AI discussion (split from Canva thread)

[JGD]

10 hours ago, R C-R said:

So then is it fair to say that to you "the bottom" has something to do with too much focus on AI tools? If so, then if Affinity adds few if any AI tools would you consider it out of this race to the bottom, so to speak?

It depends on the mix and main focus of the apps and their tools, I guess. 🤷‍♂️

On-device AI tools, using your own content and Apple's, Qualcomm's, Intel's or AMD's AI cores? Meh, whatever. I may even dabble with those here and there depending on the client, practical application, etc.

Crowd-sourced and server-side stuff, which many a creative will tell you is completely anathema from a philosophical standpoint, with no option to opt-out or as the main focus of the app/workflow or of too many of its tools? Oof, no thanks.

I'm taking the same approach to creative work as I am to my writing; or, better put, I may have a more liberal approach, because writing does hold a more sacred place in academia and self-plagiarism is way more of a problem there than in the creative arts. Sure, I may use an LLM to summarise someone else's work just to make my life easier in finding the information I need (I'm still reading the real deal and confirming its relevance before citing a word of it, of course), and I may also use it to produce some outline for a document, because I have a really bad case of ADHD and some trouble in getting work started, but do a clean-room implementation from it, with zero copying and pasting of text (heck, I may even use another Mac logged out of my iCloud account for those prompts, as I have a lot of those lying around and may be wary of its otherwise very helpful Continuity copy-and-paste feature across different devices), of whatever I was aiming to convey. Even if an LLM could, in theory, accurately reproduce my writing style if I fed it all of my academic production and the desired prompt, it would still be a machine doing it, my brain would just wither away, and having to study “my own” work so I could present it and defend it, when I can do that way more easily when it's fresh off the press and fresh in my mind, would sort of defeat the whole purpose anyway.

When it comes to the creative arts, I'm still quite conservative, so let's just say that depending on how… artistic and “authorial” I might want a certain work to be, I might use a certain mix of AI tools (or none at all!), but always based on my own input and assets. That's strictly non-negotiable for me. And, sure, no person is an island and I'm obviously not immune to external influence (you know, as they say, Ex nihilo nihil fit), but I'd rather have my natural, water-and-fat-based intelligence do that process for me. I'm okay with seeing the computer as a colleague I bring in to my process, but I'm not okay with bringing other humans into my process – even if they consented to it! – with the computer as a – IMHO, still quite dumb – mediator. Unless, of course, we humans know each other, or have some line of communication, and can team up to try and trick the computer with our inputs, or something, thus gamifying the whole thing (there's something to be said about the importance of play in the creative process). TL;DR: “AI”, as it stands now, is a bit of a cadavre exquis on a massive scale, except it isn't because people don't know each other, don't see the fruits of their labour, and the machine does all the… stitching together, and if there was a way to just revert that massification process and humanise it a bit, artists might be more willing to embrace it and the results might be more interesting.  

I might actually be on to something there, and using Canva's tools for literal and active collaboration, maybe even between teamed-up strangers, social-network-style, and the machine, could be more interesting than just letting the black-box-of-AI-doom do ALL the work for you. I'm also aware that algorithms were, at least at the outset, human creations, so in a sense we're cooperating not just with the machine but also with its programmers… OTOH, those algos are so far gone, convoluted and themselves machine-generated at this point (they don't call them “black boxes” for no reason) that I can almost put them on the same level as other digital tools I already use, of which I technically know almost nothing and which impact my creative process in ways probably more relevant than many understand or care to admit. You do get that sense of perspective when you get to do proper calligraphy, letterpress, stonecutting, etc. at least once… Then again, that sense of perspective is also what's been nagging me for years to ape many of my colleagues and mod my Parallel Pens and whatnot, but also to go and learn Python, and produce my own add-ons for Glyphs.app. That day will come, even if it's basically useless and I'm retired by then.

[William Overington]

I find the issue of Artificial Intelligence (hereinafter, AI) interesting.

My own limited experience of using AI is that it can be very good, but it can also give bad results. It needs assessing.

For example, I tried asking it to produce various pictures of a lady in a long green dress feeding an okapi. I chose that as I thought it probably unlikely to be have done before and because I have always liked okapi since I saw two at Bristol zoo in the 1960s, not having known of them previously.

Some of the pictures produced by the AI are really lovely. For example this one. (Please click on the picture to display a larger view.)

Original art generated by Bing Chat AI (Page 1) — Art & Literature — Alfred's Serif Users' Forums (punster.me)

Yet some were wrong, for example, one had an animal with the body of an okapi yet with huge horns more like a gazelle.

Bing Chat AI usually, yet not always, produces a set of four pictures.

Selection of one, or choosing to ask again and get some different pictures, is a decision for a human.

How an expert in art would assess that picture of the lady in the long green dress feeding an okapi I do not know. Yet, whatever, I am happy with the framed print of that picture that I have.

Text generated by AI is interesting.

Given a prompt with the outline of a story the AI will write the whole story.

Some of the stories are really beautiful.

Illustrated stories produced using Bing Chat AI (Page 1) — Art & Literature — Alfred's Serif Users' Forums (punster.me)

Sometimes the AI has a mind of its own as to how the story flows, producing an unexpected twist in the story.

AI asked to complete a story (Page 1) — Art & Literature — Alfred's Serif Users' Forums (punster.me)

More threads listed as follows.

Art & Literature (Page 1) — Alfred's Serif Users' Forums (punster.me)

William

 

[R C-R]

8 hours ago, JGD said:

It depends on the mix and main focus of the apps and their tools, I guess. 🤷‍♂️

I suppose so, but even if Affinity abandoned whatever plans Ash hinted about being developed for the apps back in B.C. (Before Canva) times & Canva suddenly ceased to exist, continued aggressive development of AI tools of the type widely considered to be a threat to professionals of all types will not stop.

So I think the best we can hope for is some sort of legislation being enacted in the UK, US, Australia, & so on to put limits on how those tools can be used.

[JGD]

1 hour ago, R C-R said:

I suppose so, but even if Affinity abandoned whatever plans Ash hinted about being developed for the apps back in B.C. (Before Canva) times & Canva suddenly ceased to exist, continued aggressive development of AI tools of the type widely considered to be a threat to professionals of all types will not stop.

So I think the best we can hope for is some sort of legislation being enacted in the UK, US, Australia, & so on to put limits on how those tools can be used.

I'm actually putting my money in the European Parliament or the European Commission, at this point… They seem to have an axe to grind with international big tech companies, and while some of their demands are completely brain dead (like forcing Apple to allow users to uninstall the Photos app from their iPhones… Are they for real? Nobody's asking for that! 🤦‍♂️), they may eventually hit some fair targets. And, to wit, there's a growing discourse against AI replacing jobs en masse. Unlike in the US and elsewhere, we do give two effs about maintaining a modicum of social stability.

[William Overington]

The fact of the matter is that there is no way that I could have painted a picture of anything like the quality that the AI system produced, and I would not have been commissioning a professional artist to produce such a painting for me anyway.

Though that is just my hobbyist art activity.

Where exactly is the claimed threat to professional designers? I am not in any way suggesting that it does not exist, I am just wondering in what circumstances AI is seen as a threat to their livelihood.

William

 

[gguillotte]

Any product using generative AI in any capacity makes it a business risk for me, because the creative field I work in strictly prohibits its use in any capacity for any content, writes those prohibitions into contracts and publisher agreements, and enforces them aggressively.

I'm not morally opposed to the features and personally find them fascinating, but strictly pragmatically there should be options to completely disable them in creative software, and if its usage is automatic and either opt-out or unavoidable then I will always prefer a less-capable tool that lacks any such features, not out of a knee-jerk reaction but to protect my relationship with such clients.

[William Overington]

Do you know why that is please? Is it anything to do with concern over intellectual property rights in case what is provided by AI is an unauthorized copy of someone else's work?

I asked Bing Chat AI about whether images were copied and I was told that the AI learns the technique of the artist and then paints an original picture using that technique.

I have tried Renoir, Claude Monet, Boticelli, and some others if I remember correctly.

This one in the style of Claude Monet, as an experiment to observe whether the AI would do it - and it did.

Original art generated by Bing Chat AI (Page 1) — Art & Literature — Alfred's Serif Users' Forums (punster.me)

William

 

[William Overington]

Attempts to get the AI to draw a gibbous moon were unsuccessful, even when other ways of describing that desired image were used in the prompt. A full moon was always supplied.

Also, attemots to get the AI to produce a painting of the 1813 locomotive Puffing Billy were unsuccessful, always a picture of a much later (around fifty years later at least) locomotive.

But it does not say it cannot do it, it just supplies something else without comment.

William

 

[R C-R]

3 hours ago, JGD said:

And, to wit, there's a growing discourse against AI replacing jobs en masse.

If one takes a broad view of what AI is, particularly the idea of it being anything used to dumb things down to the point that it no longer requires hiring someone with special skills to do a job, I think that has long been something most of us actually want & accept ... as long as it does not impact whatever skills we might have to earn a living.

Consider that once, just to start & drive a car with an internal combustion engine, one had to master the use of manual controls like spark advance & the choke. Automating those functions is not exactly AI but it does dumb things down to the point almost no car owner needs to hire a professional driver just to get around.

A more contemporary example is the invention of the GUI that made it possible for almost anyone to use a computer with little or no training. 

i only mention this to point out that to get any meaningful measures in place to limit how & where AI can be used, the first thing that has to be done is to clearly define what AI is & what forms of it are to be regulated. I think generative AI should be the main focus of this, particularly as it relates to ownership rights. But the legal issues are complex so I doubt we will see anything substantive from any government about this in the short term.

[SrPx]

2 hours ago, William Overington said:

Attempts to get the AI to draw a gibbous moon were unsuccessful, even when other ways of describing that desired image were used in the prompt. A full moon was always supplied.

Also, attemots to get the AI to produce a painting of the 1813 locomotive Puffing Billy were unsuccessful, always a picture of a much later (around fifty years later at least) locomotive.

But it does not say it cannot do it, it just supplies something else without comment.

William

 

It probably does not have enough data for it. Any average artist is "smarter" than these systems (so I think the term "learn" is somewhat incorrect. They don't know of anatomy like I do (or at all, better said), or any other artist who dedicated at least a decade or so to the  matter). But them having gazillions of data to 'mashup' gives them the edge in many commercial markets, which yep, destroys many artists' way of living, ends up with a bunch of artists having to quit. That while... you telling an artist to dig info about the 1813 locomotive Puffing Billy, or whatever, this person will research about the matter and do a collection of art concepts about it before even starting the project. Will also consider many more nuances (and making more sense at it) than any AI. That advantage is not enough, though. 

The concerns of clients and companies (for not using AI) in some cases are related to privacy issues, industrial secret, etc. As things can get leaked not just with text, also images. And I suppose some level of uncertainty/lack of control introduced by AI is a problem, too.

I have zero problems with AI being used for content aware or speeding up some tedious process, when it is not substituting fully the creative job (BTW, with AI, I have seen in many articles and people's comments in lots of social platforms a considerable confusion about the art concept... a prompt is not "art", no matter how elaborated the prompt). 

The Procreate app has taken a very different stance (probably unique in the industry... huge kudos to them), opposing to AI, to defend the ones who helped them bring the app where it is now (the artists). 
 

Quote

Where exactly is the claimed threat to professional designers? I am not in any way suggesting that it does not exist, I am just wondering in what circumstances AI is seen as a threat to their livelihood.

Design.. yep, it can be also challenged (the jobs matter). Less so than with artists, by far. But it is already capable of quite in several apps, and perhaps it's a matter of time.

But a designer is a lot more of a "puzzle maker", and we (I am also a designer and 3D artist, though illustration is my passion) have always been able to use such tools to focus on function (more so since the Bahaus ), composition, etc, always thinking on the end user.  Programmers are less affected, and anyway, they have many more ways to integrate with the whole AI train. 

One of the major problems is that.. Although it (the visual art world) has not (yet) seen applied as severe methods as happened in the music industry in the past (to protect the musicians IP), when you use content to build a tool (AI apps, the totally essential scrapping of content), and you don't ask for permission, do not sign any contract... you are using content in an unfair way. Whether regulation will be able to fine tune several matters to bring things to its most fair state, it is hard to guess. Also, there's a lot of money to be gained, and that makes it more difficult, as a lot of powers that be will oppose, money move mountains, way more than ethics.   

[William Overington]

Interestingly, Bing Chat AI states that it was trained on fully licensed training material.

William

 

[R C-R]

42 minutes ago, William Overington said:

Interestingly, Bing Chat AI states that it was trained on fully licensed training material.

So perhaps it is forlorn hope, but maybe they are paying decent money for the licenses for this material & a market will emerge for those skilled enough that they can make a living selling the licensing rights to the content they create specifically for training AI's?

[PaulEC]

Generative AI is quite unreliable, for example, producing images of people with extra fingers. So why assume that when you ask it a question the answer is accurate?

[debraspicher]

I've saved some AI eye candy from Pinterest, etc for myself, but yeah, the errors are glaring. I don't care for my personal usages, but I would never put it on anything commercial.

[monzo]

I’ve been playing with Stable Diffusion a bit, and the results can be pretty random, even with the same prompt, and I find it really difficult to get it to create several different elements in the same piece - it regularly misses things out.

However some of the stuff it’s churned out has been really interesting, and the font styles it makes up can be wonderfully inspiring (though usually completely illegible). 

[William Overington]

I think of AI somewhat like how I think of langauge translation programs, namely that, if, for example, I view a video of a song in a language that I do not know and I wonder what it means, if I find the lyrics of the song in that language on the web, then the translation program translating from that language to English gives me a good impression of what the song is about, but I would not use the translation program to publish something that I had written translated into a language that I do not know.

William

 

 

[William Overington]

For example,

William

 

[nickbatz]

7 hours ago, gguillotte said:

 

I'm not morally opposed to the feature

I am, and I can't state it strongly enough.

The whole idea of a machine trying to produce "art" makes me twitch uncontrollably. Art is the soul of humanity.

[Alfred]

8 hours ago, PaulEC said:

Generative AI is quite unreliable, for example, producing images of people with extra fingers. So why assume that when you ask it a question the answer is accurate?

Polydactyly is rare, but it does occur in real life.

[Bartek]

I use stock photography very often in my work (DTP, social media). From my clients' perspective, it makes no difference whether I use a purchased photo or AI-generated graphics in a designed flyer or banner. For me, it is a convenience because I gain control over the materials I use in the project.

In both cases I have to make a lot of adjustments to the images anyway, but the AI-generated images are at least more thematically in line with what I expect.

Long Live AI!

[William Overington]

A mixture of views.

Should we recommend to @Ash pursuant to Pledge 4 that involvement of AI should always be totally optional in any particular end user design project, the choice totally controlled by the end user?

Should we go further, and recommend to @Ash pursuant to Pledge 4 that there be two separate programs, Affinity Designer with no AI included in it at all, and Affinity AI Assisted where the choice of whether to use AI for any particular end user project is a choice totally controlled by the end user? The rationale for this having of two separate versions is that @gguillotte and others in a similar position in relation to contracts could then provide provenance to customers that they are using software that does not involve the use of AI at all.

William

 

[RichardMH]

If Photo goes the generative AI route, then we need them to join the content authenticity initiative and editing adjustments be in the meta data. I've been led to believe Photoshop has it already.

[William Overington]

18 minutes ago, RichardMH said:

If Photo goes the generative AI route, then we need them to join the content authenticity initiative and editing adjustments be in the meta data. I've been led to believe Photoshop has it already.

I had not known of that until I read your post.

Could you say more about this please?

I found this.

Content Authenticity Initiative - Wikipedia

William

 

 

 

[RichardMH]

The main web site seems to be this

https://contentauthenticity.org/

I'm a club photographer and I think the provenance thing will happen for competitions there sooner rather than later. My club is discussing it and as most people use Photoshop there may not be much resistance.

Its in beta for Photoshop

https://helpx.adobe.com/au/creative-cloud/help/content-credentials.html

and available  in one of the recent Leica's for the documentary photographers

https://leica-camera.com/en-AU/photography/content-credentials

 

[PaulEC]

2 hours ago, Bartek said:

it makes no difference whether I use a purchased photo or AI-generated graphics

It makes a difference to the person who owns, and chooses to sell, the images! If the images are free the person who owns the images chooses to allow other people to use them.