The AI Conundrum: Are BattleTech’s Human Creators on the Verge of Extinction?

Atlas AI 2 Courtesy of Bishop Steiner

Courtesy of Bishop Steiner

If you’ve spent even a moderate amount of time on the internet these past few months, you’ve probably come across a picture of someone with way too many fingers, or a stilted conversation between Bill Gates and Socrates, or a pixelated sitcom where there are long stretches of silence while everyone keeps bumping into furniture. Our first forays into AI-generated media have been weird, mildly entertaining, and maybe more than a little frightening. 

The core of this fear has been the long-held belief that the realm of creativity has and always will remain solely within the hands of humanity. Machines can’t paint a Picasso, they can’t write Shakespeare, and they can’t sing “Hotel California” in the voice of David Bowie. Today, machines can do all of those things. And this has huge implications for pretty much everything.

Including BattleTech. What if instead of the many fabulous artists creating ‘Mechs, the cover of the next BattleTech sourcebook was drawn by an AI? Or what if the next MechWarrior had endlessly AI-generated dialogue? These are just two scenarios that might happen as a result of the AI revolution, and it may fundamentally change the way we experience the game we all love.

Artists, developers, and even writers like me are rightly concerned that an AI might soon be coming for their jobs. Rather than wait for a robot to tell me I’ve been made redundant, I decided to investigate the state of AI-generated media and see if I really should be worried, or if perhaps AI really isn’t the robot apocalypse we all fear.

To start us off, let’s take a look at AI-powered art generators. Visual art is and has always been a huge part of BattleTech, with artists like Dough Chaffee and Duane Loose providing us our first glimpses at the multi-story death machines called BattleMechs. With thousands of ‘Mechs and thousands more variants and configurations, there is a significant cost to any developer creating art of ‘Mechs let alone the beautiful full-page spreads that give source books that extra bit of flair.

Can A Robot Make Robot Art?

Stable Diffusion Mech 12

We have Shadow Hawks at home, courtesy of Stable Diffusion.

There’s an obvious economic incentive to have all those art pieces drawn by a machine. But in my investigation, I discovered that you can’t just ask an AI to recreate an image of a Timber Wolf and an Uziel locked in mortal combat. Getting even a single useful image out of an AI art generator–at least, for BattleTech‘s purposes–takes way more time and energy than it does to ask an artist to simply draw what you want.

Let’s take a Bad ‘Mechs article as an example. Each month, I go to Eldonious and ask him to create a new and amusing cover image that portrays these machines in their hilariously worst light. I give him a few guidelines, like what’s wrong with the ‘Mech being featured, and then he comes back with an amazing drawing each and every time. It’s a one-and-done transaction. 

“Without specialized training data, I don’t think you’re going to get it to easily reproduce a known/named ‘Mech.”

That is not how an AI works.

“Without specialized training data, I don’t think you’re going to get it to easily reproduce a known/named ‘Mech,” says Jack Monahan. He’s a BattleTech fan as well as the artist/designer on another stompy robot game called Brigador. Unlike many artists, Monahan doesn’t necessarily see AI as a job killer. He’s been toying with Midjourney, one of the many AI art generators available, and he was very helpful in explaining the technology’s limitations.

Before an AI can do anything, it first has to be trained. That means millions upon millions of images that are associated with certain words so that the AI knows what to produce when it receives a request. The internet is a great place for an AI to learn because there are countless images that have already been tagged and cataloged by various websites and search engines. 

But human language is complicated, and there are lots of ‘Mechs named after things that already exist. Give it a prompt for “Atlas” and it’ll think you want a map or a Greek god. Ask for a Timber Wolf and you’ll get a wild canine. Even throwing in words like “BattleTech” or “‘Mech” will produce some sort of robot, but not the specific ‘Mech you’re referring to.

Midjourney Atlas Attempt

Atlas prompt, courtesy of MidJourney.

As an example, here’s what happened when we asked Midjourney to create a “war photograph of Battletech Atlas mech, towering over smoke and haze in the battlefield.” Although certainly recognizable as a giant robot, it’s most definitely not an Atlas

In order to get an Atlas, you need to train your own AI. Monahan notes that fed a “specialized training set,” an AI like Stable Diffusion (a competing AI art generator) could reproduce specific ‘Mechs, “but that takes a lot of sifting, tagging images, and training on that set.” 

For just what kind of training Monahan is referring to, I reached out to David Vivas of the (formerly) Everything BattleTech Discord. He has trained his AI, the publicly available Automatic 1111, to create some ‘Mechs as images, but it still takes a lot of work from the user. 

“Since you are re-rolling and tweaking your prompts and weights, your outputs can vary greatly,” explains Vivas. He shows me an image of a Cicada and a Cougar that looks passible and admits it “wasn’t what I wanted originally, but it’s what I rolled.”

To get that single image, Vivas had to first train his AI on as much BattleTech art as possible, all tagged so that the AI understands what each ‘Mech is called. Then after it creates a seemingly generic output of giant robots at war, Vivas uses a process called “inpainting” to re-draw the ‘Mechs over and over again. Each time, they get ever so slightly closer to how an actual Cougar or Cicada looks.

“Striving to figure out and bring ‘new’—even if it’s just a novel recombination of elements that tickles our brain—is what the work is about. And I don’t think AI is capable of it.”

This process is both time and labor-intensive. A single image can take a day or longer, and it requires powerful hardware. Vivas trained his AI using two GeForce RTX 3090s and then had an RTX 4090 do the rendering. The cost of these computer components alone is well into the $10,000 range (at least in Canada), and that’s just the video processors. AIs are just as hungry for RAM, storage, and raw processing power–to the point where the average person really can’t afford to have an AI art generator of their own. 

Still, if you have tens of thousands of dollars to spend and can wait a few days for each image, you can have a machine endlessly recreate ‘Mechs. But there are other limitations too.

“I don’t think AI will by definition be able to do or say anything it hasn’t already been handed over,” Monahan explains. “As certain models become more convincing, they become less ‘creative’ [and more] oddly recombinative in a way that strikes my curiosity, as a visual artist. Midjourney a couple of iterations ago couldn’t draw a rifle to save your life. It was as if you had a seven-year-old draw a rifle, and then said rifle assumed realistic coloration and shading.

Midjourney Atlas Attempt 2

MidJourney’s second attempt at an Atlas. I’m getting more Kodiak vibes.

“The latest model doesn’t do that anymore, but it also doesn’t spark anything interesting, or far less often, it will draw rifle that is more neatly recombinative of existing rifle parts (ARs, AKs, FALS, etc) and mostly put them in the right place.”

In other words, AI can recreate what it knows and it can even take elements of what it knows and scramble them together, but it can’t truly create something new. Even those “Atlases” from earlier might appear new but they’re really just elements of millions of other giant robot pictures all jumbled together so it only looks that way. 

“And that’s okay,” adds Monahan. “Sometimes you don’t want ‘new’, hence us talking about using it to render pictures of an existing ‘Mech design we like! But as an artist, striving to figure out and bring ‘new’—even if it’s just a novel recombination of elements that tickles our brain—is what the work is about. And I don’t think AI is capable of it.”

What AI is truly good at is what computers have always been good at, which is doing the things that humans don’t want to do: boring, repetitive tasks. 

“Practically speaking, AI is great for doing scut work–little explorations that you might not want to bother doing, like the way art directors like you to do 10 thumbnails of a shotgun,” continues Monahan. “AI doesn’t complain about you pressing the run button as many times as you care to.”

“There is a great desire to cut artists, writers, actors, and other creatives out of the content production process.”

AI can also be good for helping its human users think of something new themselves. “In my experience, it sometimes is very practical and fun as a kind of running parallel to what I’m working on. As an artist, I think about and approach my work in a certain way, but telling an AI via a sentence or phrase is very different, and doesn’t fatigue me the same way that sketching does (though prompting does eventually become a chore as well),” Monahan says.

There is certainly a dark side to AI art generators. We’ve already seen examples of companies using AI in place of human artists, and while sometimes that’s to perform the drudgery of artistic endeavors, that’s still a job taken from a human and handed to a robot. 

“From the basic self-interested standpoint of being an illustrator, I have serious concerns, especially with how quickly this tech evolves and how it might sample art, which does make me worry about the future of artists and art careers,” says Alex Iglesias, art lead at MechWarrior Online and MechWarrior 5 developer PGI

Stable Diffusion Warhammer

A “Warhammer,” courtesy of Stable Diffusion. Note that it seems to look more like a Tau mech from Warhammer 40K.

“More immediately, I find that yeah, while capable of doing quite a bit, the tendency towards superficially averaging everything out without understanding why things are where they are supposed to be does hold it back from making deliberate logical design choices. Though this too might eventually be ‘solved’ by AI, in which case, oh boy, I’d be in trouble.”

Our own resident artist and freelancer Eldniousrex agrees. “There is a great desire to cut artists, writers, actors, and other creatives out of the content production process,” he says. “A way for the ‘idea’ people to merely push a button to have their ‘creations’ come to life. No paying anyone. No negotiations. No delays. No moody artsy folks. Just an endless stream of content that they can claim ownership of and monetize with no middlemen soaking up the profits.”

Bishop Steiner, one of the artists that helped redesign dozens of ‘Mechs for Catalyst Game Labs in recent years, is somewhat more fatalistic about AI art generators. “It’s Pandora’s box though,” he says. “It’s been opened, and there is no putting it back, and let’s face it, most people whose livelihood isn’t potentially impacted?  Probably don’t want to put it back anyhow. So now, all we can do is deal with it.”

How we deal with it is a question that hasn’t been answered, although even the people in charge of large AI models are desperately seeking guidance. Leaving aside the existential threat AI potentially poses (outcomes like The Matrix and Terminator are suddenly no longer theoretical), what happens to copyright if an AI scrapes thousands of images to create something “new”? Is that fair use or is it illegal? Should artists be compensated? If so, how much? These questions are still playing out in courts and legislative assemblies around the world.

“It’s been opened, and there is no putting it back. All we can do is deal with it.”

And we’ll have to get those answers soon. A few years ago, we couldn’t even imagine a machine doing what AI can do today. Now, it’s not hard to imagine what AI will be able to due in a few more years.

“Eventually, AI will create higher-order exports,” warns Monahan. I offered an example of the procedural generation we see in games like Borderlands, where thousands of parts can become billions of possible item iterations. AI would be perfect for creating those parts and then iterating them into a system just like the one used in Borderlands or No Man’s Sky

But I wasn’t thinking big enough. “As in, you won’t ask it to design all the guns for a Borderlands-type game,” Monahan adds. “You will ask it to output a Borderlands-type game.”

Ironically, robots have trouble recreating other robots with the exactitude necessary for what fans demand. You better believe that if you present a BattleTech fan with a Warhammer that has one too many missile tubes or an Atlas with too many fingers (as was often the case with some early AI models), you will get called out for it. That’s something Eldonious is acutely aware of and something that CGL is likely very sensitive to. 

Stable Diffusion Mech 7

A sad robot looks on from a foggy field, courtesy of Stable Diffusion.

That won’t be the case forever. It seems inevitable now that machines will overtake a human’s ability to endlessly iterate on existing designs. But Bishop offers a ray of hope for artists who think the end is nigh. 

“Fortunately for the actual artists, I think fans, clients, etc, will always exist who want that human touch. It’s why we see even in the age of digital art being the industry norm, a number of analog artists who provide something digital can’t: a one-of-one, original piece,” he says. “There will always be those who want actual art.”

A Thousand ‘Mechs On A Thousand Typewriters

While I think the largest concerns are for BattleTech‘s imagery, AI is also more than capable of creating entire books on the fly. However, language learning models suffer from the same limitations as image generators when it comes to details. For example, here’s what happened when I asked ChatGPT to create a 500-word story in the BattleTech universe. 

The dusty plains of Galatea Prime stretched out as far as the eye could see, dotted with the remnants of ancient war machines. It was a desolate landscape that bore witness to countless battles fought between rival factions. Among the wreckage, a lone MechWarrior named Kira “Viper” Morgan stood, her Warhammer, bearing the scars of previous conflicts.

Kira belonged to the Black Talons, a mercenary company known for their fierce combat prowess. They had been hired by House Kurita to secure an abandoned research facility deep within Galatea’s unforgiving Badlands. The mission was shrouded in secrecy, and whispers of long-lost technology had piqued Kira’s curiosity.

As the hot sun beat down on her cockpit, Kira received orders from her commander, Colonel Rourke. The Black Talons were to face off against the infamous Jade Falcons, a rival mercenary unit, who had also set their sights on the research facility. It was a race against time.

The battlefield crackled with tension as the two forces clashed. The thunderous roar of BattleMechs echoed across the plains. Kira piloted her Warhammer with precision, her fingers dancing across the control panels. She fired her PPCs, streaks of energy lancing out and impacting an enemy Timber Wolf, its armor blistering under the assault.

In the chaos of battle, Kira’s cockpit was suddenly filled with warning sirens. An enemy Hellbringer, with its twisted and sinister design, had locked onto her. A volley of missiles streaked toward her Warhammer, threatening to tear it apart.

Reacting swiftly, Kira engaged her jump jets, propelling her ‘Mech high into the air, evading the deadly barrage. As she descended, she unleashed a devastating salvo of lasers upon the Hellbringer, its armor melting away like wax. The enemy pilot ejected, surrendering to the inevitable.

Victory seemed within reach, but Kira knew the battle was far from over. She directed her ‘Mech toward the research facility, pushing her machine to its limits. The Black Talons were met with fierce resistance as the Jade Falcons fought tooth and nail to maintain their hold on the site.

Kira’s Warhammer weathered enemy fire, shrugging off blows that would have crippled lesser machines. She rallied her fellow mercenaries, coordinating their attacks with precision. Slowly but surely, they gained ground, forcing the Jade Falcons to retreat.

As the dust settled, the Black Talons stood victorious. The research facility was theirs, the spoils of war awaiting their discovery. Kira couldn’t help but feel a surge of excitement, knowing that the secrets hidden within those walls could shape the future of warfare.

She deactivated the Warhammer’s systems, the hum of the engine fading into silence. Kira took a moment to reflect, her gaze sweeping across the battlefield. The remnants of fallen ‘Mechs lay scattered, a testament to the price paid for victory.

In the BattleTech universe, conflict was a constant companion, and warriors like Kira embraced their roles as MechWarriors, driven by honor, profit, or the thrill of the fight. As she stepped out of her cockpit, the dust of Galatea Prime swirling around her, Kira knew that her journey was far from over. There were battles yet to be fought, and she was ready to face them head-on.

Stable Diffusion Mech 11

‘Mechs in a foggy jungle, courtesy of Stable Diffusion.

Just as AI struggles with visual imagery and BattleTech, ChatGPT has trouble with the details. Galatea is indeed a largely barren planet, but the colonized world is Galatea VI, not Galatea Prime. The Jade Falcons aren’t a mercenary unit, I wouldn’t necessarily call the Hellbringer a “twisted and sinister design,” and I don’t think anyone within the BattleTech universe would ever refer to it as part of their internal narration. Kira’s flying Warhammer is also an anomaly, but as a mercenary, we can assume she customized it to include jump jets.

Had I not given ChatGPT a constraint on word count, it probably could have done a little better with the grammar and formatting too. It gets points for writing ‘Mechs with the apostrophe every time, though. 

The fact that ChatGPT can deliver anything at all when given a vague set of instructions like “write a story” is impressive, but it once again shows the limitations of current AI software. Give it a wide scope and AI starts to lose the particulars. However, ChatGPT shines when you ask it specific questions. If you want it to bring up a record sheet for the Warhammer WHM-6R, for example, it’ll recreate that ‘Mech’s stats without fail (including its complete lack of jump jets). 

“I’m better than anything [ChatGPT] could put out.”

Besides accuracy, plagiarism is another problem with language models like ChatGPT, although it’s much easier to catch in text than in visual media. That said, the speed at which it can produce content far outstrips anything even the fastest typist in the world can achieve. 

The potential here then is less for ChatGPT to augment or replace fiction writers and instead to assist with some of the dryer aspects of BattleTech text, ie. sourcebooks and record sheets. But, like image generators, you’ll need to coax ChatGPT to provide the correct information and also correct it when it gets things wrong.

“They’re basically just plagiarism machines.”

It’s perhaps unsurprising then that BattleTech‘s writers aren’t nearly as concerned about the unfolding AI revolution as its artists. “I’m better than anything [ChatGPT] could put out,” boasts Bryan Young, fiction writer for Shadowrun, Star Wars, Doctor Who, and creator of BattleTech‘s Fox Patrol. “It can only rehash what came before, it can’t really create something new.” 

“They’re basically just plagiarism machines,” adds Russell Zimmerman, a writer featured in BattleTech’s Shrapnel magazine and most recently in BattleTech‘s first-ever Pride Anthology.  “All they do is just go grab random phrases and string ’em together with utter confidence; if I want a confidently wrong answer, I’ll look elsewhere on social media.”

Ownership is a thorny issue when it comes to AI. I’ve briefly touched on the legal issues surrounding AI art generators, but it’s no less problematic for language models.

Stable Diffusion Mech 6

A three-legged ultra-heavy ‘Mech, courtesy of Stable Diffusion.

Depending on how it all shakes out, AI could potentially shatter copyrights as we know them, which would have a profound effect on gaming communities using licensed IPs like BattleTech. Right now, at least, it doesn’t look like we have to worry about AI breaking our beloved universe of giant robots and galaxy-spanning nations. 

The future, however, is much harder to prognosticate. There are plenty of doomsday scenarios that could play out. We could one day find ourselves in a world where AIs generate everything and the entire creative industry becomes owned by IP holders that simply feed their properties into bots that endlessly spit out more and more content. There are also more charitable scenarios where AI is just a new tool that artists and creatives learn to exploit to create better content faster. 

I must admit, I’ve on occasion asked ChatGPT to do a thing and found it did that thing better than I could. The title of this article, in fact, was partially generated by ChatGPT. 

So, am I still worried? Yeah, a bit. All of this is moving far too fast for anyone NOT to be worried. But at least BattleTech seems safe. For now.

And as always, MechWarriors: Stay Syrupy.

Initial Robot Request For Hunchback Pancakes 2

Share this:
This entry was posted in Art, Community, News, Software on by .

About Sean

Hooked on BattleTech at an early age, Sean honestly can't remember whether it was the cartoon, the serial novels or the short-lived TCG that did him in. Whatever it was, his passion for giant shooty robots never died, so now he writes about the latest and greatest in 'Mech related news.

22 thoughts on “The AI Conundrum: Are BattleTech’s Human Creators on the Verge of Extinction?

  1. Chris

    AI can help save the world, and at the same time has the potential to doom it.

    Interesting times.

    I just look forward to all those indi games getting cool AI manufactured side quests and voice over! What a game changer.

  2. Drake

    The pictures and stories are very interesting. My only problem is that I’m nerdy enough to know that the Warhammer 4L and the Warhammer 7CS both have jump jet as standard.

  3. Hyper129

    POV: You’re Chat GPT and you make an article title about how AI could be bad:

    Oh mah gawd… what am I doin wit my life?

  4. Vassily

    Curious the article coincides in time and form with a quarrel in a chat of the Spanish-speaking community, on the publication of images generated by AI, in the end it ended with the expulsion of a member of the community that abused and saturated the chat with images created with AI.
    I think that drawing AIs are a wonderful tool for people with no drawing skills, I myself am a frustrated drawer who, due to lack of time and above all talent, has not been able to develop this skill. If I had had a tool like this when I was a child, I would now have spectacular role and module manuals.
    I also believe that the artist will not be replaced by a machine, no matter how good a job it does.
    In the case of drawings, transferring the idea you have in your head to paper is reserved for a few (Kim Jung Gi, Don Bluht). Writing a promtp and having the result look like the head is a step forward, but I don’t think it replaces the human creator.
    There is something about handcrafted works that no matter how capable machines are of replicating them, they lack that spark of art that awakens in each of us.

  5. Samuel Crosbie

    Simple solution. Don’t allow AI creations that refer to itself as art. It is NOT artwork. It’s taking something already created and adapting an image to fit a set of programmed parameters. It is NOT creation. Not yet. And that’s artwork. Art is created. This AI shite is NOT.

    1. Flashfreeze

      I think the term “AI generation” gives people the idea that the output is somehow innovative or new.

      If you can get the phrase “AI component trace” to stick, I’m sure the tone will shift.

  6. John K

    Did you ask Stable Diffusion to recreate your logo, and then use that for this article? Because that has me giggling all over my house at a wannabe Grand Titan trying to make pancakes.

  7. WoK

    I’m going to just drop the (War)hammer:
    AIs do not exist. And if they did, they would be of absolute no use.

    To mankind, the idea of constructing an intelligent machine is nothing new.
    About a century ago, a devilish experiment was suggested: Put a human in a room and let him talk indirectly (via telegraph f.e.) to an unknown entity.
    If this entity were a computer, and the human could not find out about it even after an unlimited time, this would be truly a intelligent machine.

    To shatter those thoughts, i’d just ask Sean’s “AI” his opinion about the new Battletech Universe, the one with the 3-headed squirrels. It would be obvious very quickly that this is not a intelligent machine, not an “AI”.

    A new attempt was made about a generation later by Isaac Asimov, who published his three “robot laws”. Those were the most important rules for any robot (computer with legs) which he had to obey always and without question.

    Asimov’s fans readily came up with a desperately needed 4th law however: The robot must be forbidden to program itself – or it would simply erase the first three rules.

    Here is the problem: If we restrict a computer in the aforementioned way, he is bound by limits we imprint on it. But then we can not hope to get something really new from it, no art, no science, no literature. This is the current state of the “AI”: There is no intelligence, just overblown databases on steroids.

    And if we don’t restrict the computer? Then we would have an AI, but an anarchic one. It would not be bound by any rules, ethics or laws, because it would not have any guidelines about what was “good” or “bad”. One word: Skynet.

    But there is not a single computer in the entire world today (unfortunately i am not well informed about secret military projects) working unrestricted.
    The only time i ever saw computers program themselves is what most people know as the “blue screen of death” in Windows OS.

    OK folks, this text is too long already although i barely scratched the surface.
    So holler if you want me to explain anything more deeply. I’d be honored.

  8. Patrick Rich

    I agree that AI is concerning in terms of obsoleting human creators. I can see it’s merits where it would be useful for putting out quick compositions and conceptual designs, but that’s not at all how “AI artists” are using the technology, as they are using it to produce final pieces of art.

  9. Rune Wolf

    “Long-time listener, first-time caller”, first I’ll just say that starting as a guy who bought the box a year-and-a-half ago (the REAL starter box, Game of Armored Combat, not the cheap one with just 2 mechs) Sarna has been VERY helpful in learning what this setting is about. Thank you.

    The AI-generated mechs here look like something out of Supreme Commander 2. I wouldn’t be surprised if in the near-future, a new PC game featured mechs that look similar. You can guess the reason why.

    The AI-generated short story here. It reads exactly like some fan-fic where their elite mercenary company and totally awesome original character, crush the biggest badass unit they can name, (The Jade Falcons!) the most powerful mechs they’ve heard of (Timber Wolf! Hellbringer!) “melting like wax”. On a human being, power fantasy is… understandable. I know why someone would write something like that. There’s no reason for an AI to do it. It doesn’t experience any emotions. It has to have been exposed to power fantasy writing.

    To answer your question, why a Hellbringer is sinister and disturbing, I think it’s keyword-matching in the dark. It drags out a mech with “hell” in the name, it knows from the internet that words like “twisted” and “disturbing” often accompany the word “hell” when used in a fiction context. You can imagine what kind of fiction it was trained on to make the word associations.

  10. R0AHN

    Throwing my two cents in here since I’ve had to research this in my own creative stuff. I am not a lawyer but two channels that have done great breakdowns on AI and copyright I can recommend are Lawful Masses with Leonard French (who is primarily a copyright attorney) and Legal Eagle. You touched on this all a bit but just wanted to elaborate further.

    A couple of big points to keep in mind (bear in mind I’m focusing entirely on US copyright law):
    – AI cannot own copyrights because ultimately it’s not human. It’s established law here that human authorship is a requirement for registration. AI-generated images and other content have no associated copyright and can be redistributed freely. There is a sliding scale between fully human authorship and fully generated where it can be used as a tool rather than a source of the full content (for a modern example, think of Adobe’s content-aware fill that you can use to selectively “erase” something from an image). The courts haven’t tested where the big dividing lines on this scale lie just yet but will sooner or later.
    – AI image generation is not doing the same thing humans have been for millennia; they aren’t human. They break down an image into complex number arrays (vectors) pixel by pixel. Stable Diffusion doesn’t know what a cat is. It doesn’t know what a Mad Cat is or how it’s different from a disagreeable domestic feline. Ultimately this is why the examples you put up really don’t match the input; it’s putting together images based on what its training data has tagged for Atlas, mechs, etc.
    – Language models are, at their core, just a more complex auto-complete. Computerphile (out of University of Nottingham) also did a good breakdown on how it works. It looks at keywords and generates the most likely words that follow based on the last n words in the sentence so far.. It doesn’t “know” anything and is why you get such gems as the history of bears in NASA being told by ChatGPT-alikes as if they were solid fact.

    Currently there is a big push with Hollywood and game dev studios to try and replace creatives (who C-suite execs view as just dead weight rather than the core contributors they are) with AI-generated content for exactly the reasons you highlighted. “It’s Free(tm)!” But it really isn’t. It’s absolutely average and incapable of true creativity. These generative tools can be used productively as tools, such as for rapid concepting and getting the creative juices flowing, but can’t at this time fully replace people.

    Once they smarten up and realize they can’t just press a button and get a script or movie *they actually own and can monetize* I think we’ll see this die down. It will be an amazing day when and if some new movie or game or whatever that has primarily AI-generated content is pirated and in the inevitable copyright suits, the courts have to rule it as legal for free redistribution. We’ll only hit that point if there’s truly stubborn stupidity in charge of these companies (in other words, almost guaranteed).

    1. Craig

      Is there a functional difference between a search-based text producer, and a word processor?

      Would the person who wrote the software, in your case, be the content creator, in combination with the chap GPT user and their parameters?

      I regard all copyright laws as BS and in serious jeopardy anyway for the simple reason that all the revenue is eaten up by the “Gravy Train”, Randy Quaid’s “Star Whackers”, or whatever you want to call the vast paper-pushing army of publicists, agents, financial advisors, etc. who have long ago latched onto the actors and the unions in Hollyweird. Look how many movies Lucas and Spielberg film in London, Morocco, and basically anywhere else except where the studios live.

      There’s been a war on writers for years now, because they are the creative talent that’s most important. I love that scene with the communists in the movie “Hail Caesar” with the writers – because the same thing is happening today. They really were being screwed over back then, just as now!

      Disney will buy Marvel with all its IP, so they don’t have to share with the creatives anymore, or depend on them at all.

      I believe that after the courts have to deal with the first thousand or so such cases, that it all will be swept away by public pressure on the lawmakers. Giant logjams created by river floods won’t jam up the courts forever. History shows us that those logjams get unjammed eventually by giant storms out to sea.

      Witness what’s happened to the comedy world – the agents and the filler and the advisors all lost their power as comedians directly connect with club owners, fans, and often film their own specials on youtube posted freely (Ari Shaffir example I believe he spent 50 grand or 70 grand on that).

      Gone are the days when Richard Jeni committed suicide after running into wall after wall of the same agents, the same development deals, the same spiel from each of the four major tv networks. Back then comedians were always locked into the sitcom pipeline. Now he would have had his own website, his own touring company, his own “people.”

      See Adam Carolla and his frequent comedian guests as they promote their tours and specials etc. on other comedians’ shows!

      1. R0AHN

        There’s a significant functional difference. A word processor (Microsoft Word, LibreOffice, OpenOffice, Notepad++, etc.) is the digital equivalent of a pen and paper. It lets you directly put your own words into written form. A generative tool lets a heuristic algorithm (key note: not you) put words into written form based on a given prompt. In one you have complete creative control and the other you have

        In the eyes of the law, it doesn’t matter who wrote ChatGPT (or other generative tools) just as Microsoft doesn’t get any creative credit for writing Word when you type up a manuscript. One way to think about it is if you trained a gorilla to paint and to type on a typewriter. It can paint and type. You trained the gorilla and can nudge the gorilla this way and that to influence what it makes but when it comes to putting paint on canvas or words on paper, you didn’t put them there – the gorilla did. It is impressive that you trained a gorilla to do this but ultimately what they produce is not your work. Or you’re a student and a teacher hands you a writing prompt – the teacher doesn’t get any credit for the work you did.

        With how dysfunctional the Fed is right now, it would be a small miracle for any copyright reform that addresses this actually gets through both chambers. On the courts it’s going to take several (doesn’t even need to be that many, really) precedential rulings before we have a solid framework for where the major dividing lines between copyrightable and non-copyrightable stands.

        The US Copyright Office already has full say over whether something can be copyrighted since that’s their job. They already auto-reject anything that is AI-generated and have rescinded copyrights for those works that were submitted and granted but then came to light later that they weren’t human-authored (such as Zarya of the Dawn, which was AI-generated).

        For a more complete, full breakdown I’d recommend going through those legal YouTube channels I mentioned. The US Copyright Office also has a guide that goes over basics of copyright, ownership, and registration here:

  11. Craig

    If content creators fear the rise of what-we-call-A-I, then they should realize two few basic realities:

    1) Stagnant and Formulaic Output
    As original content creators get chased out of a giant interwoven complex universe such as BT, they are replaced by midwit-pleasing “safe” parodies which more closely comply with “the current year” absurdity that we’re all stapled to the bench to watch drift through pop culture right now. The cold hard fact is that “AI” formulas aren’t any different than the soulless, half-brainless message-pimping garbage which passes for science fiction in general nowadays. Take a good look at the Science Fiction Awards of any stripe over the past 15-20 years, or read any random book off the shelf – you know, the far future where characters are stuck in the early 21st century forced narratives about race and gender. It’s… hard to miss.

    2) Piecing A Universe Together Takes Time
    I came to the fiction of BT first through the Mechwarrior PC game, then the “Crescent Hawks Revenge” PC game, then the boxed set – which was good timing as I moved to a town big enough that the few geeky people I could find actually played something other than Dungeons & Dragons. When looking at the past from the 2020s, it looks like a huge avalanche of fiction – but in fact the pieces came out a few here and a few there. When I found the “House” books at a hobby shop I couldn’t believe how rich and filled-out the universe was at that point – and we are talking 3 decades ago.

    For those content creators who fear what-we-call-A-I, the answer is simple – create something non-formulaic, not pleasing to the current-year crowd (as in no pandering to what doesn’t even exist), and simply new fresh and exciting.

    You know, what fiction used to be?

  12. Kevin Killiany

    This is not a new discussion. I just wasted 20 minutes trying to track down a short story in (I think) Asimov’s or Analog half a century ago wherein fiction writers did nothing but make adjective decisions on “computer generated” manuscripts with plots and characters determined by real-time data on the tropes and genres that were selling. Nearly a century ago the Stratemeyer Syndicate paid writers $125 to write novels that strictly followed a set formula. There was a lot more freedom for details and dialog, etc., but generally speaking the Nancy Drew, Hardy Boys, Bobbsey Twins, Tom Swift, and a couple of other series I forget were rigidly an formulaically structure. And sold millions of copies because such familiar, structured stories are comfort food for a lot of readers. Though there is a lot more variation and, yes, creativity now that big publishers don’t gatekeep publication, romance novels, noir mysteries, westerns, etc., all include tropes and set pieces that fans of each genre expect and look forward to. Totally original fiction is something completely different — and even though some advise that the best way to beat AI is to be completely original, I don’t really think the BattleTech fanbase is looking for something like “A Girl is a Body of Water.” We want stompy avatars of death and conflict, conflict, conflict. Political, personal, emotional, internal, something has to confront something, and the resolution has to be satisfying — whether victory, defeat, compromise, or (my favorite) pyrrhic — even if all that’s won is growth and wisdom. What writers such as myself — or better yet, writers with real talent and skills — can bring to these stories is three-dimensional, authentic characters: People. Because every story is at its core about charcters the reader cares about (whether loathe, mourn, envy, pity, be inspired by, or identify with). That, I think, is what a living writer brings to the storytelling table.

  13. Steel Shanks

    When the AI can draw hands, and not odd seven digit claws maybe I’ll worry… Or when it can express emotion in artwork… Or show brush strokes in painting, or rough sketches etc. As an artist I am not concerned. I can draw and paint a Hunchback-4N, AI can’t… I can express emotion, and draw hands… AI can’t. By the time AI CAN do these things, We’ll be too busy fighting its Robot Army to care…

  14. Itchy

    Ha! That is a stack of pancakes, but not a hunchback! (Plus, I don’t see a chef’s hat?!)
    Also, interesting write up.

  15. Gundam Chief

    I’ve been writing and drawing for the better half of two decades at this point. I’ve been using Midjourney and Fotor for art, and ChatGPT for writing. To see what they can and cannot do, for both my own devices and concerns.

    As of this point of time, I’m not too concerned. The art AI are only as good as the prompts you give them, and even then, the result may not be close to what you have in mind. The same results come from the use of ChatGPT when it comes to writing. I’ve found it can write, but, like the article has said, it loses the details. I have also found as a result of its programming, that it tries to spin things in a positive light almost all the time. It couldn’t do dark fantasy like ‘Elric to save its life.

    Once I learned what I found to be its flaws and limitations, I was satisfied that the industry can survive for at least awhile for now, especially once laws come into being to handle issues, like Copyright, which has already been done just recently iirc.

    What I have used the Art AIs and GPT for these days, which I find them excellent for, is brainstorming.

    With the art, it’s fairly obvious how images can inspire. Once you learn how to use prompts correctly to get nearly what you want, you will go further and build from that yourself. For writing, GPT is good at presenting data on a subject that you want to find yourself, but don’t have the time or the means to find. It also can be used to bounce ideas off of in order to build up on something you have already made, and do it all very quickly.

    For example, I started with the idea of a species of beings similar to the Warforged in Eberron, only supernatural beings made by God like entities. When I started brainstorming, GPT and I created an entire background for this race of beings within a day, and it was completely unique from what it started as, while still being inspired by it. Something that can take weeks to do normally, in my experience.

    So, I see them as more of a useful tool than a threat. While there is a danger of AI taking jobs away, not just as writers and artists, but editors and data gathering as well, they cannot replace humans completely due to that ‘spark that’s needed for true creativity. Not yet anyways. Give that a decade or two.


Leave a Reply

Your email address will not be published.