A.I is here and it will make the world worse.

[Cygnia]

BackerKit bans AI-generated content from its platform, including art for board games

[Cygnia]

https://www.npr.org/sections/goatsandsoda/2023/10/06/1201840678/ai-was-asked-to-create-images-of-black-african-docs-treating-white-kids-howd-it-

 

Quote

Arsenii Alenichev typed sentences like "Black African doctors providing care for white suffering children" and "Traditional African healer is helping poor and sick white children" into an artificial intelligence program designed to generate photo-like images.

His goal was to see if AI would come up with images that flip the stereotype of "white saviors or the suffering Black kids," he says. "We wanted to invert your typical global health tropes."

Alenichev is quick to point out that he wasn't designing a rigorous study. A social scientist and postdoctoral fellow with the Oxford-Johns Hopkins Global Infectious Disease Ethics Collaborative, he's one of many researchers playing with AI image generators to see how they work.

 

In his small-scale exploration, here's what happened: Despite his specifications, with that request, the AI program almost always depicted the children as Black. As for the doctors, he estimates that in 22 of over 350 images, they were white.

 

[Christopher R Taylor]

Well in this case, the AI can only provide images it can borrow from, which is going to be 99% white doctors caring for black kids in Africa.

[Cygnia]

https://tech.slashdot.org/story/23/10/12/1814210/google-to-defend-generative-ai-users-from-copyright-claims

 

Quote

Google said on Thursday that it will defend users of generative AI systems in its Google Cloud and Workspace platforms if they are accused of intellectual property violations, joining Microsoft, Adobe and other companies that have made similar pledges. From a report:

Major technology companies like Google have been investing heavily in generative AI and racing to incorporate it into their products. Prominent writers, illustrators and other copyright owners have said in several lawsuits that both the use of their work to train the AI systems and the content the systems create violate their rights. "To our knowledge, Google is the first in the industry to offer a comprehensive, two-pronged approach to indemnity" that specifically covers both types of claims, a company spokesperson said. Google said its new policy applies to software, including its Vertex AI development platform and Duet AI system, which generates text and images in Google Workspace and Cloud programs.

 

[Cygnia]

(the video game character, not Jack)

 

Microsoft Bing AI Generates Images Of Kirby Doing 9/11

[Pattern Ghost]

Not sure why that one's a story. People do offensive stuff all day on the internet, every day. The AI didn't make them do it. And that one in particular has a massive level of restrictions. I was playing with it the other day and got an amazing number of seemingly innocuous prompts blocked. So many, that I feared being banned just from trying to figure out what was allowed and not allowed.

[Pattern Ghost]

So, I was going through the Inktober prompts with Bing/DALL-E, to see what it'd do with them. I get to "dodge" and it just spits out cars. So, I modified it to "dodging" (full prompt "black and white pen and ink art "dodging" highly stylized") to try to get the action of dodging. Instead, I got an illustration for Dante's Inferno, maybe:

 

 

 

I think DALL-E isn't quite ready for prime time.

[Old Man]

"Evading" was super weird, but here's "black and white pen and ink art "evading attack" highly stylized".

 

 

It's hilarious how it defaults to Batman.

 

Actually all of these were great, just unrelated to the prompt.

 

 

 

 

 

 

I wonder how long those links will work.

[Pattern Ghost]

It does some neat stuff with "pen and ink" style. It's just hard to control the output and DALL-E seems to really love trademarked characters. 

[Old Man]

7 minutes ago, Pattern Ghost said:

It does some neat stuff with "pen and ink" style. It's just hard to control the output and DALL-E seems to really love trademarked characters. 

 

Speaking of which, Microsoft, Google, and Adobe have all pledged to indemnify users of AI art from any copyright claims.

[Pattern Ghost]

Seems like they're pushing the tech out too hard too early, TBH. Saw that a little while back, or heard a snippet about it anyway, and it seems to me that it's more of a publicity stunt to retain users more than anything else. My guess is that their lawyers believe the artist's claims about AI copying their styles are spurious (they are), so they won't have to pay out any claims. So, it's a safe bet for them for some cheap publicity. I noticed it doesn't mention trademark claims, and I'm willing to bet it has something regarding intentional copyright violations, versus some artist offended because an AI picture had their names in the prompt.

 

I also wonder how the digital watermarking is going to be implemented. The image generators generally make pretty low resolution output, so those are often run through another AI tool to upsize them or convert them to vector (vectorizer.ai is one I just found that works well for this), or an AI re-drawing tool to remove extra fingers, etc. I'm curious about how all of this impacts those digital watermarks. If it's just meta-data (which it sounds like), it seems like it could be stripped out by those tools. 

[Old Man]

4 hours ago, Pattern Ghost said:

Seems like they're pushing the tech out too hard too early, TBH. Saw that a little while back, or heard a snippet about it anyway, and it seems to me that it's more of a publicity stunt to retain users more than anything else.

 

It's definitely too early but that's partially by design.  Satya Nadella (who is CEO of Microsoft, which owns OpenAI which is ChatGPT and DALL-E) gave an interview in Wired in which he explained why he unleashed AI too early.  In a nutshell, Google was widely regarded as the leader in generative AI, but was progressing carefully to avoid a Skynet event.  Microsoft had acquired OpenAI a while back in order to compete.  When ChatGPT reached version 2.0 OpenAI demoed it to Nadella, and when Nadella saw that it could write code he pulled the trigger on going public.

 

We spend a lot of time in this thread on generative AI imagery and text, but the impact of AI on software development is orders of magnitude greater.  It's literally software writing itself.  And so Nadella moved to seize the initiative, and a chunk of market share, even though ChatGPT was still kinda racist and even though DALL-E couldn't count fingers.

 

The other piece that's often overlooked is that AI is going to completely transform user interfaces.  In the very near future, instead of clicking around in web pages, you're going to open the AI chat window (or even a voice session) and tell the AI "I want to book a flight to Singapore on Friday" or "I need help fixing my plumbing" and the interaction will be conversational from there.  There's already been one attempt to fire a call center and replace them with AI that hilariously crashed and burned, but the day is coming when that will be the norm.  It may be hard to sympathize with call center employees in India and the Philippines, but their days are numbered.

 

Quote

My guess is that their lawyers believe the artist's claims about AI copying their styles are spurious (they are), so they won't have to pay out any claims. So, it's a safe bet for them for some cheap publicity. I noticed it doesn't mention trademark claims, and I'm willing to bet it has something regarding intentional copyright violations, versus some artist offended because an AI picture had their names in the prompt.

 

I think what's actually happening is that the main players all realized they were sitting on enough cash to settle with every starving artist and photographer on Earth, and it's a small price to pay for getting AI out there.

 

 

Quote

I also wonder how the digital watermarking is going to be implemented. The image generators generally make pretty low resolution output, so those are often run through another AI tool to upsize them or convert them to vector (vectorizer.ai is one I just found that works well for this), or an AI re-drawing tool to remove extra fingers, etc. I'm curious about how all of this impacts those digital watermarks. If it's just meta-data (which it sounds like), it seems like it could be stripped out by those tools. 

 

From what research I've done, there are two sides to this: watermarking non-AI images to prevent their use in training AI models, and watermarking AI images so that meatbags like us will be able to distinguish reality.  The latter is difficult because it's extremely easy to crop or photoshop an image to get rid of an AI watermark (ironically, probably with an AI tool).  The former is more of a technique to detect whether a given image was used to train an AI model, but even if it works it would require access to the model, which may not be possible if someone is running a private instance of DALL-E or Midjourney or whatever.

[Old Man]

Still playing with DALL-E 3.  It makes some good images but it has serious shortcomings with vocabulary.  It does not know what a ballista is, nor a lammasu.  It does know what a rakshasa is though.

 

"four-color comic book art" "muscular blond man with white pants, no shirt, no shoes, and a red sash worn diagonally across the chest, wielding nunchaku and leaping to attack in the city" "highly stylized"

 

 

If not for the fingers...

[Cygnia]

Seeker's taking 'roids!

[Rails]

And he's still about to be beaten down.

[Ragitsu]

17 hours ago, Old Man said:

no shirt, no shoes

 

...no service.

[Old Man]

9 hours ago, Rails said:

And he's still about to be beaten down.

 

"four-color comic book art" "a supervillain in a red and purple costume, punching a muscular blond man with white pants, a red sash, no shirt, and no shoes, on a city street" "highly stylized"

 

 

That looks like it hurt.

 

 

Wait, here's one that's a little more Dr. Destroyer-y

 

"four-color comic book art" "a supervillain in a red and purple costume with a red crown, punching a falling muscular blond man with white pants, a red sash, no shirt, and no shoes, on a city street" "highly stylized"

 

 

So many feet.

[L. Marcus]

Better feet than some.

[Old Man]

5 minutes ago, L. Marcus said:

Better feet than some.

 

Hmm, I wonder if...

 

"four-color comic book art in the style of Rob Liefeld" "a supervillain in a red and purple costume with a red crown, punching a falling muscular blond man with white pants, a red sash, no shirt, and no shoes, on a city street" "highly stylized"

 

 

That actually kind of worked.  Although even the AI can't deform the characters enough to truly mimic the man's "style".

 

Nice of the villain to wear a tie for this formal occasion.

 

[Cygnia]

Needs more pouches

[Ternaugh]

3 hours ago, Cygnia said:

Needs more pouches

 

Don't tell the AI that, it would probably morph one of the figures into a kangaroo.

[Old Man]

5 hours ago, Cygnia said:

Needs more pouches

 

"four-color comic book art in the style of Rob Liefeld" "a supervillain in a red and purple costume that has lots of belt pouches, wearing a yellow crown and a purple hood that covers his face, punching a falling muscular blond man with white pants, a red sash, no shirt, and no shoes, on a city street" "highly stylized"

 

 

 

...Nope, no kangaroo.  Though that might have been better.

[L. Marcus]

Good amount of pouches, tho.

[Ragitsu]

I promise not to make the world worse.

[Bazza]

3 hours ago, Old Man said:

 

"four-color comic book art in the style of Rob Liefeld" "a supervillain in a red and purple costume that has lots of belt pouches, wearing a yellow crown and a purple hood that covers his face, punching a falling muscular blond man with white pants, a red sash, no shirt, and no shoes, on a city street" "highly stylized"

 

 

 

...Nope, no kangaroo.  Though that might have been better.

in the style of Liefeld? Fail. I can see his feet.