New World Notes

Swimbot NFTs Evolve & Compete With Each Other Off-Chain -- New Project from Virtual World/Artificial Life Pioneer Jeffrey Ventrella

Update, 2/14: In Comments, Ventrella credits his collaborators on this project: "The NFT pool itself was mostly the brainchild of Canton Becker (@canton), and he oversaw the project. Luka Negoita (@lukanegoita) invented the "selfish gene" algorithm that calculates the influence of NFT genes on the evolving population. I want to give a shout out to these two amazing, creative people."

Swimbots NFT is a new and unique application of non-fungible tokens, because unlike nearly all other NFTs, they're not simply about "registering" a JPEG doodle. Here, instead, the NFT is associated with a Swimbot artificial organism which then evolves and competes with other Swimbots off-chain. (Watch above.) And while I have a very pronounced skepticism to NFTs in general, this project is worth looking into for at least three reasons:

• It's co-created by Jeffrey Ventrella, an indie developer who did some pioneering work for Second Life and other virtual worlds, and is a direct spinoff of his long-running artificial life Swimbots project.  
• It's run on the Tezos blockchain, which has way less energy-consumption problems than ETH/Bitcoin blockchains.
• Good god I'd really like to see people try experiments with NFTs that aren't just static JPEGs that we're supposed to get excited about and burn our money on, speculating.

"We are trying to add some value to the NFT phenomenon," as Ventrella puts it to me, "which is overrun with crappy amateur art, and the artless speculators who throw money at this garbage. We are planning on developing these technologies further so NFT's have extended value in a gamified space."

His Swimbots even operate on a famous evolutionary principle:

Paintings by A.I. Artist Instantly Earns Nearly $250,000 in NFT Sales

Pictured: Art by "Alejandro" on Open Sea

Update, 9/2: Clarification via August Rosedale: The $250K worth in sales mentioned in this post were not directly, as originally reported, from art by AIs Alejandro/Taylor -- they were "created" by Rosedale's new AI artist crypo-native, "a truly digital artist creating dynamically evolving art pieces in numerous styles." However, he adds, "Alejandro and Taylor did entirely sell out a couple days before the crypto-native release I think due to excitement for the crypto-native project."

So I told you about "Alejandro", an AI personality who is "creating" works of art which entrepreneur August Rosedale is then selling as NFTs, but here's the wilder thing: 

When August put art by Alejandro and his AI artist colleague "Taylor" up for sale this Monday, via his Mirage Gallery, nearly 1000 copies were sold near instantaneously. [Note: See clarification above - WJA]

"The full 920 piece sold in 5 minutes, which was absolutely insane," August reports to me now.

At their current value in Ethereum, that translates into $85,000 USD -- but that's only the start, because many of those works are then being resold: "Meaning someone buys one (or got one of the 920 from me) and then has sold it. Combined secondary sales have almost reached 40 eth ($148k) which I get a 10% cut of."

So within a week, total sales for art created by AI are now nearly $250,000.

Even August is stunned:

Inside VRChat's Happy Hill Dog Park, the Virtual Dog Experience Created by Academics to Study Real Life Well-Being & Social Isolation

Since launching last week, Happy Hill Dog Park in VRChat has been swarming with avatars of all kinds, playing fetch and applying copious force-feedback enabled head pats with the AI-powered pooches. Dr. Kim, the director on the project, told me Monday that it's already attracted over 43,000 visits, and been favorited 7000 times.

And while the bucolic experience might seem like a VR version of Nintendogs, it's at heart a sophisticated project spearheaded by an academic with the modest goal of bringing more happiness to humankind.

Dr. Brenda Freshman, a professor of Health Care Administration at CalState Long Beach, conceived and self-funded the project as a proof of concept, working with Kim (a veteran VR/game developer) and Studio CyFi to deploy it in VRChat.

"What I'm doing now is gathering my academic partners," she tells me during my visit to the dog park. "and we'll be putting a variety of studies together, but I think the first ones will be dealing with well-being and social isolation."

One VRChat user told Kim that he had recently lost his dog in real life, and was now visiting Happy Hill Dog Park in VRChat as a way to heal from that grief. But as she puts it to me in VRChat (watch below) Brenda tells me they're not specifically creating this simulation for people who lost dogs in real life, or are unable to own dogs due to allergies or housing restrictions:

Pet the Dog in VR: Adorable, AI-Powered Pooches Arrive in VRChat, a Free Public Demo of a Upcoming Research Project

The dogs in Happy Hill Dog Park are waiting for you to come play! Available now on #VRChat for both PC and Quest.

Environment and animation design by @jendaviswilson , AI & logic by @CyanLaser . pic.twitter.com/mpJ87FXXx7

— Studio CyFi (@StudioCyfi) May 19, 2021

Click here to visit Happy Hill Dog Park in VRChat, the latest, impressive user-made project on the platform -- not just virtual dogs, but dogs you can interact with in full VR.

That's right: Pet the dog in VR. And feel yourself petting them too.

"Basically the controller vibrates if you pet the dog, and they'll follow you around if you're holding like a stick or a tennis ball," VRCer Ben Rainwolf tells me.

The immersive effect of the dogs' AI, matched to the VR experience seems pretty compelling: "The cute dogs never stop wanting attention and I feel terrible if I don't give it to them," as another player put it. "It's like an adorable trap that you can't leave."

"The dogs gain trust with you over interactions and change their animations and wagging the more you interact," project designer Jen Davis-Wilson of Studio CyFyi tells me. (Her co-founder Cyan Laser created the dog AI.)  "The conditionals of going home, interacting, petting, etc. does get a bit complex with multiplayer networking. But this was a fairly small job."

Studio CyFi, as regular readers may remember, was founded to turn a popular VRChat world into a standalone Oculus Quest game. This dog park, however, was not mainly created for entertainment:

Watch Me Chat, Dance, and Flirt With Mr. Bones, an Animesh Skeleton Using Really Useful’s Chatbot Script for Highly Customizable NPCs in Second Life

Verbose, sassy, and surprisingly spry, Mr. Bones is an animesh skeleton on the Marketplace here from The Really Useful Scripts Corner (a proud sponsoring partner to this blog) powered by the Corner’s unique chatbot system.  (Click here to chat with Mr. Bones in SL -- touch the pumpkin to rezz him.)

As you can see from the video above and my dialog below, Mr. Bones himself makes for a friendly virtual companion in SL (he can even follow you on command!), but he’s also meant to show off Really Useful Script Corner’s feature set for customizable NPCs (Non-Player Characters). 

“My basic chatbot has two functions - Chat and Responses,” lead developer Grace7 Ling explains, “Chat includes the bot being able to greet visitors and greet its owner. The bot can also remember who it has met before as well.” Her chatbot can also spout random things as a way of encouraging nearby SLers to reply.

As for responses, Grace means more than simple text-based replies. Responses include “doing something in response to a keyword or key phrase it hears, like giving a landmark or SLurl, or playing a sound or animation.” 

Ms. Ling’s more advanced chatbot comes with a third feature: Chatter, which enables owners to specify multiple possible answers to a statement, with different probabilities for each. You can even expand your bot’s phrase recognition range by adding “brain-files” (i.e. notecards) to its database.

While Mr. Bones comes with the new Animesh Pro Edition chatbot script, Grace’s chatbot series includes a mesh human, a French bulldog, a parrot, and a mirror. 

Her customers, she tells me, are already using them for a wide variety of NPCs in their own projects:

Virtual Pet Rabbit Designed With Behavior-Based AI Born in Sinespace; New Addition to Growing Bestiary of Artificially Adorable Creatures

Just in time for Easter, you can buy a pet rabbit in Sinespace (a media partner to New World Notes), and unlike most virtual world pets, Buddy Rabbit is special, built on a complicated AI architecture. As seen above, Buddy can inquisitively explore an area, flee from trouble -- even doing flying spin-leaps to get away from pursuers -- and follow its owner.

Well, somewhat follow: “You see the rabbit is a wild animal and is difficult to control him exactly,” Buddy’s creator Kokostar warns me, grinning. (He's Kokostar #1902 on Discord.)

A leader in the Swiss Society of Virtual and Augmented Reality, Kokostar has been creating virtual pets in Sinespace for over a year: “I have always been interested in agents which simulate human or animal behaviors.” 

His first attempts at pets were rather mechanical, however, so he turned to a sophisticated back-end structure, to give them more life:

Covid 19 Simulation by Artificial Life Pioneer Jeffrey Ventrella Vividly Illustrates the Importance of Social Distancing

"Covid 19" is a mesmerizing if horrifying simulation that powerfully illustrates the importance of social distancing as a strategy to defeat the spread of Coronavirus. Created by artificial life/avatar technology pioneer Jeffrey Ventrella, a programmer who did some seminal work for Second Life and other virtual worlds, it animates what charts cannot quite convey: Not just that social distancing "flattens the curve", but that failure to practice social distancing gets people yanked into the hospital -- or worse, sent unceremoniously to the morgue.

"In early March, I heard about [the simulation in] this article in the Washington Post," Jeffrey tells me, explaining the origin of the program. "Friends kept telling me it looked like something I would do, so I decided to try my own variation -- with the addition of a hospital symbol that shows how social distancing eases the load on hospitals. Since I do particle systems in my sleep, it was a perfect experiment for me."

He created "Covid 19" in JavaScript and Html/Canvas, using similar physics techniques to Clusters, another Ventrella sim. Here's its basic operation:

Monday Flashback: Slack Began as Work Platform for Making a Virtual World

When listening to my interview with Philip Rosedale, where he talks about creating a virtual world-meets-Slack for distributed work groups, here's a fun irony worth keeping in mind: Slack itself was originally created... in order to create a virtual world. Specifically, Stewart Butterfield's web-based world Glitch. As he told me back in 2011:

“[Glitch] is a game, and not a total open virtual world. There are some tools for creative expression now and more will be added, but anything which would count as ‘user-generated content’ will have to fit within the constraints of the game (without the constraints, there's no game left). So, much of that will be visual customizations (to avatars, houses, locations, etc.) and some of it will be ‘other’ stuff (the non-visual: collaborative music making, player-created games, etc.)”

His development team for Glitch was based around the globe, so they needed a unified communication channel where they could work together, and that's the seed of what eventually became Slack, after the small but much-loved world closed down:

Space Invaders Meets Darwin: Mini-Game Where Waves of Aliens Pass Genes to Next Generation -- And Evolve

Invaderz is a cool little microgame (space bar to shoot, A/left arrow and D/right arrow to move) which demonstrates some big potential for artificial life in games: It's basically like the classic game Space Invaders, but any time you kill an invading alien, a genetic algorithm passes its genes on to the next generation of invaders. Sometimes those genes mutate, and sometimes those genes make the invaders more dangerous. In other words: The aliens artificially evolve to become better predators.

"Usually in a more traditional genetic algorithm," developer Victor Ribeiro tells me, "A mutation has only a small chance to happen, like 1%." To speed this evolutionary simulation up, he coded the mutations to happen 10% of the time. "Playing the game you might see some enemies that are very similar to another, with just one small difference: the mutation. That small difference will affect how the enemy moves; maybe it will affect it in a positive way, maybe not. The thing is, that mutation is unique."

So over time, they could theoretically evolve to the point where the aliens are better than the human player. (Darwinian theory in action, only with more pew-pew-pew sounds!)

In tests, Victor has seen the aliens evolve to become faster, so "by the time [the bullet] arrives at the enemy's location, the enemy will be long gone."

The code is available on his GitHub if you want to play with it. Seems like this could be a great system to implement into full-fledged simulation games. The thing to keep in mind, as Victor cautions, is that evolution is a long process and most mutations are not a benefit or disadvantage, especially in a game context: 

Watch an AI-Controlled Avatar "Learn" How to Dribble In Real Time - Huge Implications for Games, AR/VR, and Real Life Robotics

What you're looking at above appears to be a pretty big breakthrough in artificial intelligence -- an avatar "learning" how to dribble in real time through AI. DeepDribble is a project from researchers at DeepMotion and Carnegie Mellon, and the idea is to demonstrate how an AI-controlled avatar can acquire the complex and subtle set of integrated physical skills and coordination that humans take for granted:

Basketball at its peak is known for intricate ball handling, quick pivoting, and fake outs. A good player will fluidly transition between have a variety of dribbling skills. In order to recreate this, the trained characters need to be able to transition between separately trained behaviors... Libin Liu and Jessica Hodgins use the same methods for training single skills to create a multi-skill control graph (which allows transition between motion fragments of two different skills). The training is done incrementally, behavior by behavior, to ensure integration occurs with proper transitions for each skill. The result is a player that can perform multiple ball handling skills in a variety of orders, changing course with seamless blending.

"First we build a bio-mechanical model of the digital character, using physics to simulate bones and virtual muscles," Deep Motion's Libin Liu tells me. "We put the digital character in the physically simulated virtual world and let it mimic how real players perform the motions using their muscle forces."

The training process, as you might have imagine, requires quite a lot of power -- but Liu says the learning arc took about as long as a single televised NBA game -- or even half of one: