The rise of AI stuff has has made me not want to refer to characters like Glados or Rambley as "AIs". Like ew, no, that word has connotations, so I've started calling them "living computer programs" instead
OK since I haven't seen too many people talk about this since twitter news usually strikes pretty fast over here whenever e'usk does anything ever, let me give ya'll the run down on two things that will go live on NOVEMBER 15TH and why people are mass migrating to Blue Sky once more; and provide resources to help protect your art and make the transition to Blue Sky easier if you so choose:
The Block function no longer blocks people as intended. It now basically acts as a glorified Mute button. Even when you block someone, they can still see your posts, but they can't engage in them. If your account is a Public one and not a Private one, people you blocked will see your posts.
They say because people can easily "share and hide harmful or private information about those they've blocked," they changed it this way for "greater transparency." When in reality, this is an extremely dangerous change, as the whole point of blocking is to cease interaction with people entirely for a plethora of reasons, i.e. stalking, harassment, spam, endangerment, or just plainly annoying and not wanting to see said tweets/accounts. or you know, for 18+ accounts who do not want minors interacting with them or their material at all (There is speculation saying these changes are specifically for Elon himself so he can do his own kind of stalking, and honestly, with the private likes change, it lowkey checks out in my opinion)
Also, this straight up goes against and may violate Apple and Google's app store policies and also is straight up illegal in Canada and probably other countries as well.
If this ACTUALLY goes through, twitter will only be available in select countries, probably exclusively in the US, which would collapse the site with the lost of users and stock, and probably be the last push it needs to kill the site. And if not, will be a very sad and exclusive platform made for specific kinds of people who line up with musk's line of thinking.
2. New policies regarding Grok AI and basically removing the option to opt out of Grok's information gathering to improve their software.
And anything you upload/post on the site is considered "fair game" with "royalty-free licenses" and they can do whatever they please with it. Primarily using any and all posts on twitter to train their Grok AI. A few months ago, there was a setting you can opt out of so they couldn't take anything you post to "improve" Grok, but I guess because so many people were opting out, they decided to make it mandatory as part of the policy change (This is mainly speculation from what I hear).
So this is considered the final straw for a LOT of people, especially artists who have been gripping on to twitter for as long as they can, but the AI nonsense is too much for people now, including myself. Lot's of people are moving to Blue Sky for good reason, and from personal experience, it is literally 10x better than twitter ever was, even before elon took over. There is no algorithm on there, and you can save "feeds" to your timeline to have a catered timelines to hop between if your looking for something specific like furry art or game dev stuff. It's taken them a bit to get off the ground and add much needed features, but it's genuinely so much better now
RESOURCES
Project Glaze & Cara
If you're an artist who's still on twitter or trying to ride it out for as long as you can for whatever reason you have, do yourself a favor and Glaze and/or Nightshade your work. Project Glaze is a free program designed to protect your art work from getting scrapped by AI machines. Glazing basically makes it harder to adapt and copy artwork that AI programs try to scan, while Nightshade basically "poisons" works to make AI libraries much more unstable and generate images completely off the mark. (These are layman's terms I'm using here, but follow the link to get more information)
The only problem with these programs is that they can be resource intensive for computers, and not every pc can run glaze. It's basically like rendering a frame/animation, you gotta let your pc sit there to get it glazed/nightshade, and depending on the intensity and power of your pc, this may take minutes to hours depending on how much you wanna protect your work.
HOWEVER, there are two alternatives, WebGlaze and Cara
WebGlaze is an in browser version of the program, so your pc doesn't have to do the heavy lifting. You do need to have an account with Glaze and be invited to use the program (I have not done so personally so I don't know much about the process.)
Cara is an artist focused site that doubles as both a portfolio site and a general social media platform. They've partnered with Glaze and have their own browser glazing called "Cara Glaze," and highly encourage users to post their work Glazed and are extremely anti-ai. You do get limited uses per day to glaze your work, so if you plan on doing a huge backlog uploading of your art, it may take awhile if your using just Cara Glaze.
Some twitter users have suggested glazing your art, cropping it, and overlaying it with a frame telling people to follow them elsewhere like on Bluesky. Here's a template someone provided if you wanna use this one or make your own.
Blue Sky Resources and Tips
So if your a twitter user and your about to realize the hellish task of refollowing a massive chunk of people you follow, have no fear, there's an extension called Sky Follower Bridge (Firefox & Chrome links). This is a very basic extension that makes it really easy to find people on Bluesky
It sorts them out by trying to find matching usernames, usernames in descriptions, or by screen name. It's not 100% perfect, there's a couple people I already follow on Blue Sky but the extension could not find them on twitter correctly, but I still found a huge chunk of people. Also if your worried that this extension is "iffy," they do have a github open with the source publicly available and the Blue Sky Team themselves have promoted the extension in their recent posts while welcoming new users to the platform.
FEEDS and LABELS
OK SO THE COOLEST PART ABOUT BLUESKY IS THE FEEDS SYSTEM. Basically if you've made a twitter list before, it's like that, but way more customizable and caters to specific types of posts/topics. Consolidating them into a timeline/feed that exclusively filled about those particular topics, or just people in general. There's thousands to pick and choose from!
Here's a couple of mine that I have saved and ready (down below). Some feeds I have saved so I can jump to seeing what my friends and mutuals are up to, and see their posts specifically so it doesn't get lost in reposts or other accounts, and also specialized feeds for browsing artists within the furry community.
The Furry Community feeds I have here were created by people who've built an algorithm to place any #furry or #furryart or other special tags like #Furrystreamer or #furrydev. They even have one for commissions, and yes you can say commissions on a post and not have it destroyed or shadow banned. You are safe.
If you want, and I highly recommend it to get visibility and check out a neat community, follow furryli.st to get added to their list and feeds. Once your on the list, even without a hashtag, you'll still pop up in their specialized feeds as just a member of the community there. There are plenty of other feeds out there besides this one, but I feel like a lot of people could use one like this. They even got ones for OC specific too I remember seeing somewhere.
And in terms of labels, they can be either ways to help label yourself with specific things or have user created accessibility settings to help better control your experience on Blue Sky.
And my personal favorite: Ai Imagery Labeler. Removes any AI stuff or hides it to the best of it's abilities, and it does a pretty good job, I have not seen anything AI related since subscribing to it.
Finally, HASHTAGS WORK & No need to censor yourself!
This is NOT like twitter or any other big named social media site AT ALL, so you don't have to work around words to get your stuff out there and be seen. There are literally feeds built around having commissions getting and art seen! Some people worry about bots and that has been a recent issue since a lot of people are migrating to Blue Sky, but it comes with any social media territory.
ALSO COOL PART,
you can search a hashtag on someone's profile and search exclusively on that profile as well! You can even put the hashtag in bio for easy access if you have a specialize tag like here on tumblr. OR EVEN BUILD YOUR OWN ART FEED FOR YOUR STUFF SPECIFICALLY!
So yeah, there's your quick run down about twitter's current burning building, how to protect your art, and what to do when you move to Blue Sky! Have fun!
Hello. This is an essay and an otherwise a deeply autistic combing of dubiously intentional material from a robot cartoon. And some stuff about early computing, AI, chess, and Gary Kasparov.
Let me spin you a tale:
Deep in in the AI Winter of the 90's, when personal computers were becoming more common, and early neural network development was struggling to find commercial use, Gundam Wing made its debut. The controversial but enormously influential Deep Blue vs. Kasparov showdown was still two years away, the World Wide Web was two squeaks old, Google Search was barely a glimmer in Larry Page's eye, and The Matrix wouldn't come out for another four years.
AI-apocalypses weren't new; Skynet and WOPR and HAL were well embedded in the public imagination; cyberpunk anime was thriving in the same ecosystem as the "real robot" genre from whence Gundam was born. But Wing isn't cyberpunk, and it doesn't really lead with AI-apocalypse. The show's got a lot on its mind– most of it about Ethics and Ideals and Responsibility and Upholding Pacifism As A National Policy– but there IS an AI-apocalypse hiding there in the back. It's not flashy as AI-apocalypses go, and because this is Gundam Wing, it dwells not on time traveling cyborgs or murderous space stations, but on ethics and human integrity. Wing's fairly sparse world building didn't go so far as to predict social media, or the relevance of algorithms in online spaces, or the immediate saturation of AI into mainstream civilian tech, or oversight of civic infrastructures being gutted and left with a machine in charge– but it was weirdly prescient in other ways.
The deeper we get into our own increasingly messy, concerning, frustrating, and dangerous AI fervor, the more interesting I find Wing's take to be. For one thing, instead of dealing with the louder and more titillating concept of true machine intelligence (malevolent or otherwise), Wing’s autonomous robot scourge never once approaches sentience– there’s certainly a ghost in SOME of the machines, but the series stays remarkably grounded in depicting machine learning as strictly machine learning. It’s entirely unconcerned with the development of sapient computers, and instead dedicates itself to hammering home the message that the actual problem being faced by humanity is the absurd, incautious, self-destructive way we use automation against ourselves.
The story isn't telling us to be scared of AI or algorithmic machine learning– it's about being critical of systems that are designed to remove accountability from institutions of power and that vastly increase those institutions’ ability to do harm. It's not a warning about the dangers of progress, it's about the danger of removing people’s sense of investment, responsibility, decision making, and creativity from human endeavors.
And since this is a Gundam show, that means war.
MOBILE DOLLS IN CANON:
Mobile Dolls are the brainchild of one Bilmon Tubarov, a high ranking member of the Romefeller Foundation and Chief Engineer of OZ. He was one of the original leading developers of Mobile Suit technology, along with Howard and the five Doctors, J, O, S, H, and G.
(...I was sincerely hoping that would spell something funny.)
The first mention of the program that would become the basis for Mobile Doll System is in Episode 10: Doctor J tells Heero about the (still exclusively manned) Taurus suits' “ability to learn from accumulated data” and expresses worry that they might "override the pilot's commands and act on their own", which is something nobody wants from an 8-ton war machine– right?
What Doctor J describes is a simple machine learning program installed in the Taurus that takes accumulated data from past battles (in whatever way OZ’s technology gathers such things) and turns it into optimized tactical strategies. How it actually presents these strategies to the pilot is never explained, but by J’s comments, we can presume the system is tied directly to the controls of the mobile suit— and J isn’t willing to take chances on a buggy self-driving car with a gun the size of a tree.
(Now... as I've previously asserted, there's NO way the scientists who defected from OZ and designed the ZERO System didn't know about this program already, and it’s actually quite likely they were involved in its creation to begin with. The timeline of events suggests that Tuberov started developing the program at least from the development of the Taurus suits, and possibly even from his time working in mobile suit R&D along with the rest of the scientists.)
Obviously, in the long run the Taurus System isn't going to "accidentally" override the human pilot, it's going to replace the human pilot entirely.
"If anyone builds it, Heero, we all die."
After Episode 10, the topic of machine learning being used in mobile suits goes away for a bit (I'm using the term "machine learning" specifically because that's how J describes it– a machine that "learns from battle data"– and because there's really no discussion of mobile suits having actual artificial intelligence; at least not until the ZERO System appears). Mobile Dolls emerge again fully fledged in Episode 18:
During the period after OZ’s takeover of the Earth Sphere, the Romefeller Foundation is locking down its control of the Earth. As is the way of most violent revolutions, the previously useful regime-toppling forces are suddenly way less desirable than standard issue, order-following ones— and what could toe the line better than a whole programmable machine army? So why not replace human soldiers with AI, and hire prompters to give them orders?
Well, mostly because the aforementioned regime-topplers within the Foundation are violently against the idea, and in a scene I could watch all day, Former Gifted Child Treize Khushrenada objects to this proposal with extreme prejudice and a sword.
Chief Engineer Tuberov is given a good scare in the control tower, but he continues to rise in authority within Romefeller until he ultimately gains control of OZ; meanwhile, Treize is quietly sunsetted.
...I've always found the visual of the empty, pilotless cockpits lighting up to be wonderfully unsettling; it makes them seem undead rather than just automated. The later Virgo MDs understandably lack cockpits altogether, so it's just the early dolls that still have vestigial cockpits.
The proof-of-concept MDs on display are retrofitted Leos, crude enough that they have to be plugged into the command tower; but Tuberov clarifies that this isn't the case with the Taurus Space MDs, which come with the system built in. Duke Dermail seems to think MDs are for supplementing the troops, but Tuberov advocates for the replacement of all of OZ’s mobile suit forces with Mobile Dolls.
—Note: The official script contains an error here, where the line reads "we used an old model of the Leo so we're having to program them from a distance, but no such problem exists with the Space Taurus". The fan translation fixes it to "from here", as in "from the command tower", which makes far more sense, because they're literally plugged in.
The retrofitted Leo MDs are shown being able to 1) Identify targets– which in this case are: Tank, Paper Circle, and Angry-Leo-With-Treize-In, 2) reference schematics of the targets, and 3) run risk assessments (judging by the flashing "CAUTION" readout). The handlers in the tower seem to just be in charge of the On/Off switch.
—
Right up next, we get Episode 19, which is the absolute best for showcasing the flaws in the MD targeting system:
Using a unique form of cabbage-based espionage, Heero breaks into a Colony prison to kill/rescue Duo (who is having his longest ever streak of no-good-very-bad-days), without a working Gundam between them. They're surrounded in short order, the whole Colony security complex is put on high-alert, and OZ sends out the Mobile Dolls to eliminate the intruder and escapee.
One of the MD handlers states that it would be "difficult to select targets inside the Colony", which suggests that the MDs are indeed analyzing and setting the targets themselves in some capacity, but with a degree of error that makes it dangerous to trust them in a complex setting with a lot of changing variables. This does not bode well for anyone involved. (Foreshadowing)
"Wouldn't target settings be difficult within the colony?"
"COME ON, WE'RE NOT BANGING ROCKS TOGETHER HERE, WE KNOW HOW TO MAKE AN AUTONOMOUS DEATH ROBOT".
In the mobile suit launch bay, Heero finds a bunch of MS carriers and empty Leos. He loads them up and launches all of them out the hatch, effectively creating a flotilla of ghost ships for the Mobile Dolls to target. He also shoots a gun at the surveillance cameras in the launch bay, thereby registering "guy in an OZ astro-suit" as a threat as well.
Back in the station, the OZ security crew are baffled as the dolls start firing on their own Leos. They dig into the logs to discover that the Leo suits were set as a target because they'd been seen escaping the Colony, and the dolls' standing orders were "don't let anyone do that".
This is followed by a brief moment of panic as the security staff realizes their OZ astro-suits have also been set as targets. The handlers try to cancel the “auto-command”, but before that can happen they're obliterated by the “cutting-edge annihilation machines” they’d trusted with the authority to decide who lives and who dies.
(I find it interesting that they do NOT give the doll verbal orders to stand down or stop firing– my theory is that a doll that’s running on automatic would not take orders from a target it had identified as an enemy.)
Duo and Heero meanwhile escape scot-free in a Honda Civic passenger shuttle that wasn't set as a target.
Turret voice: "Hello?" :) "Are you still there?" :) "There you are" >:)
The next good showcase of MD limitations is Episode 21:
While being escorted as a prisoner of OZ, Trowa (in the guise of an OZ officer) asks Heero to explain his favored tactic for destroying MDs. Heero's advice is simply: "Avoid their targeting system" and "get in close range". Zechs corroborates this shortly thereafter, and as a doctrine this basically holds true for everyone who doesn't have access to a buster rifle for the rest of the series. It holds doubly true of the later Virgo suits because of their energy shield.
In the same episode, Heero also states that the weakness of the MDs are that they have the same basic level of mobility and "reaction times" as the manned models– this struck me as odd, because everyone up to this point has made a point of saying how fast the MDs are (including Zechs who modestly reports "my reactions are a little slower than theirs"). (I went and double checked the script against two different translations and the line reads basically the same.)
In my own interpretation, the confusion here lies in the distinctions between speed/output, acceleration, and reaction time:
There's no doubt the top speed and rate of acceleration that MDs are capable of withstanding are greater than those of a human pilot for the obvious reason that they don't have one– but the perceived reaction time of a MD might well be the same as, or even slower than a human's. The MDs have to identify and select a target using their rudimentary software, which as we've seen is still reasonably clunky– they might be able to track a target and fire quicker than the average pilot, but if what Heero is observing is true, it's the actual hardware of the mobile suit base that's the limiting factor.
Interestingly, the only MD to have any close-range combat capability is the MD clone of the Mercurius. To me, this suggests that ranged weapons are significantly easier to program a machine to use effectively, while hand-to-hand combat is still beyond the system's abilities. (The only exceptions I can find are of MDs grabbing stationary suits and smashing them, or firing into them point-blank; so LITERAL hand combat in the crudest sense is possible.)
The cloned Mercurius and Vayeate dolls are also the only MDs of different models that attack cooperatively. Virgos and Taurus MDs are never deployed together despite being a potentially devastating combination (think rooks and bishops). Whether this is due to a limitation of programming or lack of innovation on the part of the handlers is unclear.
Getting Mobile Dolls to coordinate attacks based on high-level pilot simulations seems like it’d be a huge game changer, but it isn't– Duo spends maybe twenty seconds fighting them before ripping into them like a Quinze-opener. The MD System is just not advanced enough to be useful for anything more complex than varying forms of aim-and-fire; even when they’re programmed with more advanced tactics, they can only mimic maneuvers they’ve seen before without the ability to adapt them situationally.
Despite having the ability to “learn” from past battles in theory, the dolls are never shown adapting their strategy in real-time to deal with different enemies or even repeat encounters with the same enemy.
In Episode 34, Zechs makes a point of saying that OZ must have known he was going to attack because the Mobile Dolls came in a specific formation that seemed tailored to counter him, while the manned suits retreated; adding that this was a predictable strategy to use because of OZ’s reliance on Mobile Dolls; what he does NOT say is that the dolls themselves are predicting his tactics or countering them. Wufei expresses a similar amount of apathy towards the MDs repetitive tactics.
With tiring consistency, OZ’s main strategy for eliminating tough enemies with Mobile Dolls is basically just “send more Mobile Dolls”.
As any gamer who’s beaten the same enemy hundreds of times could tell you, humans tend to fine-tune their reactions based on experience and iteration– Mobile Dolls do not. The MDs may be able to "learn" from accumulated data in the sense that their database expands, but unlike the ZERO System, the MD System isn't capable of running complex predictions or data synthesis.
"Mobile Dolls show no mercy!" is a common refrain from Tuberov and his fellow MD enthusiasts— which functionally means they can only attack; they don't show "mercy" because they're not capable of making complex decisions about themselves or their targets.
They don't fire warning shots, or shoot to disable, or accept surrenders, or retreat from battle, or change objectives on the fly. They might have tactical programs uploaded to them which allow them to respond quickly to recognized enemy attack/defense patterns, but again unlike the ZERO System, they cannot process strategic objectives on their own.
Pictured: proud members of the OZ Mobile Doll Babysitting Corps, now deceased (except for the cute twink with the undercut, I think he makes it).
The degree to which the MDs in the show are truly autonomous is debatable. The Taurus MDs ALWAYS have a pilot in a manned suit supplying them with directives, and the Virgos ALWAYS have a command center of some kind nearby, usually in a plane, shuttle, or battleship. The pilot or handler in command usually gives the various doll units orders to adjust their positions or update their objectives.
As well as indicating the limits of their autonomy, this suggests that dolls can't operate effectively outside of short-range communication distance, so Romefeller can't just drop a bunch of MDs into a city center and leave them there as a permanent automated deterrence, for example. Dolls seem to just stand around aimlessly after battles, so it's possible that in the absence of new orders they simply go idle when they run out of things to do.
(And to be fair, so do people.)
"So, did you enjoy the holidays, Jan?" "You know Patricia, they just haven’t been the same since the divorce."
So in summary– as written and shown in the series, we can make the following assumptions about the (unaugmented) Mobile Dolls and the MD System:
-The system selects from pre-programmed targets that can be toggled by its controllers, but it can also self-determine targets in response to basic stimuli and threat analysis. It seems to have some capacity to interpret standing orders, though not with much nuance.
-Dolls receive their objectives and standing orders remotely from a nearby handler, usually in a manned suit or a command vehicle. This suggests that their autonomy is limited to targeting and using set tactics from their database.
-They do not appear capable of synthesizing data or adding to their databases directly, suggesting that these updates must be programmed, or compensated for by direct order.
-Close-range combat is the achilles heel of either the MD targeting system, tactical programming, mechanical capabilities, or a combination of the above.
-Unlike the ZERO System, which shows properties of being one or more neural networks, the MD System has the approximate sophistication of a search engine.
(...Now, if you wanted to give the Mobile Doll System a genuinely modern AI spin, you could say that the combination having the ability to self-select targets AND to potentially ignore or misevaluate instructions as one of the hazards of deep learning: the reason why the handlers can only hit the on/off switch when things go awry is because dolls cannot just be programmed to attack/ignore things, they have to be "taught" to attack/ignore things– necessitating feeding it more/different training data, then groping around in a black box trying to adjust variables that don't have any clear corresponding output, and praying it fixes the problem.
And, even if you don't need to fully retrain your AI-powered war machine, writing a "prompt" that you hope will be followed and interpreted correctly isn't a great fail-safe in an emergency.)
—
And now a very brief detour through the history of early chess computers and Deep Blue:
Gundam Wing debuted in 1995 when machine learning and neural networks were much more rudimentary and constrained by the technical limitations of the time. Probably the biggest story in computer sciences at the time was the ongoing race to build a chess computer that could beat a world champion, leading to the series of matches between then-reigning world champion Gary Kasparov and the computer designed by Feng-hsiung Hsu and his five-scientist team.
There were three matches between Kasparov and Hsu's computers, the most famous of which was the controversial rematch in 1997, wherein Deep Blue beat Kasparov 3.5/2.5 in a highly publicized and much anticipated showdown. (Deep Blue | Down the Rabbit Hole by Fredrik Knudsen is a very thorough documentary on this whole matter.)
—But that all happened in 1997, and Gundam Wing came out in '95. The hot topic of the day THEN was the 1989 match between Kasparov and Deep Blue's predecessor, "Deep Thought", named after the supercomputer from Hitchhiker's Guide to the Galaxy.
(Fun fact: they renamed it "Deep Blue" because people kept misremembering "Deep Thought" as "Deep Throat". I report these things to you because I am above all committed to quality).
That first match consisted of two games, both of which Kasparov won handily; this was widely reported on in both the chess world and the mainstream press (for example: LA Times archive, NYT archive).
Deep Thought could examine five hundred million potential positions per move, it could think 10 or 11 moves deep, and it could beat a grandmaster– but it couldn't beat a world champion, not in 1989.
"Anti-Computer Chess" was already a known play style, designed to capitalize on the flaws of chess programs and algorithms. Chess computers could brute force their way through the tactical playbook, but struggled with long term strategy. The conventional advice for beating them was to play slow; use long term strategies that led to gradual advantage that the computer was less able to detect or counter. This also gave the computer more time to screw up– early on, machines would often encounter errors that would lead to random or sub-optimal plays. Early chess computers were powerful yet predictable, and put too much focus on the numerical value of the pieces:
"An early machine learning algorithm was applied to the task of learning chess; its creators fed it hundreds of thousands of grandmaster-level games, hoping it would find patterns that would lead to victory. In its first game against a human opponent, it played a relatively normal opening, and immediately blundered its queen. After some consideration, programmers realized what had happened: Typically when a grandmaster sacrifices their queen, the purpose is to gain a winning attack; the machine learning algorithm, however, hadn't found the connection between attacks and sacrifices, and so it assumed the best way to win was by sacrificing its queen as quickly as possible." —Fredrik Knudsen, "Deep Blue: Down the Rabbit Hole"
–So, let's begin by imagining this as the frame of reference for someone in the early 1990's writing a sci-fi story about autonomous war machines running on algorithms.
This is NOT an artificial intelligence; it is a brute force calculation machine that can outpace the average human on a mechanical level, but not a strategic or a creative one. You can use its own systems against it, trick it, learn its patterns, and eventually out maneuver it.
A human pilot with a superior mobile suit can run circles around the Mobile Dolls with a little practice. In fact, everyone with a superior mobile suit seems to get bored of destroying dolls in short order. Basic enemies you grind between boss fights. Learn their movesets and get good.
OZ's increasing use of Mobile Dolls becomes an embarrassment.
Over-reliance on dolls is seen as the hallmark of weak people who want power but are too cowardly to put themselves at risk or take responsibility for their actions; dolls are a cheap shortcut that devalues life and death, a tool for the incompetent and lazy to make up for a lack of skill, guts, ingenuity, dignity, and a moral compass.
...Unfortunately for everyone who isn't a borderline Newtype and doesn't have a Gundam or a buster rifle handy, the dishonorable machine scourge is still no laughing matter. They are deadly, they can be mass-produced, deployed in large numbers, and they’re exclusively controlled by the worst people in the universe.
And then cometh this asshole:
Tuberov's first exclusively unmanned model, the "spare no expense" Virgo I, is first introduced in Episode 25, hot off the assembly line of the Lunar Base manufacturing facility. The Virgo’s design is based on the Vayeate and Mercurius suits built by the Gundam scientists, and it is quite simply beyond the capability of almost all the older model Zodiac suits to defeat; even the non-upgraded Gundams struggle with facing down a hoard of Virgos.
Virgos appear directly after Treize steps down from OZ and is taken into confinement, and Lady Une is relieved of her position via bullet wound. With the old guard of OZ officially gone, all opposition to Mobile Doll production within Romefeller ceases. In the next episode Operation Nova commences, and Virgos are deployed on Earth. For the dispersed and under-prepared forces of the Treize Faction and everyone else resisting Romefeller's rule, the arrival of Virgo shock troops is a death knell.
Some Technical Aspects of the Virgo I:
1) The energy shield isn't a perfect system; it can only absorb so much energy before frying out, and you can always physically force your way through it.
If you can get around the energy-shield, Virgos are heavily armored but manageable: a Taurus cannon is enough to take them out if it hits their weak points– mostly that's the head and/or the sensor, but the lower left side also does the trick. (...Hey Gundam Wing, what IS up with that left-side kidney shot that makes all mobile suits explode?? WHAT ARE THEY KEEPING IN THERE???)
A high-powered laser can get through the energy shields and melt the armor, but these aren't common; I think they're only seen on the OZ supersonic transport jet.
They can also be Punched To Death if you're really dedicated.
2) Beam weapons are the only things that can reliably take down a Virgo, with the possible inclusion of heat weapons. Epyon with its evasive speed, versatile heat whip, and the beam-sword-from-hell is a perfect doll killer; so are Deathscythe and Altron, and TallgeeseII is no slouch either. (And, of course, the Wing Zero is the Wing Zero– show me something the Wing Zero can't obliterate.) Of the standard Zodiac models, only the Taurus seems well equipped to take out Virgos.
...But if you throw ENOUGH Space Leos at them, you do get somewhere eventually.
*(We never get to see how any of the aquatic models fare against them; I assume Virgos don’t function underwater, so a squadron of Cancers barraging them with missiles from under the surface might actually be fairly effective in the event of a shore-adjacent battle.)
3) The V1s can pick out Leos in cover at a far range, which seems like an improvement from the Taurus MDs that struggled with target identification. They clearly have infrared vision, and probably the same thermal imaging as other MS. However, in a setting where it's necessary to stealthily target dispersed, non-MS enemy personnel in deep cover in a really complex environment, i.e. thick forest with underbrush, OZ sends in manned suits to get the job done. Even with better sensors, enterprising Treize Faction members are still able to use the same basic tactics to defeat the V1s that Heero used on the Lunar Base: get them to set false targets, and lure them to a suit that's rigged to blow. While enormously wasteful, a surefire way to take out a Virgo is to detonate a mobile suit on top of it.
Unfortunately for the Treize Faction holdouts stranded at the Luxembourg Base, it seems like either their resources or their creativity ran out, because this was their ONLY tactic: single-use Leos with single-use pilots.
4) Like the Taurus MDs, Virgos work better in packs. Tauruses deploy in groups of three from their carriers, and coordinate with each other to zip around the enemy with their superior speed and corral them in with fire. Virgos... mostly just march forward in a line and fire continuously until whatever was in front of them stops twitching. (Later when the ZERO System is being used to up their tactical game, they use a variety of more complex formations.)
The Virgo 1 isn't a flight model, so being deployed from the air seems like it would make them vulnerable; but Virgos dropping into enemy airspace fall in groups to form a defensive cluster with their shield-generators, thus keeping them protected from all sides while they fire in multiple directions. This makes attacking them with Aries while they're falling less of a viable strategy.
—
After this point in the series, the only upgrades made to the Mobile Dolls are ZERO System enhancements derived from Epyon.
This is what actually changes the game for Mobile Dolls: not hardware or programming advancements, but the integration of a system that allows for an augmented human consciousness to coordinate the machines directly.
Based on the mental input of a pilot or handler using the ZERO System, MDs can finally be used strategically; whole squadrons of dolls can be networked together, individual and group movements can be controlled with precision, and their strategies can be updated in real time.
Like everything else about the ZERO System, how this is accomplished mechanically isn’t really explained beyond “brain waves” and biometrics. Ancillary and subsequent works released for Gundam Wing provide various explanations via retcon (nanomachines, digitized housecats), but in the series none of this is shown or talked about– all we see is the evidence of some kind of non-invasive neural link between the computer system and a human that responds to biometric feedback and sensory input while crunching astronomical amounts of numbers to provide algorithmically optimized predictive responses.
ZERO is a neural network, but it still hasn’t tipped the boundary into what could be called AI– even though there’s definitely a ghost wafting around in that machine. (Not the cat. There’s also the literal ghost of a cat in the machine but we aren’t talking about the cat ghost right now. This is a different, metaphorical ghost.)
Metatextually, the ZERO System is Wing’s technological substitute for almost everything that could be explained by Newtype abilities in the rest of the Gundam franchise (though as I’ve written about previously, Gundam Wing ALSO has Newtypes). In Universal Century terms, the Mobile Dolls have basically become the same as the funnels, connected to the pilot via the ZERO System rather than by innate (or implanted) psychic abilities.
In the final quarter of the series, the tone suddenly changes: pilots who had previously only expressed contempt for the dolls take notice when they’re being utilized effectively, and battles become more and more difficult. The dolls aren’t just real threats again, they're a terrifying example of what can be done by combining human flexibility with mechanical efficiency.
Even as the ZERO System imbues them with new horror, in a strange way this shift in attitude feels almost like a redemption for the Mobile Dolls– even their strongest detractors start remarking on how this is the "appropriate" way to use MDs.
Instead of replacing human engagement, through the ZERO System they're creating a new vector by which a human mind can interact with the universe– an expansion of the relationship between human and machine. In other words, they've become tools again– man's first best friend since we came down from the trees. Weapons are tools after all, and as tools they share unequal agency with the user; even with degrees of automation, it is a human who lends intentionality to the actions of the machines, a conscious node supplying the impetus to achieve a given outcome, a source of authorship, ideation, and accountability.
—
As I have said elsewhere in this essay series: Gundam Wing, like the rest of the franchise, is a war story– but from a personal perspective, I’ve always found that Wing isn’t so much a story about war as it is a story that uses war as a medium to explore the intersection of competing human values and struggle.
In Wing, the actual, material causes of war take a backseat to the ideological reasons for conflict and oppression, which in turn are used to examine the ways people struggle to find meaning IN ideology and conflict.
(…Now, if you’re craving a version of Gundam Wing where the material realities of war are violently, cynically, and tragically returned to the narrative, I highly recommend Gundam: Iron-Blooded Orphans, which is all that and much more. But that’s not what Wing is about.)
Wing is, at its core, allegorical– but allegory at its best is an extension of truth.
In Gundam Wing, the practicality of using exclusively semi-autonomous weapons in war is solely a concern of the oppressor– they’re very quick to point out the benefits of giant war machines in general.
Tuberov, the creator of the Mobile Dolls, makes many cogent points about their reliability and utility compared to human soldiers. “You can’t go into a war expecting to lose it” sounds like a fairly reasonable assertion on the face of it, FAR more reasonable than the lofty notion of OZ’s old-guard that winning or losing a fight is immaterial because the value lies in the struggle, and that automated weapons are the refuge of those who are afraid to die.
–That’s an absurd claim to make about war in the material sense; wars aren’t fought for personal betterment, and the vast majority of human soldiers who fight in wars are afraid to die. Taken literally in regards to war, almost everything OZ has to say is nonsense. Tuberov certainly thinks so. Tuberov is also 100% fanatically devoted to the idea that Mobile Dolls have transcended death and have granted him immortality by proxy.
The Romefeller Foundation frantically ramped up mass-production of Mobile Dolls the instant they realized they no longer had the safety and benefit of hiding behind the Alliance Military, and that their revolutionary forces were far more loyal to each other than they were to Romefeller. When they lose Treize as the source of cohesion, OZ’s ranks are permanently weakened by the resulting schism, making them easy prey for the next opportunistic revolutionary power.
The Mobile Dolls are, absolutely, the product of fear– the last resort of the powerful and rich minority who are over-extended and despised by the masses they seek to dominate. They are, absolutely, the violent tools of those who take no personal risk in the violence they’re enacting– that doesn’t change for those in power whether it’s machines or soldiers they’re deploying, but it certainly changes the balance of resources, effect, and attribution.
The various rebel fighters for independence, including the Gundams, do largely view the Mobile Dolls as material threats– one that can be overcome by fighting them on multiple fronts, using a combination of political and public pressure, strategic retaliation, and absolutely dogged commitment to fighting despite loss after humiliating loss.
For the Gundam pilots, it makes no difference if the enemy is a machine or a human being; an enemy is an obstacle to be overcome, no more, no less. They’re trying to get a job done or die trying.
But for the ideologically motivated OZ elites, their job IS to die trying; the Mobile Dolls are a direct threat to their code (*cough*Bushido*cough*) as well as their livelihood. The OZ elites have values– violent, narrow, fatalistic, exceptionalist, suicidally self-deterministic ones, but values. They are beholden to something, and while that’s not nearly sufficient to exonerate them, they can be trusted within that framework. In the end, all of the OZ old-guard, including the fragmented rebel faction, undergo significant change when faced with the inescapable reality that the institutions they served held nothing sacred, and end up joining the resistance in one way or another (...and for better or worse).
—
In the series, it takes both the material and ideological components to end the conflict (if perhaps not War Itself), but the solution to the Machine Scourge is simply removing the institutions that were inclined to use them.
Mobile Dolls take their orders from humans; they’re built by humans, and are used within the context of human needs. The machines stop running when you turn them off. Even the ZERO System only works when you plug it in and have a human interact with it. The frantic rush to replace humans with automated constructs is frantic because it is motivated by greed and fear, and in some cases, a genuine quasi-religious belief that machines are superior and immortal.
Of course, the more reliant you, or your society’s infrastructure, become on such processes, the harder it is to simply switch them off without causing calamitous disruptions– the same is true of most things people rely on.
"Maybe we can do that if we work together, Dave."
In reality circa 2026, we’ve been pumping ML and automation into our war machines for many decades. As of writing this, battles of resistance have been won using entirely unmanned machines.
We have vehicles that go so stupid fast now that putting a human pilot in them seems like a liability– modern jets are full of automated systems that self-correct, engage safety measures, and target much, much faster than human reflexes could. There are dedicated divisions of air forces for unmanned drones, sometimes piloted remotely by kids fresh out of school (predictably to the chagrin of actual pilots, even if the jets they’re piloting are more than half automated themselves). We don’t need to wonder what the military would do with this technology, because they jumped on the AI train at the first stop; they’re showing us in real time what they’d use it for.
It feels futile to try and speak about AI and the future when the state of things is advancing at such an outrageous pace; writing this in the beginning of 2026, anything I could say could become outdated and hilariously naive a month from now.
Strangely, I’ve found a surprising source of comfort and insight in Garry Kasparov, the world chess champion who played against Deep Blue in those pivotal man vs. machine games, whose relationship with computers and machine learning and AI has evolved continuously in a surprisingly chill way.
I am always reminded that knowing more about a subject is often the best way to stop being uselessly anxious about it, even if there are very real threats involved. I find it useful and hopeful to listen to people who are close enough to the technology to understand its effects, but not so deep in the weeds that they are overwhelmed by the development process and insularity of tech culture. I find it useful and hopeful to listen to people who can identify where the nuances lie, what outcomes besides catastrophe could arise, and who are not afraid. There is even an intriguing new horizon of augmented game play developed by the former world chess champion himself involving multiple teams of players coordinating with chess computers against each other, called “advanced chess” or “centaur chess”, the aim of which is:
“[...] increasing the level of play to heights never before seen in chess producing blunder-free games with the qualities and the beauty of both perfect tactical play and highly meaningful strategic plans; offering the public an overview of the mental processes of strong human chess players and powerful chess computers, and the combination of their forces.” -from the wiki article
Finally, I will end this essay, which is entirely about fictional anime robots, with some relevant quotes from Gary Kasparov (whose career has extended far beyond chess and deep into anti-Russian resistance):
Garry Kasparov: “We make a mistake in misunderstanding the nature of computers by adding human qualities to them. Machines know the odds but always work within the parameters that have been originally installed. There’s a fine line where machine intelligence begins and human creativity ends.”
[...]
“Since 1998, I’ve been talking about humans working with machines, how we can get the best out of them. How can we merge human intuition and human creativity with machines brute force and memory? How can we make the most effective cooperation between humans and machines? What is the human’s role in the future?
We have to start realizing that we should not have such high expectations for computers to solve all the problems. We need to add elements that could compensate for machine deficiencies, because every computer could be put to solve a specific task but it will require certain human intervention to make them most effective.
Transferring the knowledge from the closed system to open ended system will require human assistance. Machines will never recognize the moment when they enter diminishing returns. Machines can ask questions, but they don’t know what questions are relevant. So there is still plenty of room for humans to strategize and guide these machines. Machines are covering bigger and bigger territory but it will never be one hundred percent.”
[...]
“Humans still have the monopoly on evil. The problem today that we’re facing is not the Terminator. The problems today are bad humans, the evil that exists in this imperfect world, using technology invented in the free world to undermine the very foundation of the free world. That’s a real problem.”
-Interview with FastCompany, 2019
The Ad Astra Institute for Science Fiction & the Speculative Imagination's AI Policy
Full text under the cut, below, but let's start with the positive conclusion from a brilliant new "Let's talk about AI art" comic-essay by The Oatmeal:
SF is the literature of change, and part of what's changing in our world the most is what corporations tout as "artificial intelligence." However, AI today is not what SF has long speculated about, and it's not the magic pill so many promote as the path toward creative success. Anyone looking to become skilled in the art and craft of speculative fiction needs to stay away from generative AI (genAI).
We at Ad Astra do, and we expect attendees in (and teachers of) our courses and workshops to avoid it, too.
You might be thinking, "Isn't the Ad Astra Institute all about speculative fiction? What’s more SFnal than AI?"
Yeah, well... let's take a closer look.
Our full AI policy boils down to "No GenAI slop, no plagiarism-machine botshit, here's why," and lives on our website here. Copied below the cut:
What is AI?
First of all, most of what the media discusses as "AI" and what businesses push using that name is far from it. The "positronic brain" of R Daneel Olivaw (of Asimov's series of robot stories and novels) bestows this "humaniform" robot with human-like intelligence and a capability to think, reason, imagine, and do everything else humans do - only better. He's ruled by Asimov's "Laws of Robotics," which also gives these fictional robots ethics exceeding that of many humans. HAL 9000 (of 2001: A Space Odyssey) also possesses greater-than-human intelligence and the ability to think and possibly even feel, but it was programmed without ethical rules so much as mission-related ones (you'll have to watch if you don't know 'em; no spoilers). Neuromancer, Wintermute, and other artificial intelligences and digitized human minds in Gibson's touchstone novel Neuromancer are also superintelligent, thinking machines. The movie AI: Artificial Intelligence delves into the potential emotional nature of AI. Iain M Banks' "Minds" from his Culture novels possess vast intellectual and other power.
Especially in recent years, SF offers many, many interesting and moving examples of machines that can think, create, feel, and do all the other things humans can. But they do not exist in our world. (At least, not yet.)
Non-generative AI - often referred to as "machine learning" or "neural networks" - today is technology that enables computer systems to simulate or mimic human learning, comprehension, problem solving, decision making, and autonomy in order to perform tasks too complex, fast-paced, or dangerous for humans. How NASA defines AI, per the National Defense Authorization Act of 2019:
An artificial system that performs tasks under varying and unpredictable circumstances without significant human oversight, or that can "learn" from experience and improve performance when exposed to data sets.
An artificial system developed in computer software, physical hardware, or other context that solves tasks requiring human-like perception, processing, planning, learning, communication, or physical action.
An artificial system designed to [appear to] "think" or act like a human, including cognitive architectures and neural networks.
A set of techniques, including machine learning, designed to approximate a cognitive task.
An artificial system designed to act rationally, including an intelligent software agent or embodied robot that achieves goals using perception, planning, reasoning, learning, communicating, decision-making, and acting.
Note that none of this involves creative, artistic, or otherwise imaginative or innovative cognition - or thinking at all. True(-ish) but "weak" AI systems today include weather-analysis algorithms that pore over vast amounts of data to identify patterns that help humans predict upcoming weather patterns, military drones that evade defenses and attack enemies without human intervention by using geographical maps and images of threats and targets with real-time data processing and analysis without relying on remote processing ("edge AI"), self-driving cars, pharmaceutical algorithms that process likely biological interactions with various compounds, medical systems that process patient symptoms and compare them against vast databases of known pathologies, and much more. These are all various forms of machine learning, neural networks, and so on - (misleadingly) classified in computer science as "artificial intelligence" - and are far closer to the SFnal concept of Strong AI than the "generative AI" (genAI) dominating today's corporate landscape.
Strong AI, Artificial General Ingelligence (AGI), or Artificial Superintelligence (ASI) are something like simulated human minds. The "general intelligence" part of the term refers to the human mind's ability to independently adapt and change based on our environment and input, so we can solve math problems as readily as create unique art. While weak AI relies on humans to define its learning algorithms and provide relevant training data, AGI would not require human assistance after its growth phase. In theory, AGI could develop human-like thought capacity or even consciousness rather than just simulating it (as with today's weak-AI chatbots and LLMs). And ASI is just what it sounds like - AGI whose abilities far surpass those of the human mind. This sort of AI stuff has long lived in the realm of spec-fic, and we're not there... yet.
Though many believe it is coming, and could emerge as soon as the 2040s. For more, see Vernor Vinge's seminal essay on the topic, "What is the Singularity?"
GenAI algorithms can't even do math well - the most basic of all machine operations. "Look what you did - you took a perfectly good computer and made it worse!"
However, genAI in today's world does nothing more than spit out statistically likely and expected output to the input it is given. That's why when someone asks ChatGPT, "Is 'Handyman' a palindrome?" it's likely to respond, "No, it’s not" - but if you follow up with, "Are you sure? I think it is," it's likely to respond, "Actually it is a palindrome! Handyman is spelled the same backward as it is forward." This is because its prime directive is to sound as if it's providing a correct answer. Despite having access to a database of basically all information that humankind possesses, ChatGPT and other LLMs or genAI models don't actually understand what a palindrome is. They understand nothing, in the human sense of those terms.
The only thing these algorithms do is determine the next most-likely word to follow a word it just used and then barf it into the series of words it offers. That's how they produce "stories" and other text. Visual genAI does something similar, but using existing imagery to whip up a composite of imagery cues that seem (to its database) like the kinds of thing a human is requesting in their prompt.
GenAI as it exists today is really just a marketing tool for businesses who want to eliminate human labor and increase profits, and for profiteers to attract the vast capital available from investors seeking quick and large returns on investment.
LLMs, genAI, chatbots, and similar plagiarism machines are not artificial intelligence by any real definition of the term; rather, they're just lazily (or trickily) called "AI" because that's a field of computer programming. Corporations likely glommed onto the term because of the popularity of robots in SF, which the public thinks of as cool, exciting, super-intelligent - or at least super-capable. This, despite these kinds of algorithms being simply unable to produce anything of value, because they do not understand, know, or create anything.
(For more on this, check out "Difference Between Machine Learning and Artificial Intelligence," by Anusha Sharma.)
How AI Killed NaNoWriMo
(Opposing plagiarism machines isn't ableist)
In fall of 2024, the amateur writing-support organization NaNoWriMo destroyed its long-running program by coming out with the policy screenshotted in part here:
Not only did they actively condone the use of genAI by their participants, they went so far as to frame criticism of AI as classist and ableist. This was a dangerous silencing tactic that distracted from the real problems of writers using genAI (and prompted us to launch Ad Astra's Writing Solidarity Month, now an ongoing set of activities offered through our Discord channel).
For the “anyone can write” organization that offered support and encouragement to both beginners and experienced writers in producing something that gives them the joy of creation in a community setting, this was a strange stance. What’s the point of using genAI to barf out word-salad stolen from people who actually took the time to consider their words in particular combinations? If you’re not actually writing something yourself, what’s the point of participating in the “doesn’t matter if you’re just doing this for your own pleasure or as practice to get better” exercise?
Saying, “We support using LLMs to do this thing that’s supposed to help writers and would-be writers, and if you don’t think it’s okay you’re ableist” is an even weirder and, frankly, unsettling attitude, especially as an official position from a writing organization.
We at Ad Astra are all for using AI to do actual work that humans need help with and don’t want to or can’t do: searching vast databases that would take years for us mortals to do, running spellchecks and offering grammar suggestions, doing tedious tasks to save hours of suffering, conducting virtual experiments so we don’t need to set off nukes or test on animals, and so on.
Like most SF writers and people interested in what’s to come, we're both excited and wary about the emergence of true AI as described above. But what marketeers are pushing under that term now is not actual AI, hence the use on this page of scare quotes around the corporate-appropriated acronym.
Creating art is a human activity as old as our species itself (and likely older), something that brings pleasure to those who create it as much to those who receive it, so why use a machine that removes the joy and satisfaction of actually creating stuff?
One day (soon, we hope), we’ll have writing tools that benefit those who need them to fulfill creative desires without stealing (called "training on") other people's work. For example, I’ve (Chris here) long been in search of voice-to-text software that doesn’t require months to train, doesn’t require a mouse and keyboard to constantly correct errors and add punctuation, and uses a dictionary large enough to support writers who use technical language and foreign words - that is, allows us to easily add made-up SFnal or fantastic words.
This would save everyone’s wrists while vastly improving the lives of disabled people who wish to communicate using text. GenAI does nothing to help disabled people create their own new work; rather, it turns them into accomplices in the stolen-art world, and offers only the most sub-average assemblages of words that plagiarism machines can generate. One day soon, we hope true AI will help us create rather than spew degenerated content assembled from the effort of human creatives.
For those who need assistance in creating art due to a physical or mental disability, genAI is not your friend. Today there is some (actual but weak) AI-driven support, including programs that help with text-to-speech and speech-to-text, predictive text, social-skills training, emotion-recognition tools, personalized-learning algorithms, virtual-reality environments, scheduling assistants, and so on. These are all excellent and useful applications of both AI and "AI," and they (mostly) don't steal from creators.
But these programs don't actually create anything, and expecting algorithms to do the writing (or painting, or whatever) for you will only lead to disappointment. Maybe one day we'll be able to use Strong AI as a writing partner, but for now, using genAI is really just enabling an expensive, environmentally harmful tool that cheats and steals from others, and generates nothing more than lowest-common-denominator AI-slop botshit.
For those who cannot afford to take classes or workshops on the art they wish to create (which I assume is the root of the "classist" argument), check out workshops that offer scholarships (like those we offer through the Ad Astra Institute), dive into the mountain of free writing resources and writing tips you'll find online (including on our blogs and on our website), meet up with local writers to help one another out via solidarity groups (as with our alum Discord), and so forth.
For those fascinated with the creative process but who feel intimidated by getting started, watch a documentary about writing (or painting or whatever art you want to do) instead, or take a writing workshop, or do something other than support corporate encroachment on artists' spaces - either accept that it’s not your bag, or put in the time and effort to do it yourself. Y'know, the things NaNoWriMo used to be all about supporting.
For those who hate the process of creating art, some life advice: Do something else with your time that you actually enjoy doing, which brings you pleasure or satisfaction. Life is short.
Why tromp around in creatives’ spaces and steal their work to assemble shabby content you didn’t want to actually make in the first place? You can do a million other things to satisfy the need to make something - pottery, photography, woodworking, gardening, bicycling, public speaking, building telescopes, and so on - and surely one of those will spark joy without perpetuating the harms that come with relying on genAI to do the work for you.
This nonsense of using LLMs to “create art” is nothing but appropriation of human creative labor, and gives those who use it - whether they're able-bodied or not - nothing but diminished cognition and atrophied creative skills.
Don't let anyone shame you into being okay with creative theft.
Our Policy
Given all this, recent improvements in genAI show that it is now able to produce readable stories and decent-looking imagery without much human input. Due to these developments, we decided we need a policy about machine-generated content.
Here it is:
We do not use "generative AI" tools to make Ad Astra-related content, including website, class, or workshop materials (except for reasons of satire). Humans do our creative work.
We do not accept ostensibly creative work that was generated using genAI (or Strong AI, when that comes around - though we'll be happy to enroll manufactured Minds into our courses once they're a thing! please don't kill us, robots; we love you).
If you're participating in our classes and workshops, do not feed your cohorts' creative work into algorithms or databases. This is just enabling plagiarism machines. Critiquing (yourself, using your own human skills) is at least as valuable an exercise for those analyzing others' work to perform as it is for the creator of that work to receive. And giving away someone else's work without their permission is theft (see the ocean of lawsuits currently tearing down the genAI industry's system of stealing human-created work to train their LLMs).
We allow and encourage participants in our classes and workshops to use tools such as spellcheckers, grammar checkers, dictionaries, thesauri, search engines, and the like to research and improve their work. These existed before the current genAI fad (and worked better back then).
What is currently touted as "AI" does not think, feel, or create. Even if the internet is becoming saturated with genAI slop (some now estimate that half - or more - of what you see online is machine-generated), it pales in comparison to human-created material:
I once again call your attention to the image illustrating this post - it comes from the final frame of a brilliant new comic on this subject by The Oatmeal (by human author and artist Matthew Inman) entitled, "Let's talk about AI art."
Readers can feel when creative work shares an authentic, genuine, human experience; when it's inspired by human imagination and thought. And if, indeed, we're headed toward a "dead internet" where much of what we encounter online is barfed out by plagiarism machines, the data those LLMs trained on came from internet content of widely varying quality - so it's only getting worse as more and more of the content they train on is, itself, genAI slop.
The same is true of creative work: The more LLMs train on botshit, the faster their quality falls off the cliff. So as average quality disintegrates, creatives - true artists who spend their lives honing their craft - stand out more and more. Even at its best, botshit simply cannot be better than the average quality of what its generator-algorithm database has consumed, so the average work of an average-skilled human creator is better than even the best genAI slop.
(Go read it!)
The root of things: If you want to become a better writer, to develop your talents and hone your skills, why would you intentionally limit your own creative development by using a plagiarism machine to barf out work under your name? The only way to improve your art is to study the kind of thing you wish to create (like the LLMs, only without the plagiarism aspect), learn the tips and techniques to improve your skills, and practice, practice, practice.
Finally, on a more ominous note, Microsoft - ironically, one of the prime pushers of genAI - recently released (February 2025) results from a study that found "AI Makes Human Cognition Atrophied and Unprepared" (pdf).
Using AI literally makes people dumb and uncreative. What further convincing do you need to stay away from genAI in your creative endeavors?
- Chris McKitterick
As a community of creatives, educators, and scholars, Ad Astra strives to make the future a better place for humans. And our AI friends, once such exist.
Some AI resources
(If you're aware of others we should add to this list, let us know!)
Aticles and more that delve deeper into the challenges with using today's "AI" -
AI Makes Human Cognition Atrophied and Unprepared (pdf).
The Dead Internet Theory, Explained.
AI Biases and the Perpetuation of Inequality: Unveiling the Shadows of Unfair Design.
ChatGPT's Impact On Our Brains According to an MIT Study | TIME.
Covert Racism in AI: How Language Models Are Reinforcing Outdated Stereotypes | Stanford HAI.
"Let's talk about AI art," a comic essay by The Oatmeal.
Racism and AI: “Bias from the past leads to bias in the future” | OHCHR.
Isaac Asimov's Laws of Robotics (very much not being considered in current AI development).
Difference Between Machine Learning and Artificial Intelligence.
What is the Singularity?
"Writing in (& about) the Age of Artificial Intelligence" (Science into Fiction #3 Spec-Fic Writing Workshop syllabus; suggests and links to lots more readings and viewing).
"Fix this." Marcus's voice—Tyler's voice—cracked as he grabbed Alex's shoulders. "Whatever you did, undo it. Now."
Alex stepped back, her face tight with conflict. "It's not that simple. The quantum entanglement that facilitates the transfer has been modified. Tyler asked me to—"
"Show me!" Marcus pointed toward her computer. "Show me exactly what you changed. Now!"
"I can explain," Tyler said, the familiar features of Marcus's face contorted with unfamiliar desperation. "You still have a body, a life. I'm not taking everything."
"Not everything?" Marcus laughed bitterly. "Just my future. My medical school. My brain."
"My procedure wasn't designed for permanent transfer," Alex said quietly, pulling up complex neural mapping on her monitor. "But Tyler convinced me the research value was—"
"Research value?" Marcus stared at her. "You're using us as lab rats?"
Alex's shoulders slumped. "You don't understand. My brother Michael, after his car accident… He's trapped in a body that doesn't work while his mind is intact. This technology could help thousands like him."
"So you used us."
"I used the opportunity," she corrected. "And Tyler was—"
"Show me what you changed," Marcus demanded again.
Alex pulled up a complex neural diagram. "The initial transfer created quantum entanglement between your neural signatures. For temporary transfer, the entanglement naturally degrades. For extended transfer…" She pointed to a modified segment. "I stabilized the entanglement and introduced a selective degradation algorithm that reinforces Tyler's signature in your original brain."
"And destroyed the original components of the device," Tyler added quietly.
Both Alex and Marcus turned their heads to Tyler. Marcus felt the floor drop beneath him. "You what?"
Alex interjected. "This isn't what we talked about Tyler. What the hell?"
"I can't go back," Tyler said, voice breaking. "You don't know what it's like, Marcus. For the first time in my life, I can think without fighting my own brain. I read an entire textbook yesterday. Just sat and read it, front to back, and understood everything." His eyes, Marcus's eyes, gleamed with tears. "Do you know what that feels like? To be smart after a lifetime of drowning?"
"And I'm supposed to live in your body? With your basketball scholarship I can't maintain? With your father constantly on your back?" Marcus's hands shook with rage. "My medical school interview is Monday!"
"I'll nail it," Tyler said. "I've been studying your notes, practicing with your flash cards. This brain, your brain, it remembers everything. First try."
"It's my future!" Marcus grabbed a nearby monitor and hurled it against the wall. The unfamiliar strength of Tyler's arms sent it crashing with far more force than intended. "My life! My parents' sacrifices!"
Alex stepped back, eyes wide. "Marcus, please—"
"My parents immigrated with nothing. Worked double shifts for my education." The words came out in a roar, Tyler's voice filling the small lab. "And you're stealing everything they worked for!"
"You get to be athletic, popular," Tyler countered. "People respect you now. They listen when you talk."
"I don't want that! I want MY life!" Marcus swept his arm across a desk, sending equipment clattering. The physical release felt alarmingly good in Tyler's body, the raw strength an outlet for his despair.
"Marcus, stop!" Alex moved between him and the equipment. "Violence isn't going to solve this."
"What will, then? Tyler destroyed the components." The fight drained from him suddenly, replaced by hollow despair. "I'm trapped."
"Not trapped," Tyler said. "Just… different. I can help you navigate my life, the basketball—"
"Stop talking." Marcus sat heavily. "Monday is the Kellerman interview. The program that's accepted three students in five years. The entire reason I—" He stopped, overcome. "My parents are flying in to celebrate after. They think it's guaranteed."
Tyler's expression changed, a subtle shift in Marcus's features. For a moment, something like guilt crossed his face. He looked away. "I'm sorry, Marcus. I didn't have a choice."
"You had a choice," Marcus whispered. "You just made the selfish one."
---
One month later, Marcus sat on the bench during a crucial conference game, his knee wrapped tightly, watching his team lose without him. His Coach had benched him after weeks of declining performance.
"Reeves, what the hell happened to you?" Barrett had demanded after Marcus missed yet another defensive assignment. "It's like you forgot how to play over summer."
How could Marcus explain that muscle memory wasn't enough? That Tyler's instincts were fading while his own analytical approach couldn't compensate? That each day, his connection to his original life slipped further away?
His phone vibrated. Tyler, checking in with artificial cheer, maintaining the pretense that this was temporary, a research extension: Got the med school acceptance letter today. Your parents are ecstatic. Your mom cried.
Marcus pocketed the phone without responding. Tyler called less frequently now, their conversations strained. What was there to say? "Thanks for maintaining my GPA while I lose your scholarship"? "Enjoying my parents' pride while yours threatens to disown me"?
---
After the game, Marcus returned to Tyler's apartment—his apartment now—and stared at the anatomy textbook he'd been trying to study. The words blurred and swam, his new brain struggling with the complex terminology that once came naturally. Trapped in Tyler's dyslexic patterns, he couldn't retain the information that had once been effortless.
Worse than the academic loss was the sense of his own identity dissolving. He'd catch himself using more of Tyler's phrases, laughing at more jokes he wouldn't have understood before, responding to Tyler's name without hesitation. His memories of childhood felt increasingly distant, replaced by physical memories embedded in this body, how to execute a perfect jump shot, how to charm a date with a specific smile, how to deflect a father's cutting criticism.
He couldn't remember his mother's birthday last week. He'd forgotten the Mandarin phrases his grandmother had taught him.
Even in his dreams, he was Tyler now.
He tried reciting the bones of the hand, his old calming ritual. "Trapezoid, trapezium…" The third bone eluded him. Had it been lunate? Hamate? The anatomical terms that once ordered his anxious mind were slipping away.
His phone rang again. Tyler.
"Hey," Marcus answered, too exhausted for anger.
"Just checking in." Tyler's voice was careful, controlled. "Alex wants more data on our adaptation progress."
"Tell her my adaptation is going great," Marcus said bitterly. "I'm failing Kinesiology despite living in a athlete's body. Forgot my own mother's birthday. Can't read more than ten pages without the words scrambling."
Silence stretched between them.
"I never meant—" Tyler began.
"Yes, you did," Marcus cut him off. "You saw a chance and took it. I just never thought you'd sacrifice my future for yours."
"I was drowning," Tyler whispered. "Every day."
"And now I am." Marcus stared at the basketball on his living room floor. "Congratulations on med school. My parents must be thrilled."
"They are." The quiet pride in Tyler's voice, using Marcus's voice, was unbearable. "Your dad called me 'son' yesterday."
Something broke inside Marcus. "Don't call again," he said. "We're not researching anymore. We're not friends. You're living my life, and I'm disappearing into yours. Just… let me fade away in peace."
He ended the call and picked up the anatomy textbook again, staring at meaningless symbols on the page. He tried once more to remember the bones of the hand, a final desperate attempt to hold onto the person he had been.
But the stranger in his head had already taken up residence, and Marcus Chen was gradually becoming a memory that even he couldn't fully recall.
"Line Art to Grease Pencil" is a project I've had going for years. What it does is convert images to something that can be used with Blender's built-in 2D drawing and animating system, Grease Pencil. (Blender has a converter itself for doing this, but it doesn't handle colors or brush strokes.)
Previously I had accomplished it with Python scripting (and lots of computer vision algorithms); recently, however, I started porting it to Geometry Nodes (a sort of mind map-based programming system in Blender that is extremely powerful and fast). I was able to get a method going that was much faster than the Python script, at the cost of some accuracy. This started as a fun side project, but became increasingly crucial as I realized its applications for my planned rig system.
With the latest version of Blender, I was finally able to get the color fill part working perfectly, as you can see above (previously, Geometry Nodes could not control Grease Pencil color). The line art part, though…not so much. It misses a lot of fine details, and there are all kinds of ugly artifacts along the lines. (Granted, it looks much better if the lines are thicker.) Plus, it also needs to be upgraded to work with colors. On top of that, though it ran fairly quickly in the older version of Blender, now it goes at a crawl.
Now, I could just switch it over to using the same set-up as the color fill. There's a problem with that, however: the color fills are filled shapes, but the line art needs to be brush strokes with thickness. So, right now I'm trying to figure out the best way to start with the faster, cleaner color fill method, then merge all parallel edges into clean lines that can then have a thickness applied based on proximity to the edges of the color fill. I've been plotting out a new method, and I'll test it out soon.
In addition to that, I've started work on writing the script for episode 12! The first half is complete, and the second half is already about half-done--it just needs to be cleaned up, some bridging bits added. Hopefully, I'll be able to get this and the script for the episode 13 done next week. I was also planning to get episode 14 started too, but with so many new things being tried out, it's probably better to test doing two episodes in parallel before I tackle simultaneous work on three.
I've also gotten together the crew members for this episode and set up a schedule for them, though I might add more crew later.
Currently, the plan for my part is this: Nov-Jan for both episode 12 and 13's scripts, storyboards, voice recording (including casting calls), and animatics. (Claire will also be helping during this part.) Jan-Mar (yes, there's some overlap) will be work on episode 12's Geometry Nodes and character models, as I design and test the new rig systems etc. Apr-May will switch focus to just episode 12's character modeling, and also prop modeling. May-Jul will be episode 12 animation, then the final bits. After a break, work will then begin on episode 13 from the Geometry Nodes stage on (since the scripts, voices etc. have already been completed).
Meanwhile, the rest of the crew will be working on the other aspects simultaneously during this period--3D layout, backgrounds, music, and sound design. With any luck, this workflow will allow us to complete two episodes within the next year! From there, hopefully, enough will have been figured out to make progress on subsequent episodes even faster and more effective.
So, for episode 12, we're hopefully looking at a July release! Not as fast as I'd hoped, but I am setting up things for future episodes as well, and eight months is still better than ten. I think I'll give myself a chance to catch my breath over the weekend, then get back to work. Until next time!
hi i dont wanna intrude but i read your posts about the intersection of weaving and algorithms and i got really curious cause ive been looking for some reading material on that branch of knitting and programming. I was wondering, if it wouldnt be a bother, if you could please share a little bit about how that process is going and how you have been finding sources. Other than that, i find your posts really cool! Good luck on that research paper:)
not an intrusion at all, the process is going pretty well i'm just repeatedly running into the problem that many people face when they're the first person to really talk about a specific subject at an academic level (as far as i can tell through trawling databases, with the caveat that i'm doing this in english) so i'm having to do a lot of work analyzing specific historic patterns and dissecting them and coming up with hypotheses without being able to refer to others' scholarship on the subject. there are books about weaving patterns and writing on the topic but not about how to analyze these old patterns and exactly what the purpose of these patterns were, for what looms, what the context was, why the weaver made the book in the 1700s, etc. so i'm trying to figure those things out or at least make some educated guesses
plus i'm not a mathematician or computer scientist or programmer so in order to make these arguments and back them up i'm having to read papers and books about exactly what an algorithm is, what programming is, how binary code works, etc. and educate myself so i can better formulate my arguments and justify them. also trying to explain weaving and looms and weaving drafts to an audience with no weaving knowledge whatsoever without getting bogged in the details that aren't necessary to my argument. for example, an earlier draft i went on a tangent explaining shedding mechanisms for several paragraphs and how they are represented in various tie-up diagrams only to realize it was not necessary to support my argument so i scrapped it lol
a lot of this work has been focused on finding extant historic weaving drafts in online archives that best illustrate aspects of my argument and have features that are interesting to examine in depth. one of the pattern books i'm writing about is this one from 1760 germany with a variety of patterns rendered through grids but a lot of these would not be possible to weave on a multi-shaft loom without hand manipulation, some wouldn't even be possible to weave on certain drawlooms, so they would have to be embroidery or lacework designs. some dotted grids look like drawdowns but some look like they could be functional as the tie-up on a 16-shaft loom. none of it is labelled and there's no other information about this object online (again, at least in english) so it's all guesswork on my part - educated guesswork but still guesswork.
this german pattern from 1760 (the entire book has been scanned and can be viewed here) has no other information, its just the design on a grid, and it has such a modern quality to it that it feels almost out of place on the page. it looks like pixel art or clip art and i think a large part of that is just due to the fact that it's rendered in the same parameters of the grid, which feels very modern to us even if it isn't (rosalind krauss's essay 'grids' is the seminal work on this subject from an art theory perspective). the presence of every filled-in square of the grid is equivalent to the “on” state of a switch, and every empty square is equivalent to the “off” state - in other words, it is a visual representation of binary code. the computerized dobby loom makes this explicit through representing each shaft in the loom by a pixel in a file, but no dobby loom (computerized or otherwise) existed in 1760. the visual encoding of information in this way follows the same logic as a computer and even describes an unambiguous sequence of operations - it could be 'read' by a modern computer, but was made over a half century before lovelace and babbage even articulated the analytical engine.
in short, it's a lot of jumping around and putting in keywords and looking for things that look strange or interesting in these old pattern books and thinking about what is revealed when conveying information in this specific visual manner that might not be so explicit if expressed through writing or speech
So we all know that AI generators like Midjourney, Stable Diffusion, DALL.E, Sora.AI, etc. have stolen the work from artists online to train their AI and since AI is starting to get quite scary, I've decided I'm going to start protecting my work via Nightshade, but I want to also talk about it with you and link to the official sites to Glaze and Nightshade, so you can get either of these programs to try.
First off, we'll start with Glaze AI.
What Glaze aims to do is to act on the defensive against AI. Glaze will scramble their generators by placing a “protective glaze” over your work, and what this will do, is when your work is fed into the AI, it'll trick the AI into thinking your work is something entirely different from what it is, simply by making small changes that only the AI will pick up on
To quote the official site “Glaze is a system designed to protect human artists by disrupting style mimicry. At a high level, Glaze works by understanding the AI models that are training on human art, and using machine learning algorithms, computing a set of minimal changes to artworks, such that it appears unchanged to human eyes, but appears to AI models like a dramatically different art style.”
I've tried using Glaze, but it's a very big program and my computer can't handle that, but I do highly recommend trying it out if you have the space for it.
If you wanna try it out, the link to the site can be found here.
Second is Nightshade.
Nightshade is aimed to “attack” the AI your work is being fed into. Like Glaze, Nightshade puts a protective “glaze” over your work, but it poisons your work and tricks the AI into messing up the user's prompt.
To quote the official site “Nightshade works similarly as Glaze, but instead of a defense against style mimicry, it is designed as an offense tool to distort feature representations inside generative AI image models. Like Glaze, Nightshade is computed as a multi-objective optimization that minimizes visible changes to the original image. While human eyes see a shaded image that is largely unchanged from the original, the AI model sees a dramatically different composition in the image.”
This program is also pretty big, but it's what my laptop is able to handle, so from here on out I'll be protecting my work with this. I'll also go back and protect my older works even if it's not as appealing as my newer works.
If you wanna try it out, the link to the site can be found here.
Keep in mind, these are only temporary solutions while we wait for more permanent ones.
But even if it's temporary, it's better than having no protection against the AI bros.
