AI/LLMs People called an actual Monet AI slop

[The Firebird]

so here's an article that goes into more detail including people deleting their responses

Someone Shared a Real Monet Painting as AI and Asked for Critiques

Someone shared a real Monet painting as an AI image and asked for critiques as to why AI art is inferior to the real thing. Hilarity ensued.

petapixel.com

There's also this study that is also linked in the article that says people think positively about an image or even music until they learn it's AI Understanding how personality traits, experiences, and attitudes shape negative bias toward AI-generated artworks - Scientific Reports

which also applies to other things like actual creators for example you love a book or show, then you find out the person or company behind it is horrible and you no longer like it.

Yeah this seems just a more illustrative demonstration of something that is already scientifically proven (people are biased against AI artwork).

There are a range of responses to learning you don't like the creator of art; some people are bothered and others don't mind so much. I wonder if that dynamic is the same regarding AI bias? I.e., if you care a lot about a creators actions, you care a lot about AI?

 

[Deset Gled]

This classic XKCD comic has been on my mind a few times in the past week, but this thread seems like as good of a place to post it as any:

 

[trappedslider]

not worth it.

 

[Umbran]

Seems like politics in that people are now so divided into my tribe being absolutely right and yours being so wrong it cannot possibly have anything right or truthful.

I think it is more that people talk about things they don't know about.
In this case, Monet. Like, the number of folks who have studied Monet enough to legitimately comment on the OP is probably pretty small.

That's the gotcha here - not about AI, but about knowing less about art than you claim to.

 

[Deset Gled]

I think it is more that people talk about things they don't know about.
In this case, Monet. Like, the number of folks who have studied Monet enough to legitimately comment on the OP is probably pretty small.

That's the gotcha here - not about AI, but about knowing less about art than you claim to.

Some of the comments are actually pretty spot on if you consider they are judging a low quality jpeg instead of anything close to a hi res scan of the original art.

For comparison, here's the one from the thread:

And here's a 240kb (still pretty low res, IMNSHO) from Wikipedia (Water-Lilies, 1915, Neue Pinakothek, W.1796 )

Keep in mind that the original is actually ~2mx1.5m! Just imagine how much we're missing in detail, color, physical effects of paint, etc.

Now, look as some of the criticism from the meme:

"Brushstrokes for me don't feel like Monet" - Spot on. Because the low res has digitized that level of detail straight out of the image.

"It feels to flat and soft and blended. I don't know." - Exactly what you expect from a highly lossy, scaled down image.

"There is something off to me. Like some things are almost upside down? IDK how to explain it. And I don't know monet well enough to pick out his stuff specifically. But the image has something off." - Yes, there is actually something off.

Frankly, I think some of the critics are doing a decent job based off the info they were given. And these are some of the ones the jokesters hand picked as showing the most bias against AI.

 

[The Firebird]

I think it is more that people talk about things they don't know about.
In this case, Monet. Like, the number of folks who have studied Monet enough to legitimately comment on the OP is probably pretty small.

That's the gotcha here - not about AI, but about knowing less about art than you claim to.

A few of those folks did comment, and clocked it pretty instantly.

I think this relates to @aco175's point. There are a lot of reasons why people are quick to jump to conclusions when they know only a little of the story. And tribalism (here pro- or anti-AI bias--we've all seen instances of people who are far too credulous w/rt AI output) is a big reason for that.

The fact that the AI haters jumped to "this is souless and bad" rather than "this is not AI" says something.

 

[Morrus]

I think it is more that people talk about things they don't know about.
In this case, Monet. Like, the number of folks who have studied Monet enough to legitimately comment on the OP is probably pretty small.

That's the gotcha here - not about AI, but about knowing less about art than you claim to.

Yeah. All this tells us is that people on the internet are full of c**p. But we already knew that.

 

[Basic Expert]

It looked like a late-period Monet to me, or at least a low-rez image of one. However, even if we accept the premise as true, that this is an AI generated "faux" Monet, it leaves the following pretty clear distinction:

Real Monet 'Waterlilies': Worth at least tens of millions.
AI-Slop 'Waterlilies': Worth nothing.

I know which one I would want...

 

[Willie the Duck]

Not commenting on the results at this point, but I wanted to comment on the experimental design (treating this as a layperson natural social experiment).

The first point is that it is an inherently deceptive social experiment, but there is no equivalent of an Institutional Review Board. There was no outside body tasked (or ceded authority to) guaranteeing the rights, safety, and welfare of the human subjects. No one keeping it from being another Milgram Obedience Experiment (or even a Sokal/Sokal-squared hoaxes). Obviously there will be some retro-active social judgement if the process was too over the top, but overall, it would be nice if there was someone else with veto power over a deceptive act.

The second is that they only reported results after they knew what they were (and likely that they supported their existing personal notions). This is an ongoing issue in academia (in no part because an experiment with a no-result is harder to get published), but is more extreme here (and complicates matters, in that it's hard to use reproducibility as a metric instead, since successive deceptive experiments are hard to pull off). By doing so, it makes hash of the p-value testing (effectively "this situation is likely to have shown up by chance only 1 time in 20. Since it showed up given those odds, we are at least dubious that it was by chance"). This is because, if you waited until you got the results you wanted before you let the world in on there being an experiment, you've made it showing up not by chance at all.

Thirdly, yes, quality of the art (in terms of pixelation, etc.). Yes, a huge number of respondents indicated that the Monet piece they were being shown (under the understanding that it was AI) looked bad. However, it did 'look bad' in that it was a poor-quality pixilated reproduction, and it isn't generally clear that that isn't what many-to-most of the respondents were keying off of. This is part of a larger issue with 'experiments' on social media, in that they often end up being designed in a way that makes what-has-actually-been-shown very ambiguous, yet they are treated as clear-cut evidence of {whatever one's pre-existing socio-political notion is}. In this case, it seems the intended notion is that people will call a masterpiece terrible if they think it was created by AI. That may be the case, but it seems incredibly unclear that this specific social experiment really shows it (to any high degree of certainty). A tighter experimental design could have reigned that in, and made it such that there were fewer explanations for the resultant responses.

So, from my perspective, this experiment is something of a nothingburger (crossed with some apparent cynical manipulation both within the experiment and in the promotion of the results). None of this speaks to the actual experimental question of whether many people would change their opinion on the quality of something if they thought it was created by AI (based on their predetermined notion of the quality of stuff made by AI). I think that's almost undoubtedly true (if it follows existing patterns of how people perceive things), this experiment just didn't do a great job of showing it one way or the other.

 

[Morrus]

Also, of course, nobody consented to this “experiment”; they just got lied to then mocked online for believing the lie. There are major ethical issues with this “experiment”. And I use the word “experiment” in quite because it was no such thing—it was a prank, at best.