New World Notes

Look: More Fun Turning Avatars Into "Real" People With Artbreeder AI

In case you missed it, here is Ugleh Ulrik's tutorial for using the AI-driven platform Artbreeder to transform avatars. If you have questions about his tutorial, clickthrough here to ask him on YouTube.

On Second Life Friends, a large Facebook group for SLers have been using this tutorial t0 turn their avatars into "real" people. With permission, here's some of my favorite before/after pics:

Tutorial: How to Turn Second Life Avatars Into "Real" People With Artbreeder

Want to turn a Second Life avatar pic into the "photo" of their real life analog? Watch this Artbreeder tutorial from Ugleh Ulrik, the SL creator who first started the whole buzz around using the AI-driven platform to transform avatars.

"I do a lot of experimenting with new AI tech," Ulrik tells me, "and once I saw Artbreeder my first thought went to Pixar characters but then after that I realized making Second Life avatars more real was a possibility and I tried it out on a few close friends and it worked well."

Here's his key tips to keep in mind when using Artbreeder on an avatar:

I'm Mesmerized by These Avatars Turned Into "Real" People Via ArtBreeder AI

My post last week on Artbreeder, which uses Generative Adversarial Networks to create "children" of an original image, generated a mini-burst of excitement among many in the SL community, who started experimenting with the process to turn pictures of their avatars into "real" people. And the results are pretty mesmerizing! Above is one by Devin Vaughn, whose avatar looks in real life, I guess, like the clone of Ronan Farrow blended with Anderson Cooper?

Some more below:

Will Real Celebrities Expand Their Brand With AI-Based Clones Like This?

What you're watching above is a video from social media influencer Taryn Southern, except not quite: It's a "clone" of Ms. Southern, created via an AI platform called HourOne.

"We take a small sample of video footage (ideally captured in a studio, as we did with Taryn), and use our proprietary AI algorithms and generative adversarial networks, to generate an exact digital replica of that person," explains Natalie Monbiot, Head of Business at HourOne. "We can then program the digital twin by way of text or voice inputs, to generate new content delivered in an entirely lifelike way. Uniquely, we don’t do any stitching with our characters – they are fully generated, in the cloud, to deliver with flexible and natural facial expressions -- and since the videos don’t need to be ‘handled’ the potential to scale is enormous."

Monbiot estimates it's possible to save 95% of production costs on future videos, using clones this -- since after all, once Taryn's clone is created, a real life studio, camera crew, make-up, lighting, etc. are no longer required. To be sure, there's a certain dead-eye loss of humanity in the transition from biological Taryn to digital Taryn, as you may notice from the video. But Monbiot says HourOne's AI clones will become more lifelike and expressive in future iterations:

"Today we specialize in face-to-camera video with nuanced and natural facial expression and head motion," as she puts it. "You will certainly see bigger ranges of movement and emotion in the coming months."

From the point of view of well-known online personalities like Taryn Southern, not to mention major pop stars, film celebrities, and beyond, this opens up the possibility of their AI clones continuing to perform, endorse products, make press appearances, and beyond, even while their organic selves are sleeping, or on vacation.

Afrofuturist Nettrice Gaskins on Black Algorithms & Her Neural Net Portrait of Stacey Abrams

Afrofuturist artist Nettrice Gaskins, last seen here hosting an art gallery in Second Life since her LA show was cancelled by the pandemic, has a thought-provoking and actually rather thrilling post linking Black Algorithms of the past to the present:

Kuba cloth’s complexity derives from semi-symmetric compositions of simple geometric motifs. The semi-symmetry in the motifs are similar to Ghanaian Kente and other African textile designs. You can also find these motifs in African American quilts. The knowledge and practice crossed the Middle Passage where it became codified and passed down through generations. These designs are also computational; they use a generative (shape) grammar that, in effect, formalizes an algorithm that generates strings in a language...

This algorithm is also used to create music, as evidenced by James Brown (and others) who pioneered funk and hip-hop/rap. James Sneed wrote about it in a 1981 essay titled “On Repetition in Black Culture.” Snead says that human culture evolves and things from the past emerge in the future as either improved or retrogress. What is interesting to me is how, in computer science, the generative grammar used to create the textiles and musical sequences are linked to artificial intelligence, specifically neural nets and machine learning. I use the latter to create art such as this piece.

She means the one here, an amazing portrait of Stacey Abrams she created using Gatys image style transfer, which is (as explained here), "[N]eural representations to separate and recombine content and style of arbitrary images, providing a neural algorithm for the creation of artistic images."

She relates the Abrams portrait to improvisation: "Improvisation depends on the ability to extemporize new melodies (training data) that fit the chord sequence (model)," she tells me, citing her essay here. "Call-and-response participation is happening, too, as the medium (machine) responds to the data input." 

Dr. Gaskins has a book touching on this coming soon from MIT Press, Techno-Vernacular Creativity and Innovation.  Meantime, here's my profile on her portraits of Black artists using Deep Dream AI algorithms from 2019, including Sade, Nnedi Okorafor, and (now tragically poignant) Chadwick Boseman (also below):

For Auction: Art Created by Artificial Intelligence from Artists Also Created by Artificial Intelligence

This striking abstract painting is by an up-and-coming artist named Taylor; you can bid on it for $200 or more soon, in hopes it becomes a rare collectors’ item that symbolizes the start of a new art movement.

Because the thing is, “Taylor” is not an artist and this is not a “painting” as we typically understand those terms. The painting was generated by artificial intelligence… as was the painter herself. 

You can see more works by “Taylor” here, part of a fascinating project by rising technologist August Rosedale, whose development I’ve been casually following since the early 2000s.

Starting soon, Rosedale’s Mirage Gallery will auction off AI paintings like this to the highest bidder, and confirm their authenticity by registering them on a blockchain.

“I think that people will bid on AI-made paintings for the same reason people bid on human-made paintings,” August tells me. “People enjoy having beautiful art pieces around them and also can be invested in the idea of the artist's work increasing in value over time.”

The painting, to be sure, is not actually painted, and will come with a blockchain token.

“For the physical pieces, it will be a 24x24 inch canvas print. Each buyer can choose to sell the token and canvas print separately but the value will come from keeping them together.”

To create Taylor’s work, August compiled ten thousand or so abstract art paintings in a wide variety of styles, then fed them into a generative adversarial network, or GAN:

“The GAN works by having something called a discriminator whose job is to determine if an input image was from the original dataset or not. What this means is that the model that is training will create an image, and then feed it into the discriminator and see if it can convince the discriminator that the inputted image was from the original dataset. Over time, the model is able to create more convincing outputs that start to trick the discriminator.”

The only human input in their creation happened at the beginning, when August curated the datasets the AI then worked from:

“It is important for the datasets to have similarities in the images or else the outputs will be much more difficult to interpret,” as he explains. “Once the dataset has been made, the GAN plays a sort of game with itself to try to trick itself into believing that its own creations were from the original dataset. With powerful GPUs and lots of computing time, this process can result in amazing pieces of art. The Mirage Gallery artists will continue to refine their art with more hours of training as time goes on.”

Same goes for the personae and appearance of Taylor and “Alejandro”, the Mirage Gallery’s current representational artist:

New iPad App Uses Machine Learning to Capture Real World Objects & Convert Them Into Standard 3D Files

Caveat emptor and all that, but the iPad LIDAR Scanner just published in Beta seems seriously impressive, able to "capture" real world objects and convert them into imperfect but pretty usable 3D models. It works, says developer Laan Labs, by automatically extracting objects via machine learning (i.e. extrapolating data for unseen surfaces).

More info from iOS app page: 

I Was Going to Blog an Uncanny Valley Think Piece About Samsung's NEON "Intelligent" Avatars But Then Laughed My Ass Off Instead

Samsung's NEON! It's getting much buzz at this year's CES, so much so that I planned to write a thought provoking think piece about them with a mandatory Uncanny Valley reference, but then I watched this video and got to 4:20 (no I'm not high) and just laughed. Does any human in the history of the world say "pizza" like that, unless they have a gun pointed to their head?

Anyway, more on the official homepage:

Listen: Music Inspired by Gödel, Escher, Bach Co-Created by an Artificial Intelligence Trained on Bach and Human Interaction With the AI

Click above to hear something fairly amazing: Music created by a merger of Bach, artificial intelligence, and live interaction by a human audience. It's from the new album "The Eternal Golden Braid" by my pal adaptive music/sound artist Robert Thomas, who's been creating works of music shaped by virtual reality, digital interaction, and AI for well over a decade. (As here with Chris Nolan and Hans Zimmer and more recently here with the LA Phil).

This current project, as the title suggests, was inspired by the landmark book Gödel, Escher, Bach: an Eternal Golden Braid, and incorporates music from Bach himself. As Robert explains in his liner notes:

This piece started life as a collection of renderings from a machine learning process which was trained on a large database of works by Bach. I then curated some fragments from it and used them as a starting point for a composition for string trio.... We wanted to explore having the audience interact live with the music in a collective strange loop, so I constructed the piece in two "levels" which the players could move between at any time. Level 1 was calmer and more introspective and represented by Blue. Level 2 was more chaotic and extroverted and represented by Red. When the piece was performed the audience held up colored cards, which were recognized by a machine learning camera on the stage. This fed a signal to the players which told them which musical notation to read. 

So this is, as Robert puts it to me now, "music composed in collaboration with an AI, which was trained on Bach, then performed by the London Contemporary Orchestra, with the audience interacting, guiding the music. Aspects of the music were inspired by themes in Gödel, Escher, Bach, though: At the start and end you hear kinda shepherd tones played by the strings; these are one of the strange loops referred to in Marcus du Sautoy's talk about [Hofstader's book]."

The live performance looked like this:

John Carmack, VR Pioneer Who Once Described Developing VR as a "Moral Imperative", No Longer Focused on Developing VR

Oculus CTO and VR pioneer John Carmack is stepping almost completely away from the company and virtual reality development in general to go in a totally different direction:

Starting this week, I’m moving to a "Consulting CTO” position with Oculus.

I will still have a voice in the development work, but it will only be consuming a modest slice of my time.

As for what I am going to be doing with the rest of my time: When I think back over everything I have done across games, aerospace, and VR, I have always felt that I had at least a vague “line of sight” to the solutions, even if they were unconventional or unproven. I have sometimes wondered how I would fare with a problem where the solution really isn’t in sight. I decided that I should give it a try before I get too old. I’m going to work on artificial general intelligence (AGI).

I'm not sure how many people in the VR industry grasp the significance of this move. The announcement comes only days after Carmack said this about the current state of VR:

"I'm often kind of grumpy around the office because I really haven't been satisfied with the pace of progress that we've been making. When I'm in VR I see the magic there, but my brain is always throwing up these giant 'to do' Post-It Notes on top of everything, reminding me of all the work that's yet to be done."

Putting the two statements together, it's hard to escape the conclusion that Carmack has decided that virtual reality won't or can't be improved, or sufficiently matured into a mass market product, within the span of his career. And so instead, he's devoting the rest of his work energies to developing AI.

The statements also stand in sharp contrast to how Carmack fairly recently described the importance of bringing VR to the masses -- as "a moral imperative".

As he explained to me for my Wired article back in 2016: