Larian Studios Forums General Chat AI controversy in games

First of all my stance on AI. Whilst I do use AI as a tool and to generate art for my own personal use the best to frame and put on my walls I do not agree with it replacing human artists. Nor should people sell AI art; theres loads on Deviantart.com and Etsy which I think is wrong and takes money away from real artists.

You can see why. I created the Patriarch from the divinity games sat on a throne. Honestly it's very good. Link to my deviant art as I cant link directly to images.

I would be greatful for anyone's opinions.

https://www.deviantart.com/adriansmith31

I'm a little confused. You don't like AI replacing artists, but you use AI to replace artists?

There are a ton of fantastic artist selling REAL art (both prints and originals) for very reasonable prices. You'll get much better and more meaningful art to frame and hang on your walls this way, AND you'll be supporting real artists instead of pushing them out of the market. And in my experience, the artists *really* appreciate it and will give you personal notes about the piece, etc., especially if you're a repeat buyer.

I personally don't care about companies using AI for placeholder assets or whatever; at present, nothing that Larian has revealed about their development processes so far has me worried that their games' quality will suffer or that anyone's job will be replaced.

But stuff like what you're describing really bums me out.

Dragon looks cool bro. Keep it up.

Yeah it seems as if that came across as a bit hypocritical but I thought I'd share what I created on Deviantart.com even though I don't really agree with it. But that art isn't for sale, its freely downloadable if anyone wants that sort of thing. But they have a choice on deviantart to see AI content ir not with the site filters.

Like you say I dont care if Larian use AI if it helps them during production as long as it doesn't make it into the final game.

Remember, humans can produce slop just as much as AI can. Just look at any of EA's latest releases. Dragon Age cough cough.

Edit: I used AI to create some new pictures for my house. Not finding what I wanted I made my own. I will take your comments onboard though and look through deviantart to find something I like to buy.

Direct link to image for those who don't want to go looking for it. The image was edited manually, this isn't the raw AI generated version. https://www.deviantart.com/adriansm...s-throne-from-Divine-Divinity-1274853183

Thanks

I just see it as another tool, ultimately. It's a shame there isn't more nuance involved in the typical discussion surrounding it.

On the one hand, I fully agree that a lot of AI is outright slop. I used to enjoy searching Google for fantasy art for inspiration when creating and writing characters. It's now increasingly difficult to find drawn art that isn't made by AI and some are very obviously full of uncanny mistakes, such as weird faces and hands.

At the same time, many other industries have been impacted negatively by the rise of technology. Artists are not unique on that front, yet many demand that AI is not used at all whilst demanding boycotts...even as I imagine they quite readily and gleefully use technology that has negatively impacted other professions. There's also a monetary bias to it, in that because artists stand to make money through commissions they are liable to speak out against AI purely on that basis.

I would, perhaps have more sympathy if a lot of the same artists lashing out at AI were not also prone to defiling established aesthetics in various settings, such as demanding that cool fantasy races are hijacked and redesigned to resemble the same demographics as modern day big cities.

AI is a tool. As far as I heard, Larian knows this.

Other companies are trying to replace human work with AI "work" and its failing for obvious reasons. AI is just a machine repeating patterns it has seen before. It cannot actually think in any way.

As short as I cand add to this (even if I'm half a year late to the party), is that I personally do not see what we use today as actual "AI", since all I see is some code that alters itself and needs a huge amount of memory to store all the possible logical paths it sees. Kind of a decision tree of decision trees of... more decision trees?
I've more been following the neurological network-based AI development, area with actual potential to create a digitalized intelligence akin to ours, but the catch there is that we do not yet have the ability to even handle the amount of information that would be needed to just digitalize the neurons of a single brain. We're talking I think some billions of neurons, *each* of them having some 10k possible connections, a weird kind of permanence, with the way memory works, the whole shebang would be kind of impossible to simulate even if given the entire processing power of all hardware on Earth at present. We're just not there yet in terms of hardware.
It gets even funnier when we ended up calling computer-driven NPCs or players in strategy games "AIs", when they were not much more than a set of algorithms that worked against each other. Sure, the algorithms got much more complex and we gave them the ability to change themselves and each other, but that is still not actually intelligent...

Which is reasonable since it is objectively not "Artificial Intelligence"

It's literally just a bot that runs pre-scripted code.

That's all current "AI" is, in its various forms.

It's only being called "AI" because it tricks people (Notably, investors) into thinking it's actual AI, but it's not.

This isn't bad in of itself (Outside the economic concerns of this investor hype, with it massively impacting prices of computer components and causing lots of Data Centres to get built which suck up all the water and use tons of electricity)

The main controversy surrounding "AI" comes from the way that the current fad iterations function. Which is by training a dataset to regurgitate to users.

For LLMs this means that it will often confidently state wrong information making it a terrible source of data. With the worst aspect being its programmed "Agreableness" where it reinforces particularly dangerous ideas (Like pushing someone towards terrorism or self-deletion - Things that have literally happened)

For Gen-AI this means it goes around and steals tons of real artists work in order to copy/paste it into the stuff it "Produces". Thus not only is its usage tied directly to replacing artists, it's also outright plagiarising them to do so.

And this exactly is what I keep trying to tell anyone that listens. It's not the technology evolving, as it never is. It's the horribly delayed and long-term lack of regulation. IMHO, technological progress should be treated something like cyber security. As soon as some new vulnerability is identified, it's a race to fix it up before too many bad actors exploit it. Replace "vulnerability" with "technology", "fix it up" with "regulate it" and that is the principle I would personally *love* to become internationally-spaning law.

What the layman calls "artificial intelligence" is called "machine learning" by the expert.

The later term much, much better describes what we have.

---

We simply dont actually know what intelligence is.

Thus we cannot create an artificial version of it.

What we would want is create artificially the genius of our actual grand thinkers. Like, say, Albert Einstein.

What we have is a logical machine that can repeat patterns it has seen before.

---

About the brain, we dont fully understand how it works.

What neural nets emulate is a trivialistic version of the brain, as it was understood decades ago.

---

Either way no computer has ever actually experienced even the most simple of sensations, pain. Something even the most simple biological brains can experience; in fact we see pain reactions in biological systems without a brain, like plants and unicellular organisms. There is something fundamentally different about living beings compared to machines, and I doubt any increase in computing power will ever overcome this. A machine is simply not a being.

We do. It's cognitive ability.

Nominally, it's the ability to interpret and solve problems.

It also comes with a level of self-awareness and the ability to utilise logic rather than instinct.

That we know what intelligence is, is how we are able to classify various animals as intelligent; Dolphins, Octopodes, Corvids, Elephants, Chimpanzees etc.

They stand out because of their use of logic to interpret and solve problems. Using a higher level of thought than the pure instinct of other animals. With Chimps notably being in their own Stone Age with them regularly creating and utilising stone tools to assist them with tasks.


Our inability to create an artificial version of it stems from our lack of knowledge on the exact processes behind it. Notably, we don't fully understand how a brain works, nor can we recreate one. As such, we cannot reproduce the result.

We have only been able to create rudimentary facsimiles of "Learning" by way of "Reinforcement" which is reliant entirely on applying a reaction to specific actions as opposed to any actual understanding of what is occurring.

With the most obvious example being MLMs whereby you just get a program to run thousands of iterations and any one that doesn't perform as desired is deleted and ones that do perform as desired are kept for future iterations until you get a copy that has randomly acted in a way that you wanted without any knowledge of why or how.


At the base level, we don't necessarily need genius AI.

All we need is something that is capable of thought, deduction and adaptation.

Like, if we take current usages of "AI", we have things like ChatGPT whereby you'd want an intelligence that is able to take a question you ask it and then respond to it using its own logic - Such as searching online for credible sources of information, or providing an actual opinion based on accrued knowledge.

GenAI would want an intelligence that is able to understand prompts and then actually create something based on it (Not just mash together existing media, but actually make something brand new)

Automatons would want an intelligence that is capable of adapting to change, providing the intended outcome despite variable inputs (I.e. Being able to do QA in production settings)

Of course, one of the dreams that mankind has had for yonks is robot companions, which require enough capacity for thought and learning to be able to have a unique personality and be able to express this.

Genius AI is only a small sample of what we could use (And has its inherent flaws in that it can outpace our own ability to understand rendering it either obsolete or causing us to rely too heavily on it and become complacent and regress our own capacity for intelligence)


I mean, theoretically, we could create machines that feel pain.

Pain is a relatively simple process that is based on sensory input being interpreted and resulting in the feeling of distress.

Emotional pain is more complex due to it being based around chemical reactions influencing cognitive pathways (The latter being not fully understood).

The whole thing about "Artificial Intelligence" is that when actually achieved, there would be little difference between machine and organism besides the structure of their body.

Literally, "Artificial Intelligence" is such a misnomer because if we create it, like truly actually create an actual intelligence then it wouldn't be "Artificial" it would simply be intelligence.