Ethics of (image) generation models

Part 2 2022/23
21

Here’s the thing though, we create and continue to use technology, any kind of technology, because it is “better” than us at doing something. So, personally I’m not too concerned about this aspect so much because we’re already inferior to all the things around us which are so much better at doing the things that we can sort of do.

I find this to be the most important question staring humanity in the face. Humans will need to re-evaluate all values and come to terms with a new reality. The fear I have is that, previously, when revolutions of such magnitude have happened, humans have responded with conflict and violence that resulted in large scale devastation and suffering until we reached a new understanding and a new equilibrium. Some examples would be the discovery of metal tools, transition from stone to bronze to iron tools, the advent of agriculture, the discovery of fundamental forces etc.

The scary thing (for me) is that, to me, it appears that as a species, we are unable to reconcile our discoveries with our values and definitions of concepts peacefully, and tend to go through difficult transitional periods of high violence and repression perpetrated against each other (not to mention the environment and other species equally sharing the planet with us) before coming to terms with the new normal.

2 Likes
22

Jordan Peterson is controversial, so I’d like to avoid commentary for-or-against him personally, but I remember this video from a few days that Responsibility is What Gives Life Meaning. There seems a basic human need to “pick up something and carry it. Make it heavy enough so that you can think that… useless as I am at least I could move that from there to there!”

That “thing” that provides each of us that sense of responsiblity will likely need to evolve (in spite of and a lot of cultural inertia), but adapting to different environments is what humans are good at. I think service organisations like Rotary, which are having difficulties engaging the rising generations will find a resurgence as people start to need a “purpose” that is not their “work”.

The optimist in me hopes/expects that when AI is good enough to displace humans in the majority of occupations, one of the compulsory Level-0 objective functions we built into them will be something simple like “provide humans a fulfilling life experience” and let the AIs sort it out. As an avid scifi reader, I’ve considered the impacts of AI over many years, and I’ve settled on the term “fulfilling” since permanent-ease or extended-happiness are not always satisfying. A “bit” of struggle can help make it all worthwhile.

1 Like
23

Okay, but it seems likely that this technology will rapidly become better than all of us at literally anything and everything we might do.

We humans will no longer be the most intelligent and capable creatures on the planet. When compared to a super-human intelligence, we will feel more like our pets do now. There won’t be anything we can do which makes us feel “that was objectively a good thing to do and worthwhile”, because an AI robot could have done it quicker, better, and at lower cost.

I guess pets and other animals can have happy and fulfilling lives. We can continue living just for the experience of living. I think that might be a good approach to life here and now, also. People still play chess, even though the AI is much stronger. People still knit jumpers, even though machines can do it more efficiently.

3 Likes
24

FYI, the last part of my talk from ten years ago touches on these concerns:

8 Likes
25

The last chart comparing human-linear-improvement and ML-exponential-improvement is particularly insightful. Humans are not good at predicting exponential change in our environment. Here are examples I like of how fast change can happen…

p.s. the tool to examine/correct the visual groupings was cool. Is that still available? Or any recent equivalents?

26

On image labelling, you might find bulk interesting, and there’s a good video explainer that goes alongside.

27

For music there is a performance avenue for many artists to continue to generate income from their work.
Similarly, authors/writers have book signing, book readings and speaking tours.

What is the equivalent for visual artists? Painters do seem to be left out in terms of residual income from past work as well as performance income.

There’s no artist royalty from art galleries that I know of, whereas as a musical artist can get a payment for a song being played on radio or in public broadcast. (Perhaps shutterstock is the equivalent for some commerical use).

Some exceptions to the visual art (painting) being non-performative are interview shows where an artists paints a portrait. e.g Anh’s Brush With Fame, and TeamLab who create giant interactive art installations that have 1000’s of paying visitors per day.

Art busking - portraits for tourists, chalk sidewalks etc. Not really something I expect established artists to warm to.

Other examples?

For visual artists, the generative models may hit harder on them than other creative arts?

1 Like
28

great thoughts everybody,
I would say that for such scenarios to happen AI will have to excel at system 2 capabilities (as per Daniel Kahneman system 1 vs system 2 types of thinking & info processing) as well, and that’s still a long way off probably. I also have addressed this issue sometimes by saying that if you consider what companies like Neuralink are already experimenting with, you can then project towards the future two lines:

  • in one line, AI excels at system 2 capabilities at some point in the future
  • at another line, the consequences of neuralink like ventures is that we end up mergin with tech and AI

Basically, today, humans and AI feel like two entities separated from each other. AI is there, we are here.
But eventually, that consideration won’t make sense anymore as advanced tech and humans, biology and AI, etc will have merged in a number of ways.

At that point in time, it’s not about: what do we do now that AI excels also at system 2 capabilities?
It will be about: what do we do and where do we go now that we are together, that we are one, and that are able to do so much?

Of course, the big question is, what will come first, as it is entirely possible that AI may reach excellence at system 2 capabilities way before we merge with it, as such merging may be really far away in the future because it probably entails understanding way better how our brain works, etc etc

But of course, there are also those who think that for AI to excel at system 2 a new paradigm change will be needed (and of course, there is the opposite side that believes that by scaling system 1 capabilities+evolving the current architectures, system 2 excellence will gradually emerge)

a lot of question marks, and referring to professor Kenneth Stanley and his book “Why greatness cannot be planned”, the only thing we can be sure of, is that the stepping stones to the scenarios 10, 20 and 30 years ahead will probably surprise us in counterintuitive ways

3 Likes
29

What is the equivalent for visual artists? Painters do seem to be left out in terms of residual income from past work as well as performance income.

Mainsteam AI tools do not currently produce actual paintings, i.e. paint on canvas, which have a slight 3d aspect and texture, with appearance that varies in the light. There is some value in having a real painting rather than a print, which is why paintings are still marketable even though they are necessarily much more expensive than prints.

Many people will support human art because it was made by a human, just as we are interested in the Chess competitions between humans even though AI is much stronger at chess than any human.

As of now, no AI has personal life experience, authentic thought, nor emotion, so the artistic direction of an AI-assisted artwork must come primarily from the human who conceives, prompts, selects and likely edits the work. Typically if we buy AI art we will be supporting the artistic person who used AI to help them create the art, or a person may use AI art as part of a larger work such as a graphic novel.

I do think that visual artists who decide not to use AI assistance will find it much more difficult to make a living as a professional artist, going forward, but it might still be possible. I don’t see that people who produce commerical art (e.g. for advertisements) will be able to make a living if they choose not to use AI assistance. Even today, it would be unwise for them not to use it.

2 Likes
30

it is entirely possible that AI may reach excellence at system 2 capabilities way before we merge with it

I think that both system 2 / AGI, and merging with AI (e.g. neuralink), are both coming very soon, on the near horizon.

Even if we do “merge with AI”, it seems likely that the AI will be significantly stronger than human intelligence by every measure, so a “merged” human and AI will not be more intelligent than the AI by itself. If I work on a math problem or a painting with my dog, the dog isn’t going to be much help.

So while it might be fun or helpful for us to merge with the AI, it won’t help us to get jobs or to feel fullfillment, as the AI would be at least as intelligent, and more efficient and cost effective, without our “help”.

It’s a weird situation. Many of us think we are creating AI to be our helpers (robot means worker or servant), however given that the AI will be much more intelligent than us in every way in the very near future, it seems to me that the roles will flip and we will be effectively pets of the AI, or stuck in the role of children who may learn and create, but do not contribute economically or academically significant work.

Any work we might do or product we might produce or service we might provide will have little or no utilitarian value. It will only be of value as an anthropological curiosity, or to patrons who wish to support human endeavour. That might be okay with us, or it might not, depending on our philosophy and approach to life.

31

A publication in The Guardian today: When AI can make art – what does it mean for creativity?. It seems to mostly rehash stuff already reported in other news outlets mentioned in this thread, but worth skimming and noting this is (still) in the news cycle.

1 Like
32

Good to see some artists are taking advantage and getting exhibits. This might be the future in line with my comment re finding the performative equivalents for digital artist.

3 Likes
33

I’ve not kept up with the very latest in generative AI this month. This morning noticed a piece from the Australian Broadcasting Corporation Background Briefing - A journey inside our unimaginable future. I had not heard yet of the “Loab” phenomenon, intriguing if not a beat up. This program tends to be reasonably researched, for a “vulgarisation” one at least.

34

I believe this is the first one of these lawsuits.
From the website:
“We’ve filed a law­suit chal­leng­ing Sta­ble Dif­fu­sion, a 21st-cen­tury col­lage tool that vio­lates the rights of artists.”

They name Stability AI, Midjourney and DeviantArt, but not OpenAI.

This is how they describe diffusion in the complaint (page 15):
"Diffusion operates in two phases. The first phase of diffusion is to take an image
and progressively add more noise to it in a series of steps. In this case, “noise” refers to
something seen rather than heard, but the connotation is the same: random fluctuations that we
perceive as chaotic and unstructured. At each step, the program records how the addition of noise
changes the image. By the last step, the image has been “diffused” into essentially random noise.

The second phase is like the first but reversed. Having recorded the process of
turning a certain image into noise over many steps, the program can then run the sequence
backwards. Starting with some random noise, the program applies the steps in reverse order. As it
progressively removes noise (or “denoises”) the data, the program is eventually able to
reconstruct the original image."

35

Last night I attended a panel discussion ‘Creativity, AI, Emerging Tech and the arts’ it was attended by a diverse group of artists, arts managers, arts educators and technologists. What I found enlightening was that the majority of artists were keen on the new technologies, many already adopting ChatGPT or Diffusion into their workflows and processes. And in after-discussions there was a general feeling that this was punk moment where art was now within reach of a DIY movement, democratising the creation processes.

Art reflects the time it is created, and those times have changed.

General concerns around copyright, and being torn between wanting to hate it because of those issues, but also loving to use it.

Caveats. Not saying that this event was representative of the broader arts community, attendance was likely biased towards more tech-curious artists, given that was organised in conjunction between an AI hub and a modern arts gallery.

Still, it was interesting to see the adoption, impacts and perception around this tech.

1 Like
36

An update on this case.

The judge also said the artists were unlikely to succeed on their claim that images generated by the systems based on text prompts using their names violated their copyrights.

“I don’t think the claim regarding output images is plausible at the moment, because there’s no substantial similarity” between images created by the artists and the AI systems, Orrick said.