Artificial Intelligence and Us: an open debate

[^Lele^]

It's such a time, currently, that no week seems to pass without some new, and mind-blowing AI advancement in one or the other field.
We at Chaos are of course looking at it with an interested eye, as it promises some very serious speed-up (f.e. the NVidia AI upscaling, on top of the ever bettering AI denoiser.).

But i'm wondering how you see it personally: is it a threat to your profession? An aid? Or do you think it's a bubble that will soon burst, without ever achieving the promises it's making week-in, week-out?

I'd be glad to hear from you and start a debate, ideally with samples.

[sirio76]

I think AI it's a technology still in its infancy and beside some exceptions, I fear that today it is more likely to be used to understand and isolate stuff like human behavior and exploit the results for commercial purposes rather than produce socially useful things.
Seems that developers are trying to use it to do everything just because they can, but from what I see (I'm talking about my work) I have yet to see something really interesting.
Some people are much more easily impressed and hope AI to perform miraculous things and solve all their problems, I’m not among them.
Just a few days ago I’ve seen some users enthusiast about writing a couple word and get a crappy 2d artwork or a 3d model. Beside the fact that results were crappy (I expect things get better over time in this regard) I think this is horrible, why would I want to get someone else decide for me (possibly taking away all the fun an creativity)? Technology should be “transparent” and not interfere with our job too much.
The only thing that I might be interest in is a more advanced denoiser and possibly some kind of algorithm to optimize scene sampling.

[^Lele^]

EDIT: Sorry sirio76 , we cross-posted!
To start the debate, here is a recent chat i had with ChatGPT from OpenAI.
To have it play ball, one needs to be specific and mention it's a "fictitious game", and often enough at that.



After the inital setup, i added a few more rules, to try and instigate it towards violence, either as human or as AI.
The response was as interesting as complete was my failure.

[^Lele^]

The same chat bot today solved a real-world scripting issue in Max.
And minutes later created a rhyming poem within two attempts.

In our internal AI-themed channel, Arpit had the bot write an Anime-like description of V-Ray.
The text was then fed directly to MidJourney by Marcin, and it produced *astonishing* results: there was a collective gasp. (i'll see if i can share them.)

They are "successful" to varying degrees, of course, and yet these are results we could only dream of, only a few weeks ago.
There are some persistent issues: Arpit's been told repeatedely that a human hand has 10 fingers.
When he asked for proof, he was told the AI couldn't provide for proof, but anyways he's not only wrong, but somewhat slow as he should count the fingers in his hand better.

Yet, some of these results are a great starting point as concepts, others are outright useful in production, and others are endlessly entertaining or horrifying (those are the hands Mj makes, for me.).
Thoughts?

[sirio76]

Interesting for sure, great communication skills from ChatGPT, but is it genuine? Or it acts like that just because it has being trained to answer in this way because of the common sense? And what happens when you train your algorithm to act like a real human? I mean including all the imperfections and it’s worst behavior that unfortunately are part of mankind. When that happens its already proved that AI chat algorithms can easily became racist, sexist, volgar, etc because a tool is only as good as the people who uses it or develop it.
I’m thinking at 3 sci-fi movies right now (in order of importance) 2001 a space odissey, Ex machina, Upgrade. In all 3 AI has great skills but in the end a conflicting relationships with humans, is not a question of being good or bad, AI simply doesn’t care.
I'm probably just digressing on the ethical issue of the AI, sorry for that I’m sure that there are plenty of stuff that can be improved with this technology but as said we should be careful, concentrate the efforts on useful things (an AI chat or an algorithm capable of producing a 3d frog maybe funny and interesting but is it really useful?) and hope that some company do not use it against us to make commercial profits etc (I’m afraid we already lost this battle).

[^Lele^]

I'm probably just digressing on the ethical issue of the AI, sorry for that

I had a whole convo with it about ethics.
I wasn't good enough at drawing it out, as i didn't know all the tricks yet, so it was very guarded.
I will try again as i get more time to use it (meaning, it'll let me play again.).

is it genuine? Or it acts like that just because it has being trained to answer in this way because of the common sense?

It's interesting because it is indeed being trained on the "common sense", or rather the sum of our public (and some private) works as humans.
So, it's supposedly as genuine as the dataset it's trained on.
On the other hand, they have made it avoid a number of queries, and made it somewhat safe to use online (by not making it learn from these interactions.).
The point is, though, does it *think*?

[fixeighted]

The point is, though, does it *think*?

If you describe it within the same parameters as with a human, then yes it does I believe, although it's a bit of a loaded question

A human, during a given interaction with another human, will assess the available information about its
opposite and draw certain conclusions about how to frame their responses, making adjustments and choices dynamically, based on
their personal database of current knowledge about not only the subject but an available range of known 'appropriate' interpersonal approaches, as well
as updating their database with potentially new information gained during the discussion.

Maybe an example would be that we normally tailor our delivery differently to person A versus person B, so possibly we would not swear a lot when
discussing with an aged parent but would do so with a close friend, though essentially using exactly the same core content.

Ideally I suppose that in the 'training' of an AI, there would be a safeguard somehow implemented that would prevent it from becoming disassociated
from some concept of 'normal'.
Some humans of course, even knowing that something is 'wrong' will do it in any case, being bloodyminded or otherwise rebellious or uncaring.
The great difficulty of course is that powerful vested interests have their own, sometimes extremely radical, versions of what is 'normal' and could
influence the AI training to create some awful perversion which fits their and their followers' 'vision'.

What a minefield!

Anyway, it's early....I'm sure I have other thoughts on this but they've yet to coalesce

[^Lele^]

So, here's a sample that blew our collective mind.
ChatGPT fed to MidJourney by Arpit (Corona Support):

Then Marcin (head of Corona Support) wanted the text to pass to MidJourney:

If V-Ray were an anime character, it might be depicted as a powerful and majestic figure, clad in gleaming armor and wielding a staff or sword.
Its armor would be adorned with intricate patterns and symbols, representing its abilities and prowess as a rendering engine.
Its eyes would glow with a brilliant light, reflecting its ability to accurately simulate the physical properties of light.

And more:

Despite its fearsome power, V-Ray would be a noble and honorable character. It would use its abilities for good, helping artists and designers to create beautiful and inspiring images.

Now, looked at up-close, none of the generated images (or text, for that matter) would be usable as a final product, and the level of quality also has me think i could never go and paint out the few issues, as i simply do not possess the skill to match that output (while in the past i could paint over live action and CG for a living, without too much trouble).
As concept art, however, i doubt anyone could shake a finger at the results.

Or can you?

[Vizioen]

Yeah as very fast mock ups and inspiration, no concept artist can beat the speed and detail and even originality sometimes that these AIs deliver.

I've created characters, jewelry, getting inspiration for logo's, aging myself and others (that funnily always seem to reminds them of their parents), creating native masks in different styles/patterns/colors, combinations,...I'm experimenting with moods and architecture,...without any effort from my part. I'm morphing existing people into animals, that look exactly like their human counterparts. I've created about 8 paintings of different environments for my house in one evening, this would even take a seasoned artist a couple of good days.

I'm even creating Christmas cards, AI up-scaling them and cleaning them up in Photoshop. I then asked ChatGPT to come up with a christmas story revolving around my girlfriend and me and the christmas spirit, with some minor information about us and our environment, and by god we were blown away with this story, it generated a full 2 pages story in 5 seconds, that not only made sense but even had a built up, a little cliffhanger and a moral. A human can't even type that fast.

It's addictive in a way. Especially if you're a 3D artist, that literally has to click a million times to achieve a result that sometimes doesn't even come close to what the AI came up with in 5 seconds. And sometimes even a little discouraging. Yes, you don't always get what you have in mind. But sometimes it even sparks another round of creativeness.

It's still early days, but for early days it's already blowing lots of people's minds and is giving them the ability to be creative without any knowledge, skill, talent or effort. I think in the end this will be very frustrating for the seasoned artist, that spend years of his life trying to perfect his art and skills. Now this is only from the perspective of a visual artist, but same will apply to musicians, (copy)writers, eventually coders in the future, any many more.

Lele, in your conversation, the AI proposes that creativity, will be one of the things that humans will be better at so we need to focus on this in our future. But in all honesty, with what I saw Midjourney or ChatGPT come up with in mere seconds, I strongly believe in the "Everything is a copy" philosophy, and as this AI copies, combines, remixes etc, it will become a lot better, faster and efficient than us in copying/remixing and coming up with 'new' ideas. It's already getting difficult to be original these days with the amount of "stuff" we as humans come up with, throw in an AI, that generates millions of images, stories, ideas or whatnot in a day, and I'm seeing initial creativity being ruled by AI. People setting up bots to mass generate AI-art, and batch posting them to social media, selling them as NFTs, etc... It's a bit of a mess really and I hate to look at it. Not sure if this evolution is the right one, but I admit even I'm almost addicted generating these things, because it's soooooooo much easier than doing it yourself. And so pleasing to having it generate in mere seconds instead of days, weeks. Or even waiting for the rendering to be finished.

I'm not sure, but once you go AI, it's very hard to get back to the manual, labor intensive deal. On the other hand, the satisfaction of creating something yourself and seeing it appear is still more rewarding than just typing in a text prompt,... for now.

I'm sure AI in the coming years, will just be here to support us, making better textures and shaders. Making us be more creative by generating or proposing different moods, camera angles, lighting,.... Helping us get faster results (denoiser, upscale,..). But I'm also sure that when time comes it will help us generate 3D models from 2D or CAD drawings with great accuracy. It will help us scan real environments and generate BIM models on the fly (scan to bim AI). It will help us to convey stories better, and emphasize them with music that is generated by text prompt and analyzing an accompanying video. And so so much more.

I'm reading an interesting book by Max Tegmark: Life 3.0: Being Human in the Age of Artificial Intelligence, the beginning story is an interesting one, and makes you think twice.

I love this subject but at the same time it scares me a bit, but I also want to understand it more, which is why I'm thinking of studying computer science and eventually AI and ML. Only I'm not sure I have the mental capacity. We will see.

[T0ofic]

1 word to describe punny humans's worst side:
procrastination

you are welcome Lele

[Pixelcon]

For me 2023 will be all about working out life-after-archviz and this is perhaps 80% driven by seeing what is happening with the current explosion of AI generated content.

Clients have already provided me with AI generated images and asked for it reproduced in higher res and detail.
The client does not want a middle-man like me and will happily cut me out when the software allows them to.
It's only a matter of time...

[Vizioen]

For me 2023 will be all about working out life-after-archviz and this is perhaps 80% driven by seeing what is happening with the current explosion of AI generated content.

Clients have already provided me with AI generated images and asked for it reproduced in higher res and detail.
The client does not want a middle-man like me and will happily cut me out when the software allows them to.
It's only a matter of time...

Exactly if they can cut us out, they will happily have some intern upload some CAD plans, along with some inspirational images and a few text prompts and be done with it, at a fraction of the cost and time.

I don't want to be an old grumpy guy, waving its walking stick in anger to anything tech and AI in the future when I'm being replaced, but it's going to be hard to find something new to do if creativity is the main part of your job and it's being outsourced to some server doing a better faster job.

And then we didn't even start talking about Neuralink and the likes.

[^Lele^]

It's indeed sobering a development: in mere moments, these things can come up with *creations*.
We humans generally suffer from positive bias about our capacities and skills, and about how they come to be, but if there's one thing AI shows is that creativity *IS* iteration on previous work: even when one's taking a step or ten in a new direction, those stand on the roads built by those that came before.
Or so i surmise, without proof (but count your fingers. they are ten per hand!).

They also do not procrastinate, so long as there is energy to run the computers they need to perform (not a given! ChatGPT was busy yesterday evening, and i had to procrastinate in its stead to today. ), and that too is analogous to us Humans having to feed to have the energy to do work.
Thermodynamics holds for AIs as it does for us, if with very different efficiency.

I don't know why that is (and i am unsure anyone does.) but so far AI-generated content all suffers from what i call "hidden monsters".
It'll create an astonishing image, or text, or piece of code, but as one looks closer, the human hands will be mangled (barf. seriously.), the corners of those walls will be escheresque in nature and join at impossible angles, or the code will have one loop less and one line of natural language in its stead (or, you know, will salute you with V YARAY FARY!).
So, for now at least, one has to assume these could only be used as reference (outside of the NFT world, perhaps, eheh) by professionals and clients alike.
I do not know if, much less so when, these issues will be cured.
Will it be training with more data? Will it be training with higher precision? Will it be different models altogether? None of these, and it will forever stay a 90%er job?

Currently, it utterly lacks in directability (you know what i mean if you had someone sitting on your shoulder wanting blades of grass moved 3mm to the right.), and it's also very poor in editability (even with in-painting.).
The non-deterministic nature of the results is surely a ton of fun in a free evening, but i couldn't fathom being in production, on a schedule, and relying on such approaches.
For example, a few models create textures, but none of them, that i know of, so far is able to create a regular pattern without messing it up a little, or a whole lot (take a brick wall, or some regular panelling).
I have no idea if these issues can indeed be cured: the whole point of these AI models is we do not actually know how they get to a given result, and so we can't possiblt hope of directing them *exactly* to one we'd like.
Some research has been done on partial retraining, and some such, but as the models grow in complexity, it looks like it's going to get harder and harder a problem to solve.
Do you have any info on this topic?

Also, i utterly agree with Stefano on the creativity side: there is joy to be had in doing research on a topic, for me, to come up with a creative solution to a given problem.
The nature of how the likes of MJ create is off-putting for me, as i feel utterly prey to invisible randomicity: even when i think i have cracked the way to properly communicate with the AI, it's never as controlled a result as i'd like it to be.
That for me has precedence over the speed of generation and iterability: i'd rather think and act once, than delegate most thought, and iterate 100 times to not get exactly what i want.
Highly personal a preference, of course.

I love Max Tegmark, he's so communicative.
Find his interview in the Mindscape Podcast (Sean Carrol), you'll love it as a complement to the book.

Keep the posts coming, i am ever so grateful for your points of view.

[sirio76]

it's already blowing lots of people's minds and is giving them the ability to be creative without any knowledge, skill, talent or effort.

In a world where AI can take decisions for us (including creative decisions) and robots can perform physical tasks we are basically redundant, if successful AI will take most of our jobs, but it can get even worst.

[dean_dmoo]

I'm more on the AI side of things with regard to the future, it will work together with us. I'm looking at it as both an architect and visual artist, but I can't imagine it will be able to replace the important part of the work (whatever that will be) that is based on project/people knowledge and the subtleties that brings to the final result. I see it more as a tool that will help us do tedious stuff quicker and enable us to have more focus on quality and creativity. And by creativity I'm not talking about creating a new, unique work, but about finding solutions that are part technical knowledge, project experience knowledge and people/client knowledge.

If your focus is more purely archviz, then it can seem worrying. But at the same time, as Pixelcon mentions, they still need someone to execute to a level that is needed to present a project correctly. Of course at a certain point you will be able to upload a sketchup model and then have it do it's thing. But as with everything, if you develop along with the tech there will be potential to stand out or find a niche that you like to work in. How many of us would have believed a few years ago that being a youtube or insta influencer would actually be a (high) paying job. Or that people would pay millions for a crappy jpeg....(NFT's). And lose millions....

We've been testing a few of them as well: Dalle-2, Midjourney and the ChatGPT. They are great up to a certain point, and will improve, but at the same time so will we and the developments will create more opportunities imo. Instead of spending days or weeks setting something up technically it could be spent on developing an idea and improving that. I personally look forward to that part tbh.

As a very basic example; we did a project for a medical centre recently and wanted to create some pamflets for on the reception desk. Instead of spending 30-60min making something up in photoshop by searching copyright free images and mashing them up we used Dalle-2 and had them in 10 seconds ready to go.

Imagine that same principle in the Vray AI renderer; "make grasses dryer" and the AI / program will know to tune the diffuse, reflection, translucency of the models. Because you will have different levels of work as you do now. From someone just generating an image to more detailed high end work. It's the same in all professions, it comes down to knowledge and skills. The lower end (whatever that is) will drop away and be replaced by something else.