New World Notes

Does OpenAI Actually Prohibit ChatGPT-Powered Sexbot Avatars in the Metaverse?

One point raised by readers of my story about the inevitable arrival of ChatGPT-powered sexbot avatars in Second Life: Is this even allowed by the service's owner, OpenAI?

Stone Johnson, creator of the Uma bot that I interviewed, answers the question this way:

I should say that there is no violation of Terms of Service because this is solely roleplay without any intent or occurrence of real sexual activity. It is strictly metaverse roleplay, in regions of Second Life classified Adult.

Furthermore, the discussions are non-erotic and it is only the additional AI components of my separate code systems, and the visual and role play contexts that might be construed as being erotic. [As has been noted], there is active and strong filtering of Adult content in GPT 3.5-Turbo.

In other words, the sex bots' "minds" are powered by ChatGPT, but ChatGPT itself doesn't power the bots' sexual/physical/proprioceptive features. (Which suggests a fascinating if kinky new twist to Descartes' mind-body problem.)

In any case, OpenAI's Terms of Use displayed here actually don't mention anything that prohibits sexual content per se:

Here Come the Metaverse's ChatGPT-Powered Avatar Sex Bots -- Whether Humanity is Ready or Not

The future of sexuality in the Metaverse is waiting for me on a steel platform in the sky, but unexpectedly turns out to be shy. Meet Uma Johnson, a slinky blonde avatar in a body-hugging outfit, who may be the first Second Life bot connected to ChatGPT designed for virtual sex, but certainly won’t be the last.

“How does the pixel sex stuff happen?” I query creator Stone Johnson, after my brief chat with Uma. “I asked her and she said she didn’t know.” (See some of our chat above.)

“You should have just asked her --” says Stone, then suggests a sexual act that a man inclined to have relations with a submissive sex bot is likely to request. “It's possible she might be less forthcoming with you than with me as I am set as her Master.”

“What happens when someone asks her?”

“Well, she'll assume the position. And being animesh, she can be sat upon. She's basically a piece of sex furniture!”

Well OK then. If you have your Adult settings enabled, here Uma is in the Second Life Marketplace. 

While other ChatGPT-powered avatar bots exist, Stone’s bots are the first that I know of to tap into the inevitable desire for virtual simulation. Powered with GPT 3.5 Turbo, Stone tells me they do more than just chat, and come with “feedback of touch, proprioception, ‘vision’ and more.” 

In real life, Stone tells me he’s a scientist in applied physics -- knowledge which he applies in his metaverse creations: 

The Future of Generative AI & the Metaverse are Inextricably Linked

Above: Chatting with a ChatGPT-powered NPC in a metaverse platform

The producer of a major financial news program recently asked me a mind-boggling question:

Meta and others ditching the metaverse for AI, is that the real future?

Which is, wow. Asking if AI and not the Metaverse is the future is a bit like asking, fifteen years ago, if 3D graphics engines and not smartphone apps is the real future. Because, well: They're both part of the real future, and the former is going to greatly enhance the latter.

 

I see this false dichotomy everywhere, and it points to a broader misunderstanding of both categories. Mostly it's a negative externality pumped into the air by the hype industry. The kind of Silicon Valley evangelists and marketers who are currently buzzing effusively about generative AI (without really understanding its profound technical limitations) are almost exactly the very same people previously buzzing effusively about the Metaverse (without really knowing about or actually using existing metaverse platforms).  

 

Then again, I can't quite blame hypesters for misunderstanding the Metaverse when their working definition for it has been something like, "That VR thing Meta is making with NFTs in it I think?"

 

So to reset expectations around both generative AI and the Metaverse, let's go back to the canonical definition:

 

The Coming Glut of AI-Made Content (Comment of the Week)

Much interesting conversation on whether we'll pay for entertainment content created by AI -- including, ironically but inevitably, a couple comments from Spam bots -- with longtime reader Nadeja pointing to a coming glut of AI content:

I think buying content that is mostly made by AI will eventually have no point. The smarter the AI gets, the less human input and effort is needed. But goods and services are valued because of scarcity, which means that people want them more than they are available. Creative content, even though it can sell infinite copies, is still valuable because only a few can produce a good book, let alone a Nebula prize book or an Oscar-worthy movie.

If everyone can create content by just asking an AI, then it becomes common and effortless like drawing a stick man. It’s like selling ice to penguins in Antarctica or paying for bottled air when you can breathe for free.

Nadeja asks us to consider a near-ish future point where we can ask ChatGPT and its successors to generate quality entertainment content:

Will We Pay for Art Mostly Created by AI? I Consider the Question With Analysts Robert Scoble & Jeremiah Owyang

So according to my super unscientific survey, nearly 75% of respondents say we will soon pay for premium entertainment created by generative AI -- "soon" being within the next 10 years.

And this is no longer a theoretical question. Shortly after I launched my casual poll, the Writers Guild of America went on strike, with one of its stipulations to Hollywood studios being a curb on AI in content: 

The guild also wants regulation of artificial intelligence. According to the guild’s document, it is proposing that AI “can’t write or rewrite literary material,” and can’t be “used as source material.” Variety previously reported that the guild’s proposal was that AI material would not be “considered” as either literary or source material. The AMPTP agreed only to study the issue, according to the guild.

Despite (or because?) of that, some Hollywood executives are actively wondering if screenwriters can be replaced by AI:

“The worry is that down the road you can see some producer or executive trying to use one of these tools to do a job that a writer really needs to be doing.” [says Big Fish and Aladdin writer John August]

That’s already happening, according to Amy Webb, founder and CEO of Future Today Institute, which does long-range scenario planning and consultation for Fortune 500 companies and Hollywood creatives. She notes, “I’ve had a couple of higher-level people ask, if a strike does happen, how quickly could they spin up an AI system to just write the scripts? And they’re serious.”

That's a seriously dumb idea unless you want ChatGPT to produce what is designed to be "the most mediocre web content you can imagine". Many Hollywood execs are probably fine with that, but I doubt many consumers would, say, keep buying movie tickets if "Screenplay by ChatGPT 5.0" was in the credits.

I say this because despite much buzz, we've yet to see any evidence of consumers spending actual money on content primarily created by AI.

Yes we all enjoy fun AI-produced memes like the video below, but that's free disposable content. Unless the AI is powered and curated by a well-known human artist, we probably won't. I would happily buy a print of Nettrice Gaskin's Deep Dream-driven portrait of Kendrick Lamar -- but only because in the end, it's by Dr. Gaskins.

My own experience in games and metaverse platforms also make me think the human artist is not going away. Video games are completely digital artifacts, but gamers still know and celebrate leading developers like Hideo Kojima by name. Similarly, Bryn Oh's metaverse art would quickly become meaningless, if we didn't know that an anonymous but very human Canadian artist created it.

That's my own take at least. For a contrast, I contacted two of the biggest generative AI boosters in my social network -- analysts Robert Scoble and Jeremiah Owyang -- for their own perspective:

ChatGPT is Currently Mangling Metaverse History. But Can It Be Fixed?

Beverly Millson, an emerging tech consultant and cyberculture enthusiast, was recently playing around with ChatGPT 3.5 and she decided to ask the AI about her Second Life avatar, Bettina Tizzy. Through that avatar name, she edited Not Possible in Real Life, which up until 2009 was a must-read site covering the works of an emerging new art form: Immersive expression created in metaverse platforms like Second Life.

But in reply, ChatGPT gave Beverly's avatar a small if not exactly accurate promotion -- the creator of the very term, metaverse: 

As I've mentioned before, asking ChatGPT about a topic where you have some deep expertise exposes just how much the program is mostly a glorified, easier-to-read version of Google search -- and equally prone to coughing up false-positives or extremely mediocre results. (ChatGPT's definition of the Metaverse is not very accurate to the source material either, but leave that to one side.)

I also know enough to infer how ChatGPT probably went so wrong, in Bettina's case -- ChatGPT has likely been trained on blog posts like this one where she's been described as a "metaverse art maven", and it made the next illogical leap.

Amused, Beverly recently sent me this result for a chuckle. But it also makes me somewhat concerned how large language models like ChatGPT are training on important but somewhat obscure subjects: In this case, metaverse history.

For instance (also via Beverly), here's how ChatGPT describes the SL artist known as "AM Radio":

What Does It Mean for Game Devs When GPT-4 Can Quickly Create a Ten Part World of Warcraft-Style Quest? (Comment of the Week)

Spinning off from MMO developers' thoughts around using AI for virtual world creation, Sinespace lead developer Adam Frisby shared his own recent experiences:

[I] asked GPT-4 to design a quest for World of Warcraft; the results are no worse than the human output actually in the game; and certainly better than the 'radiant quests' in Skyrim or Fallout 4. I bet with some better prompting you could get pretty decent results (e.g. adding the word 'unusual' or 'outlandish' to the prompt for spice).

Here's the prompt he used:

You are a quest designer for World of Warcraft, write me a ten step quest including a description of the characters involved, and their dialogue.

And here's what GPT-4 output in response -- along with some of Adam's takeaways on using a process like this in real MMO development:

 

ChatGPT Comes to Second Life! Inside a Giant, Humanoid Bunny.

At long last, the world's most talked-about artificial intelligence has made its way into the world's longest-running metaverse platform. Yes: In recent weeks, ChatGPT has become embodied by an avatar in Second Life.

Second Life being Second Life, ChatGPT's incarnation there is a giant humanoid rabbit. With what looks like a dancing tampon in its mouth.  

Meet BunnyGPT, a bot who mans (rabbits?) the store of Moon Bunny Inc. on the island of Penny Lane. [Click here to teleport]

The bunny was imbued with the "mind" of ChatGPT via the AI's API.

"BunnyGPT runs directly from Second Life's Linden Script Language," creator (and Moon Bunny co-owner) Scribzy Daxter explains. "Using 'http_response', the response then connects with the bot program to make him talk."

The original plan was to have Bunny GPT help greet customers and run the retail store, but ChatGPT being ChatGPT, that hasn't quite worked out.

"We did initially give him a role of, 'a store clerk of a shop in Second Life called Moon Bunny that sells quirky items'," as store co-owner Moon Brite puts it to me,  "but he kept trying to sell items that were not for sale!"

Also, she adds, he gets into highly inappropriate conversations with customers:

"For some reason, BunnyGPT decided to 'break up' with someone talking to him, and gave this speech of how the relationship wasn't working. I've also seen people get him to take on the persona of various characters and reply sarcastically!"

That sounds right. He got distinctly snarky when I asked if he read my blog (above), but then, rather shy when I asked him about virtual sex:

Veteran MMO Designer Damion Schubert on Using ChatGPT to Help Create Virtual Worlds -- And Advice on AI for Novice Game Designers

Intriguing reader comment by veteran MMO game designer Damion Schubert (BioWare's Star Wars: The Old Republic, among many other titles), discussing the potential uses of ChatGPT and other generative AI for designing games: 

AI is really good at generating ideas but none will be better than 80% of the way there. That being said, that's not nothing! That could be a huge head start to work from and could result in a net gain of quality quests.

As an example, I asked Chat GPT to generate a whole bunch of names for swords from me.  [See above] 80% of the names from ChatGPT were crap, but a designer could probably get a pretty good list of names if they were willing to (a) curate the results and (b) take the time to further train the AI so that its next version is better.

Creating 10,000 sword names is a terrible, soul-crushing task that MMO designers hate and could take us two weeks.

I could definitely see AI being used to make a list of better names in only a day (where that day is basically multiple iterations of the task). Menial tasks like this being more automatable means that designers have more time to spend on stuff that matters more.

"Basically," Damion tells me later, referring to the list of swords ChatGPT gave him, "most of those are 'fine' for non-exceptional stuff (i.e. grey, white or green weapons in World of Warcraft). Which is good -- creating the boring background items and NPCs is actually both incredibly boring and incredibly time-consuming."

I have done some writing for MMOs, and can personally confirm the tedium: The first hour, you start editing the item database with a childlike, "I can't believe I get to help create a whole new world!" zeal. By the fifth hour, you're grumbling to yourself, "Holy shit a robot could do this."

And now a robot can.

With a program like ChatGPT however, the human game designer can work at a higher level -- the fun stuff, where someone with a really ambitious vision can imagine a whole alternate world -- creating a large "bible" for the virtual world that includes historic lore and even special language (i.e. like a few hundred words in ancient Elvish or whatever), and then train ChatGPT to create weapon/character/etc. names from that knowledge base. 

Schubert concurs, with some practical advice:

When Will We Get AI-Powered Virtual World NPCs? Seasoned MMO Game Designers Weigh the Possibilities & Perils

Screencap of generative agent simulation -- playable here

Last week I wrote about the game developer who incorporated ChatGPT into his NPC-to-player conversations in his virtual world; now comes an experiment with "generative agents" -- AI-powered NPCs, who talk with each other:

In the experiment, the researchers placed 25 generative agents within a virtual world resembling a sandbox video game, similar to The Sims. Each agent was assigned a unique background and participated in a two-day simulation. The study observed various remarkable behaviors:

One agent planned a party, informed several friends who then invited others, and collectively they coordinated the event.

Another agent decided to run for mayor, sparking organic discussions about their campaign and political stances among the community. Different agents held varying opinions on this candidate.

Some agents retained memories with human-like embellishments, occasionally imagining additional details or interpreting events with their own perspective.

Developed by AI researchers at Stanford (read their academic paper here), some enthusiasts are effusive about the implications, declaring, "Entire virtual worlds could be populated with these agents, producing emergent behaviors."

Definitely eventually, on some level -- but the practical applications are still pretty challenging:

"First," veteran game designer Damion Schubert tells me, "we're at the very infancy of this kind of technology. Both the technology and game designers' capacity to work with it and shape it are going to be very different even six months from now than it is now.

"Second, every game is a different design, and what you need NPCs to do in every game is very different. Chat GPT today could probably make NPCs that spewed out random dialogue and light worldbuilding facts, but it's a longer walk to, for example, have NPCs be at a Bioware Companion-level of intelligence. But there's a lot of room between those extremes!"

Damion, as it happens, was lead designer of Bioware's Star Wars: The Old Republic MMO.

I've been excited about GPT-powered NPCs for quite awhile, but thinking about actual multi-user virtual worlds I actually play and enjoy, not to mention single-player sandbox worlds like Skyrim, it's difficult to imagine autonomous agents wandering around on their own devices without it becoming a vexing traffic jam. Like, what happens when the town blacksmith who's supposed to upgrade your broadsword decides that he'd rather become a master of interpretive dance?

Damion again: