Answering Gaming Questions with AI – Are You Just Stealing Stuff?

good artists copy; great artists steal

-Pablo Picasso, who very well may have stolen that quote

This week I wanted to do something a little different and maybe see if I could find something that AIs ought to be good at.  I’ve spent a few weeks throwing questions, some of which I cannot answer, at the three AIs I have chosen just to see how they handle them.  Results have varied, but a good number of the responses would at least get a few upvotes on Reddit.

AIs thinking about video game questions… drawn by an AI

For those interested, these are the past posts:

• Define a Sandbox MMORPG
• A Tank in EVE Online
• Finding a Warm Ocean in Minecraft
• Why So Many Fantasy MMORPGs?
• MMOs vs MMORPGs

So my idea for this week was to lean into what I thought should be an AI strength, generating lists of names or ideas or concepts or whatever.  No deep thinking or analysis, just give me some things to, say, name some NPCs or make up some new crafting materials for me.  This is where being trained on a large corpus of data should allow the AIs to do well at.

I started with ChatGPT and asked the following:

Create a list of unique names for a warrior in a fantasy setting

And it gave me a list of 30 single word names, about a third of which I recognized from other sources.  “Stormbringer” has a bunch of past usage, including being the name of a Michael Moorcock novel.  Likewise things like “Thunderstrike,” “Runewalker,” or “Nightshade” hardly seemed unique.

Well, maybe I needed to tune up the question.  The joke is that creatives are safe from AI because AI requires the end user to ask for exactly what they want, and anybody who has done project work with people knows how infrequently that happens and how inarticulate customers can be about their real needs.

So I first added a gender to the request:

Create a list of unique names for a female warrior queen in a fantasy setting

That got me 30 responses that were even more heavily borrowed from literature.  Uniqueness didn’t seem to enter into it.  (And I am not going to clutter up this posts with most of these lists, you can go generate your own.)

I changed up again, going for something even more specific:

Create a list of unique names for a dwarf in the arduin grimoire universe

This gave me a decent list of dwarven names, though I didn’t go pull any of my Dave Hargrave rules or modules off the shelf to go check how many were borrowed.  And uniqueness didn’t seem to enter too much into it.  The last names were all of the classic “material+object” format, so gave up “Granitebeard” and “Ironaxe” and the like.

I tried a few more, but I was already a bit torn on how useful ChatGPT was really being here, as it just seemed to be regurgitating tropes of the genre.

That, in turn, brought me into current events.  Right now in the US the Writers Guild of America (WGA) is on strike.  They represent the writers who work on most of the movies and TV shows that come from the US, and one of the bones of contention is AIs like ChatGPT.

Studio execs, being more of the Wall Street business ilk, have been hungrily eyeing the advances in AI because they would like to have content auto-generated to feed the ceaseless demand from the public for something new to watch.  And, of course, they would like to spend less money doing it because, if burning down an orphanage would increase shareholder value by 0.25% they would at least argue that it was their fiduciary responsibility to grab some torches and hand wave about how they were helping encourage the orphans to find their path in life.

In case you have missed it, I am clearly not a fan of Wall Street obsession that everything must be sacrificed for immediate shareholder value.  People were complaining about the lack of long term vision of US companies back in the 80s and 90s, but compared to how things are now that time seems like an era of farseeing stewards of the economy.  But I digress.

So I support the WGA and their strike and am glad that their sibling unions like SAG are on their side.

But I am not 100% behind their AI stance, which has become “AI is theft,” at least from some of their membership.

This is not an uncommon opinion in the push back against AI, the idea that training AI with copyrighted material is unethical and should be unlawful, with the conclusion that it is the equivalent of theft.  The theory is that, while AI pretty much sucks now, at some point it may be good enough to convincingly produce artwork that is indistinguishable from that of the original artist that produced the content with which the AI was trained.

And, at that point, it will be too late.

The thing that always jumps into my head at that point is that this is the path that a human artists also takes to learn.

You start out looking at what other people do, experiment with imitation, and, if you are lucky, you find your own style/voice in which to express yourself that makes the art uniquely yours.  And 99% of those who aspire to the arts probably stop progressing somewhere during the imitation phase, somewhere between being unable to achieve even a passable imitation and being able to imitate another’s style in a different context.

With the way AI is currently being approached… which seems to be something along the lines of “if we throw enough data at it maybe it will be good or become sentient or something” … it is barely making it to the imitation phase, and even then it requires human input to set a context in which it might work.

Mark Twain in the style of Roy Lichtenstein

All of the best, or most interesting, bits of AI art are derived from the human input required to guide the AI towards some outcome.  And decent outcomes are pretty rare.

At this point you might be asking if I then support the idea of AI being used to run roughshod over artists, letting it appropriate their work.  No, I certainly do not support that.  I think it is incumbent on the humans guiding the AI to ensure that that does not happen or to compensate the artist if it does.  We have a whole legal mechanism already in place for that.

Granted, that mechanism has its own issues.  Did George Harrison really copy from He’s So Fine when he wrote My Sweet Lord?  Did that nut who wrote a series of orphan boy goes to wizard school stories before Harry Potter really have grounds to sue J.K. Rowling.  Is Digimon just a blatant rip-off of Pokemon?  We already lack clear guidelines for how much one is allowed to be influenced by other works… and any famous artist who says they had no such influences is a liar… so drawing a hard line that AIs may not copy or be influenced seems naive at best.

We are worried about AI doing what humans do all the time already.  And I get the concern that AIs can do this sort of thing at scale.  But the idea that somebody is going to create an AI and not train it on the works of humans is a non-starter as well.  It doesn’t work for humans… everybody I know repeats things they’ve simply heard all the time, and original thought is so rare… well, one of my college professors once yelled at our class that the chances of even one of us having an original thought in our lives was infinitesimal.  That certainly applies to me as I have spent more than 16 years mostly regurgitating things I’ve read or making simple, first level connections between established ideas.

So if you use AI for things, I feel it is on you to make sure it isn’t infringing in any substantial way.  And if an AI is infringing on your stuff, then I suppose you respond the way you would if a human did it.

All of which is kind of a ways from where I was headed when I started this.  Flow of consciousness posts can be like that.  So let me return home, back to the idea of AIs straight up stealing stuff.

One of the attempts to get to some useful ideation out of ChatGPT had me submit this prompt:

Generate a list of futuristic sounding fictional metals used in spaceship construction

I want to make a spaceship building game or something and I need some ideas.  So ChatGPT came back with:

1. Neutronium Steel: An ultra-dense, highly durable alloy made by fusing neutron star material with terrestrial metals, offering unparalleled resilience against extreme forces and temperatures.

2. Plasmanite: A lightweight, semi-transparent metal capable of manipulating plasma fields to enhance the efficiency of propulsion systems and generate advanced energy shields.

3. Flexilum: A highly adaptive, morphing metal that can self-repair and rearrange its molecular structure in real-time, allowing for customizable and rapidly changing spaceship designs.

4. Gravitite: A metal that actively manipulates gravitational fields, enabling advanced propulsion and artificial gravity generation, as well as improved maneuverability in deep space.

5. Starforged Tungtherium: An alloy crafted from tungsten and exotic, interstellar materials, resulting in a heat-resistant, incredibly strong metal able to withstand the rigors of FTL travel.

6. Cryonium: A metal with innate superconductive properties at room temperature, enabling high-efficiency power transmission and advanced energy storage for spacecraft systems.

7. Nanochrome: A self-assembling nanomaterial metal that can form intricate patterns and structures, providing dynamic and efficient weight distribution and hull reinforcement.

8. Spectralite: A metal capable of changing its properties according to the electromagnetic spectrum, allowing for advanced stealth capabilities and adaptive radiation shielding.

9. Etherium Crystal Alloy: An interdimensional metal that manipulates the fabric of spacetime, reducing the effective mass of a spaceship and enabling advanced FTL propulsion methods.

10. Quantum Resonance Mesh: A metal formed from quantum-entangled particles, granting improved communication and data transfer rates across all spaceship systems, as well as enhanced sensor capabilities.

And I was pretty happy with that.  I mean, some of that needs to be work shopped into something a little less ponderous, but some decent work.

Then I went and used Google to see what the AI had cribbed from the internet and found that “Tungtherium” was unique, but everything else had at least some precedent while a few of them were brand names of actual products, or close enough to such brand names that Google felt it ought to bring them up at the top of the first page of results.

Now, to be fair, none of them had the properties described, and I am sure that is a whole additional field of research to work on, to see if those concepts themselves were stolen.  So maybe useful as a starting spot for making up your own metals, but not something you could safely use straight out of the box.

Grade: B-

By that time I was ready to move on to Google Bard, which completely misinterpreted my intent and just rolled up a list of metals used in other stories:

• Duranium is a fictional metal that is often used in starship construction in the Star Trek universe. It is said to be incredibly strong and durable, and it is also resistant to radiation. Duranium is often used in the hulls of starships, as well as in their weapons and other components.
• Dureum is another fictional metal that is often used in starship construction. It is said to be even stronger than duranium, and it is also said to be lighter. Dureum is often used in the hulls of large starships, as well as in their engines and other critical components.
• Dust is a fictional material that is found in the world of Remnant in the popular web series RWBY. Dust is a crystalline energy source that can be used to create a variety of effects, such as fire, ice, and wind. Dust is often used in weapons and other technology, and it is also used to power vehicles, such as airships.
• Inerton is a fictional metal that is used in the construction of airships in the novel The Windup Girl by Paolo Bacigalupi. Inerton is said to be incredibly strong and lightweight, and it is also said to be resistant to gravity. Inerton is used to create the hulls of airships, as well as their engines and other components.
• Kryptonite is a fictional mineral that is found on the planet Krypton. Kryptonite is said to be poisonous to Kryptonians, and it can also have a variety of other effects on them, such as weakening their powers and even killing them. Kryptonite is often used as a weapon against Kryptonians, and it is also used to create technology that can control them.
• Orichalcum is a fictional metal that is often mentioned in Greek mythology. It is said to be incredibly strong and durable, and it is also said to be impenetrable by weapons. Orichalcum is often used to create weapons and armor, and it is also used to build ships and other structures.
• Plasteel is a fictional metal that is often used in starship construction in the Star Wars universe. It is said to be incredibly strong and durable, and it is also said to be resistant to blaster fire. Plasteel is often used in the hulls of starships, as well as in their weapons and other components.
• Radium is a fictional metal that is often used in power sources in the Fallout universe. Radium is said to be incredibly powerful, and it is also said to be radioactive. Radium is often used in power armor, as well as in other technology, and it can also be used to create weapons.
• Unobtainium is a fictional metal that is often used in science fiction. It is said to be incredibly strong and durable, and it is also said to be incredibly rare. Unobtainium is often used in the construction of starships and other advanced technology.
• Vibranium is a fictional metal that is often used in the Marvel Comics universe. It is said to be incredibly strong and durable, and it is also said to be able to absorb vibrations. Vibranium is often used to create weapons and armor, and it is also used to build vehicles, such as the Black Panther’s suit.

Google Bard has become the comedy relief of these posts, though I did get a note that it was getting a significant upgrade soon.  We’ll see how that plays out.  But for now…

Grade: INC

Meanwhile, BingAI… look, I’m going in the revers of my usual order… seemed to grasp what I wanted, though wasn’t going to generate too many options:

• Nanosteel: A metal alloy composed of iron, carbon and nanobots that can self-repair and adapt to different environments.
• Plasmite: A metal that can store and release plasma energy, making it ideal for weapons and propulsion systems.
• Quantium: A metal that can manipulate quantum fields and create wormholes, allowing for faster-than-light travel and teleportation.
• Xenonite: A metal that is resistant to extreme temperatures and radiation, making it suitable for exploring hostile planets and stars.
• Zyrium: A metal that can change its shape and properties according to the user’s will, enabling versatile and customizable designs.

 

BingAI, like ChatGPT, pulled things it had read elsewhere to compile its list, adding some fancy attributes to the list.  It just made a shorter list and wasn’t as expansive on its options.

Grade: C-

Basically, to answer my question, all three basically stole their ideas.  Only ChatGPT came up with a new word, and even that was a mash up of other words.

But that is all you can get from AIs at this point.  We have given them a bunch of LEGOs and they can only assemble them in ways they have seen before… and only with human intervention.  An AI simply can’t assemble something in a way it hasn’t come across in its training because the whole thrust of the training is just to give the AI the maximum number of examples.  It cannot do anything new, and it certainly cannot conceive of a LEGO brick outside of the set it has been given, something an 8 year old can probably manage.

So it is all theft.  But so are most human efforts.  We just draw a line between what is allowable for AIs and what is not for humans I suppose.