Article: Responsible AI Use, Darktable, and DDT

Software darktable
41

I think that’s a reasonable approach from an artistic viewpoint, because it truly leaves the image untouched by AI generative or “partially generative” algorithms. Cameras for fifteen years have had facial recognition (even my old point and shoot did) and it’s a bit misleading to call that “AI” because it’s so different from the AI that has been developed in the past 4 or so years.

I wonder if that’s really worth it, given the amount of toxic waste and greenhouses gases that will be caused by AI, which will in turn lead to more reduced health. There is no free lunch.

That’s a good point, though I think that once technology becomes sufficiently advanced and complicated, there will always been a big corporation waiting in the wings to take advantage of it and push it to immoral extremes.

I don’t think they can be separated. When very powerful technologies emerge, powerful people tend to take advantage of them and thus they grow into powerful entities, often corporations. Fossil fuels grew into big oil, computers grew into big tech. And since technology generally moves much faster than regulation or laws, it’s practically inevitable.

What would happen if someone discovered a device that allowed them to read minds or kill with thought? No person could be free from the temptation to use that ability to become all-powerful. One cannot separate a device that grants power from the necessary consequence of that power being realized.

I think a lot of people tend to think along the lines of “if only we could use this tech for good”, because it’s too horrific to believe that we have no control. But I do believe we do not have control when the power of technology becomes sufficient.

I’d disagree with that. Politicians only juggle ameliorations and who gets a bigger slice of the pie and who pays taxes. But the impact at large cannot be controlled by governments or politicians or lawmaking. Once power exists, people will want to take advantage of it to their own benefit, regardless of politics. And of course, because technology moves much faster than politics, there’s not much hope for politics except damage control. (In the long run. I don’t discount local power to make certain things safer, but as I said, those are just ameliorations, not a disruption of the general trend of the impact of any technology.)

That’s the key, isn’t it? No matter what minority exists using AI in a responsible way, there will always be the unscrupulous. And the effect of that unscrupulous behaviour is only proportional to the power of the technology. At which level of power will that effect be too dangerous to compensate for?

I agree with that. I love darktable and I greatly respect what has been done with it over the years. I’ve used DT since 1.4 (around that anyway), and I think it’s such an amazing piece of software. I’ve enjoyed editing with it since Day 1. But the recent introduction of AI within DT, even if it’s optional, made me lose some enthusiasm for it. Especially seeing how it’s shifted the emphasis away from learning to edit photos to a more automatic point of view.

I do think one of the primary reasons why it’s so polarizing is because never has an automatic tool before come so close to threatening creative work. It’s not really creative, just a large statistical thing, but it replaces creativity in a practical sense in many arenas and that’s just offensive.

That dinstinction is definitely important, and one I’m aware of. Still, I do think the spirit of those models is similar at the core to other types of AI in the sense that they can lead to similar outcomes. For instance, what would happen in the future if:

  1. The “world” agrees on “safe and ethical AI standards” that use only open training data, and these standards satisfy even the most staunch open-source advocate

  2. Facebook or some such company develops a new, amazing AI subject recognition model that verifiably adheres to these standards, but the model required 100 gigajoules to train

  3. Facebook releases the model as open source, which could be integrated into DT.

Would it be integrated into DT and other open source projects? Would the feedback and modifications from the open-source community then be used to develop further models that require even more energy to train?

I know, everyone must make their own choice. I’m just saying that I don’t put much stock in open vs. closed vs. ethical vs. unethical models because at the end of the day, it’s a trend to use machine learning up the freakin’ wazoo for very little gain, and that entire philosophy in my opinion is suspect.

1 Like
42

But I’m not sure at all that AI is a bad thing generally .

The main problem is the speed of change - for more than one reason.
Many people strongly dislike change, and have no time to adapt.
Many changes aren’t worth being done, or even worse have really bad societal consequences which will take a long time to solve - because societal changes always are slow.
Human societies always had problems dealing with that, but now change is supercharged, and the amount of power (money) behind the push to change makes it extremely hard to slow the changes down. Which will either make the world a living hell for many, or bring on a bloody revolution - or maybe both.

There are lot’s of good things to be accomplished with it.

Yes?
I’ve a constant load test on my servers because i’ve over 50% AI traffic. Not good - because i know too well what my users write, and i don’t want anyone too train future AI systems with that (one user is the exact opposite: i want AI to be trained from his texts - but that’s one user of 400 to 500 monthly users). Did someone say robots.txt? No, AI bots mostly ignore them. The people behind them are really awful.
(and don’t get me started on copyright).

Then there are the AI bots “testing the security” of my software. I’m quite sure at least some of the successful tests would end with an exchange of some bitcoins against encryption keys), and others would end slightly more legal in a “disclosure” of the security whole after some months if i don’t pay to keep it undisclosed (mostly bitcoin, if course).
Now i’m not finishing the redesign i was working on, but adding layers of hardening. Maybe that’s good in the long term, but right now it is incredibly annoying.
And since i have no second developer for the project, i now use an AI to audit the source code. Meaning i have to pay the same disgusting industry ruining society, because buying an GPU able to run one of the top models is out of the question now.

Which brings me to the next point: the price of hardware. No good.

But yes, i know, we are now entering the golden age of software. There will be more software than ever - that is what they tell us. But who cares? 99% of the software ever written lives in niches or rots in private or public repositories.
Who needs more of that? Especially in light of the fact that maintenance eats 80% of the engineering time or tokens of software projects?
Yes, i know, AI might kill SAP and Oracle, and that would be a good thing. But killing these companies will not be as easy as just writing software and fixing a few bugs. SAP is an ecosystem, and has stamps of approval by more regulators than i can count . Replace that with AI? Fast? No way. And Oracle? Exactly the same. The companies might die (and i would never bet on oracle surviving Ellisons AI bet), but not the products.

The golden age of software… wait… what about the next generation of coders who now don’t get the experience needed to control the LLMs work? Control which will still be needed…

But, yes, LLM can do good things. Like any newbie in their job can do. Like a very talented newbie doing clearly structured jobs bound by logical rules, or an annoying newbie in other environments.

But I do see a need for regulation and verification.

The main problems with regulation are speed and money. Regulation is a slow process at the best of times - and the amount of money on the side opposing regulation will make it even slower. btw: most of the time regulation comes after a damage was highly visible.
And verification will not happen. Its an arms race, against a technique without any built-in verification hooks. LLMs just aren’t build for that.

Quite possible we are all just frogs in a pan with slowly warming water, not jumping out in time. Which quite probable is NOW.

Frogs notice when they are boiled. They just have nothing to jump from, if they are already in the pot.
Maybe the race who did that to the frogs deserves AIs.

I hate using AIs. Its just like having a junior developer who never learns.
And i love them. Because i’m blind for the mistakes i made in my (quite old) code base, and my users deserve protection.

Sorry for the rant. But i’ll press the reply button anyway.

Regards, Uwe

2 Likes
43

I’m nearly that old that I can remember the fear of steam engines… :wink:

Many died because of those! Yet - in the end - they were beneficial to our world. Even when their appearance more or less mark the beginning of industrial CO2 production now causing serious problems because of serious and long negligence.

The lesson of this:

  • change is always difficult.
  • a good argument is mostly not decisive.
  • it’s very hard to foresee the course of a development.
  • when the powerful hijack a development trying to monopolize its benefits they get themselves in serious trouble sooner or later.
  • current oligarchs will probably run into trouble as well.
  • just as before there is a great lack of answers on urgent questions.
  • we’ll have to work hard and organize ourselves to get things better.
44

I’m not really sure if they were beneficial. I’m pretty sure cars were not. And if steam engines contributed to being locked in this mode of production for the sake of production, then I don’t think it was worth the environmental catastrophe.

Negligence? I’d say it was not at all negligence at all on the part of many, but rather willful destruction because those doing the destroying knew they would not have to face the most serious of the consequences.

1 Like
45

Exactly. This is not negligence. This is decades of Big Oil lobbying and manipulating the public through mass media. Exchanging the future of the planet for very short term profits. Not only exchanging the future, but the present too for some people. It continues to this day, only now they even managed to get the president of the US to shamelessly and publicly further their interests, quite the feat…

Big Oil did private research and climate modeling in the 1970s and knew about the disastrous outcomes.

46

I will respectfully disagree. AI gives an additional tool for troublesome noise. I don’t need to use it often as the non-AI options are usually good enough. AI masking can sometimes be used to select objects that are difficult to select by non-AI means. But then the edits applied to the masked area is still in my creative control. When I want to mask I can choose a brush, a path, a parametric mask ,an AI mask or whatever I like. That is the freedom of choice that I love. It is masking capabilities that had me select DT over Rawtherapee.

AI in DT is very restrained at this point of time. I don’t see that changing in the near future.

1 Like
47

I never said it was universal for everyone. Just shifts the emphasis in general, a little. I know there are people for whom the emphasis is not shifted.

2 Likes
48

Can you explain this? With respect, but I really don’t understand what you’re saying here.

  1. what has AI masking changed in the sense of editing the image? What edits exactly are automated here? It does not edit the image in any way, it only helps to easier create a mask. Nothing magic, I could this by hand, it just takes a lot longer.

  2. AI denoise is of course slightly different. Yes it does edit the image. Yet, the aim of feature is not to do automatically edit the image or create a certain look: denoise clearly has the aim of delivering the image as if it was taken with a lower iso. No change in how image itself looks or its composition or whatsoever.

  3. Can you point me to editing modules that are not in darktable 5.6, but would have been if the ai modules were not developed? I can’t see anything here in pixls forum that say: o we don’t have the capacity to include this module in 5.6 because we are focused on ai modules.

So please explain your reasoning, I can’t see darktable development that shifts away from ‘learn how to edit’ to ‘the software will do it automatically for you’

1 Like
49

I think we’re getting a bit away from the topic because my greater concern wasn’t even editing process, but I realize I made the remark so: automatic mask generation does not do the same thing as what a human would do. For some, that’s not a problem but many people will tend to just use the mask the computer gave them and not think about it. And that promotes more automatic thinking over time. There’s no use in giving me your own example because likely you are a person who thinks more carefully, but most people will just use automatic masks even if they don’t make sense.

When I draw masks for instance, I would likely never want the exact same auto-generated mask. I tend to draw them in a way that corresponds more naturally to drawing. Sometimes of course the mask will be “exactly the sky”. But even then, I think automatic masking encourages people to focus more energy on photos whose light is suboptimal or not that good anyway (in some cases), because it’s so easy to make more wild modifications. Like I frequently say, if you’re using tons of masks and local editing that require very precise masks, there might be something wrong behind the camera. Automatic AI masking discourages people from asking what that could be.

I hate to beat a dead horse but I’m not saying that holds true in all cases, and it’s less likely to be true for people who have done it “the hard way”. But in the future, people who grow up on automatic masking are likely to be more lazy in their photography.

That’s wrong. I already demonstrated in this Photography Life article how ML denoising can interpolate and change the image. It does interpolate on a micro-level and of course by definition it reconstructs patterns from random noise so yeah, it does change the image. It looks different for sure.

But that’s again not my main point within the scope of my previous comment. It seems to me that AI-denoising is way overblown. It’s just a fact that beginner photographers (and this is not meant to be an insult, everyone including myself has gone or will go through this process) is likely to take photos at suboptimal settings that introduce much more noise than is necessary. Yes, that’s not always the case, noise can’t always be helped. But in many cases it can, and AI denoising just encourages people to be lazier with their camera settings. Or else just encourages people to try and salvage more photos that could be improved in other ways with a more thoughtful process.

It was never my concern that other modules would not be developed. Only that AI encourages more automatic photo editing, that it encourages people to buy upgraded hardware to edit photos instead of improving technique, that it takes lots of energy, that it normalizes the use of AI in general.

Finally, in my opinion, this comment box is too short to give a complete reply to your question. But Jacques Ellul does a good job in his book “The Technological Society”. It’s worth reading.

Like I’ve said many times, much of this is not about AI but AI seems to represent a huge leap in these nasty effects and it is still a nascent technology and therefore it’s one of the best examples of technology gone past the point of diminishing returns. I’ll leave just one quotation from Jacques Ellul from his book “The Technological Bluff”, the sequel to the aforementioned book:

The more hazy and hypothetical or contested a problem is, the less it affects the public. People prefer not to see it or to hear about it. Those who speak about the dangers are labeled ignorant or pessimistic. The advantages are sure and plain to see; the disadvantages are diffuse and uncertain, all the more so since there are experimentations.

1 Like
50

@JasonTheBirder, I think you wrote a good, thought-provoking article, and it’s a shame that so many of the comments underneath the article (not here) were so combative, with one person saying they would cancel their subscription and only resubscribe “if the team develops more well reasoned positions on various aspects of ai”.

That’s essentially saying “I will only read views that align with my own”. This is exactly what the fabled “echo chamber” looks like.

I’m still developing my views on AI, although I’m highly pessimistic of what it will do to us in the long run. But I also acknowledge that it’s something we have to live with now, so I’m working out how to live with it.
It’s a constant struggle between how I want to live my life in an ideal world vs the reality of just getting by, earning enough money, taking care of my family and looking after my mental health.

51

It is the same kind of idea as LLM deskilling. The more you use it, the higher the likelihood that your own skills deteriorate.

6 Likes
52

AI helped me learn to triangulate my position on a paper map using pencil, compass, and ruler. This was after several abortive attempts with YouTube videos. (I lost 190 lbs and took up back country hiking where this skill is necessary.) Those who want to use AI tools to aide or accelerate learning, will. Those who are lazy and wish to avoid learning will also find a use for AI. Human nature.

Same problem with fire when it was first discovered, I am sure.

My personal line is that I like using AI to help me share my photography and take the administrative burden of that off myself, but I prefer to create my own images and compose around things rather than remove them in post with Adobe Firefly.

1 Like
53

Thanks for the feedback. Most of them were willing to read it and debate though, and that’s still pretty good. But yeah, one or two people got upset. I think one component of that is that there IS fearmongering elsewhere by the media, and hype by the companies themselves, so that sometimes a more rational viewpoint can be drowned in what at first appears to be more of the same stuff.

That’s absolutely a good point. We all have to take care of ourselves first. That’s why I make videos on YouTube, even though I hate consumerism and Google. It’s one of the only ways I know how to make a living and it’s preferable to working in a 9-5 office which I hate even more, and arguably the 9-5 is even more destructive, at least when it comes to my previous line of work as a mathematician. But without YouTube I wouldn’t even make enough to break even.

Darn, it’s tricky indeed. But the only way to make a living for most is to participate in this twisted society in some way, and even if you “exit” it like buying a plot of land and tilling it yourself, you still have got to buy that land and that requires money, which sometimes also requires advancing what we don’t agree with. That in turn creates a cognitive dissonance which is hard to face, and that’s probably another reason why I’m quite frequently attacked.

In many ways, it’s a depressing situation but I’ve taken it to be something to fight for, a purpose - to question civilization. One must make do with what one has. The alternative viewpoint of just finding a small corner in the world and ignoring it and being joyful all the time (quite a different thing from happiness, I’d say) is not acceptable to me, though of course I am trying to find a safe corner for myself…

That is true. That is why I think one should be very careful on relying on any tool, not just AI, but especially the tools that work in an opaque manner. For example, I edit my audio by hand but I recently wrote a macro to help me repair problems in short sections of audio. However, I’d say that’s quite a different thing than AI because (a) I wrote it myself only after understanding the problem and (b) as I continue to use it it, I still resort to the manual technique that I already learned and improve the macro.

But even if one doesn’t write or make their own tools, and nobody can do that for all such tools, it still makes sense to avoid feedback for which the outcome isn’t completely known or understood. That’s why basic autofocus is not something I protest about, because it’s easy to articulate exactly what the outcome should be and even a rough idea of how it should work, even if one cannot implement it.

Unfortunately in some ways, the effect of automation and its complexity on human beings is quantitative. No one can really give any exact point at which automation becomes overpowering because it’s a proportionality: the effect is proportional to the power of the thing. And AI has one other attribute that goes beyond many existing tools: it can grow without bounds in multiple dimensions, which makes it especially difficult to handle.

2 Likes
54

Hi and welcome to the forum!

Sounds pretty dangerous, as there is a possibility that it could have taught you incorrectly, leading to you putting yourself in danger. Glad you seem to have learned it though. And congrats on your weight loss, that is something to be celebrated.

This is, of course, the danger with LLMs and their tendency to completely make up things when you’re not already skilled in the domain in which you’re seeking information – you won’t necessarily know when the LLM is wrong.

I don’t find this very compelling, sorry.

We’re a forum dedicated to Free and Open Source Software. If you’re plan is to use that software here, please don’t.

1 Like
55

I went hiking with a friend who is former military. He said I nailed it. I’m hyper cautious about safety when hiking solo (e.g. I carry a SpotX and also a ResQLink.

I use the tool that fits the job. I also use GIMP, RT, and I’m a huge fan of DarkTable. I’ll take a Linux desktop over Windows whenever possible. I don’t proscribe how you should live your life, so perhaps extend me the same freedom?

56

You completely missed the point. This forum doesn’t welcome any non-FOSS software (with some freeware exceptions I guess). No one is dictating you anything, chill.

57

What I’m truly worried about is that yes, AI can help one learn, but it also tends to focus learning on very narrow-domain problems and it tends to give a person narrow-domain skills that aren’t easily generalizable. Some fastidious people will become better programmers, have a language conversation partner, but what will they not do?

That’s the question. General learning on one’s own with a more active approach often encourages going into random directions that result in cool stuff happening. That might not have happened in your case, but it often will in others.

The greatest problem with AI and learning here is that one doesn’t really know what is lost because AI is so direct that it actually removes random avenues for exploration. In other words, I posit that AI trains people to become very competent without much creativity, and to take a problem as given, rather than to question whether the problem is really the right one or whether it should be solved in the first place. That’s of course in the long run and on average, which makes it exceptionally dangerous because the early problems it does solve don’t give a clear indication of this general trend.

We have been trained to view things only at a personal level, because technology for the past 2-3 hundred years has made us overly focused on the self. That can be a good thing up to a point, but it’s long past the point of diminishing returns now. Unfortunately, the dopamine stimulation of contiuing this nightmare is probably too seductive for humanity to turn back simply because of the momentum of the process. A rather sorry state we’re in.

1 Like
58

What else is your “What many people might do” line of argument, but fearmongering? There is no logic to it.

ok… I’m thinking that one might follow such a path of thought if the topic was, for instance, people using AI tools for psychiatric care and counselling. I cannot name sources, but it is my understanding that this has already led to deaths.

We are talking about drawing a mask to match a shape. It is unlikely to kill anybody. In terms of automation, it is not far from parametric masking by selecting sample areas.

It’s not easy to talk about AI. There are areas which I consider to be a useless product; there areas where I think it is dangerous. And people say stuff like it has to be regulated. Who’s going to do that! How?

But in this tiny corner of a limited photo-editing tool, I think there was, and is, too much protest. But I can’t complain, because that that protest caused boundaries to be set and established.

It’s good to talk. But perhaps, now, I out to be editing some photos instead :D.

59

There’s quite a lot of logic and precedent for it, but I’n not going to write more essays about it because I’m just exhausted for now. But anyone may consult a lot of carefully-reasoned opinions in David Skrbina’s book “The Metaphysics of Technology”, which summarizes the philosophical thought on the matter.

I’d like to remark that no where did I ever say anything about AI taking over the world, AI killing people, or other AI crazy disasters. Just about what AI has done so far, which is already sufficiently disappointing, and the guaranteed fact that AI uses more energy than we ought to be using, and that alone is enough reason to reject it because the attitude towards it is much more important than individual use cases, which I’m not interested in.