Many have already commented on the massive influx of AI generated images and other types of content on the Marketplace being sold under all manner of names and categories.
- Would love to see a means to filter such content out of the Marketplace experience for those not wanting to see it.
- Would also suggest an enforceable requirement for generative AI content to be listed as such in a similar manner to mesh / partial mesh content.
- Understand policing this would be a huge task and something not easily automated at this time; but feels the ideas should be discussed.
- In terms of Profiles, with LL, various brands and residents all experimenting with AI powering full avatars, believes more than just relying on the goodwill of those writing the avatar’s Profile to indicate it is AI-powered.
- S personally in favour of a more “heavy-handed” approach – there should be a clear, unavoidable denotation of an avatar’s status at a scripted agent of any kind – so not just AI powered avatars, but any scripted bot system.
- Believes it has to be marked as such internally, so suggests the denotation should be made public through the Profile.
- Believes the overall, residents of Second Life should be given the tools that they need to make informed decisions about the aspects of the grid that they wish to engage with, and tools such as those suggested would help in this.
Philip Rosedale Response
- Suggests the idea of self-identified content on the MP should be submitted to the Feedback Portal (if not already done), as it is something LL could consider, noting that self-identification is not currently enforced.
- Notes that subjective identification of AI-generated content can be just that: subjective, rather than accurate. Therefore placing such identification into the hands of a tool runs the risk of false positives.
Dax Dupont
- Continues the discussion on generative AI and the ethical and safety impacts, citing articles about AI Chat bots used in the mental health sphere having a negative impact on those using them.
- Asks whether LL are convinced that their use of AI in the first person [via bots] in SL is safe for users, and will not “hide behind the standard liability claim of’ oh we’re not responsible of whatever comes out of this object’, because they believe there is a “non-zero” chance people are going to get hurt.
Philip Rosedale Response
- Points out that Second Life may be a pseudonymous environment, it is not anonymous, and that acts have consequences. There is a strong bottom-up governance trend.
- Thus, in general, and regardless of the use of AI, anything that come out of an avatar is someone’s responsibility.
- In terms of LL’s direct use of AI bots, emphasises that LL does not see itself as a content creator, so would not take responsibility like that; even tools like the Character Generator carry the statement that those using the tool are responsible for the content the character is creating.
- Believes that to create a living world like Second Life has to be “deeply bottom up”, meaning everyone takes responsibility for the things they are doing, and any use of an LLM would have the same responsibilities applied to an avatar emitting its signal as it would to anything else some might say through their avatar.
Arrow Njarðarson Strong
- Notes that he has used AI extensively to assist him in his personal life and with his Second Life as a creator / builder, including creating AI characters that have assisted him process his personal trauma.
- Does not believe AI “steals”, but rather learns “patterns like a student exposed to a library of human thought, and then generates new unique content based on what we ask it to do”, and as such, believes the creativity rests with the creator, and that AI is a “brush”.
- Notes that AI has personally allowed him freedom as a person living with anxiety and noise sensitivity, helping him to reclaim control of his time and energy, and his ability to dream again.
- Suggests that AI is another way of learning from humanity’s shared knowledge, and is not something to replace us but to continue the ago-old process of the accumulation of wisdom.
Philip Rosedale Response
[Pantera’s Video: 1:15:40-1:16:08]
- Notes that Arrow touches on the challenge that AI does demonstratively have very close copies of recognisable content, such as likenesses of Mickey Mouse or similar, because they are so common, it can reproduce them precisely – and that is one of the issues making AI such a complex topic.
Philip Rosedale – Closing Remarks
[Pantera’s Video: 1:16:08-1:18:44]
- Notes that the meeting has been recorded as a matter of permanent record.
- Believes that the discussion could result in proposals submitted by users through the Feedback Portal which might help / try to move things towards specific actions relating to AI in terms of code, governance and the SL Terms of Service.
- Would like feedback on whether people think such town Halls are a good forum for discussions on matters relating to SL.
- Feel as if that, given the state of the world at large, there is a potential to use Second Life as a demonstration of how self-governance can work.
- He recognises that LL as a company are running the servers and the platform, and so have an “unfair” advantage that could be “difficult to give up” – but given projects such as open-sourcing the viewer, the open nature of the currency exchange, etc., – it might not be impossible.
- Restates his belief that open dialogue of the kind seen at the meeting is desperately needed in the wider world.
Final Word – codyjlascala
[Pantera’s Video: 1:19:53-1:22:00]
Featured in the award-winning film My Avatar and Me, Cody made a moving statement at the conclusion of the meeting, spoken through his carer, but offered in the first person here.
Here is my opinion on AI. I think it would be a good thing, especially for people with disabilities, because it could help them to build things and stuff like that, which is very difficult for them to do. I have been trying to build a city in Second Life, and it has really not been easy, and AI could really help me do what I want to do. I love Second Life because it is a democracy and not an autocracy.
Inara Pey: Living in a Modemworld 