AI Art?

Ah, October, the oddly named tenth month of the year. Probably the month where the on-rush of inertia towards the end of this year and the start of the next is the strongest. After the last entry involving tiny fonts, I didn't really feel a need to say anything. Until now.

The recent release of DALL·E for unfettered public use combined with how an artist winning top honours at a fine arts competition with AI generated image has fanned flames of arguments for and against the recognition of AI generated images as art.

I won't list the arguments down here, but instead share my perspective on this and on related aspects of this in general.

See, scientific and technological progress leaves an indelible mark on the overall psyche of the human race. In many cases, the belief that something is possible has proven to be enough of an impetus to drive someone to make that belief a reality. If they can explain it, we get new science. If they can replicate it consistently, we get new technology. In that sense, the progress of science and technology cannot be halted.

In the case of having faster computational hardware, or the ability to make more money with less labour in the world of manufacturing and business, almost everyone is in agreement that such progress is beneficial. And the near optimism of the ingenuity of humanity to solve problems in the future (especially the problems that the solutions of today have created) creates this self-feeding and self-expanding loop of advancement.

Due to the generally dehumanised aspects of manufacturing, business, and computation, most people do not see a threat of sorts. There were once those who fought against the rise of the machines in industrialisation, but their fight was more about how the former labourers were ill-treated in the rise of higher productivity tools than how the tools were of higher productivity, despite their group (the ``Luddites'') being used these days to mean an opposition to industrialisation/automation.

So, why the brouhaha over AI generated images as art?

It's not just an issue about labour the way the original Luddites were fighting against (though I suspect that this will eventually become the real argument in time to come), but it seems to, at some level, impinge upon what some claim to be a quality of what is essentially human---art. ``Art'' is a loaded term, and with the post-modern movement being a thing, is even more loaded than ever before.

At the risk of oversimplification, ``art'' can be thought of as the outcome from the sum total of an expression of qualities that have not been adequately expressed/explained in a quantitative manner that permits a consistently deterministic and predictable outcome. You know, the thematic opposite of ``science''.

Art is older than science---we have been expressing/creating things long before we knew how to characterise, explain, and replicate them at scale. Despite all the new gadgets that we have created over time, art still exists, though to be fair, art has also evolved.

Evolved to make use of the newest gadgets to assist in the expression.

Finger painting was among the first forms of visual art. Then different pigments were discovered, and other media to use them were also created. Art evolved from depicting the past to depicting the future, from symbolisms of reality to abstractions to hyper-realism, from the still to the animated to full multimedia extravaganza---you name it, art has and will continue to explore it.

AI generated images is just another gadget that is created that incumbents have not gotten used to. After all, no one seems to bat an eyelid when the ``clone stamp'' is used in art, or even more dumbly, when three-point perspective is used---both of the cited examples are gadgets/technological tools that artists have accepted as part of their toolkit of expression.

AI generated images are going to be that, eventually. There is no running away from it, and I am sure that many artists know this.

But what they are angry about is the same as what the Luddites are angry about back in the day: the ill-treatment of the specialists of the art by the Johnny-come-lately wielders of the latest geegaw brashly proclaiming deceptively that they were also artists without acknowledging that their apparent powers came about through the use of a new high-productivity tool as opposed to walking the hard path that the specialists of old used to get to where they were.

Thus, it is about the deception that are ruffling the feathers of the naysayers against AI generated images as art, and not necessarily the AI itself. There are some arguments about how the AI model is somehow stealing intellectual property through amassing the images of artists to train from, but I do not see how this argument can be fought comfortably, considering that even the regular training of a human artist involves studying the artwork of other artists.

As someone from tech-land, I want to point out that the representation of something does not make it that something, i.e. ``a map is not the territory''. If the representation used in the AI model is a literal copy of the image, then there are strong grounds for how the AI model is truly a copying-plagiarist. However, if the representation used is an abstracted form of the image, then the argument is severely weakened---even in the case of human-land, when an artist is sued for copyright infringement through what is claimed to be a derivative work, the burden of proof for the actual copying aspect is subtle enough that a lawsuit in front of a judge is needed to sort it out, because factors other than the representation in the head of the allegedly infringing artist will come into play to determine culpability.

There may be strong arguments about attribution of the training data source images, especially if they are digital in nature, mostly because in the land of the discrete, a map of the territory is the cheapest when it can be a literal copy of the territory itself. For the confused, this means that if I want a ``map'' (i.e. representation) of the digital ``territory'' (say an image, or a video, or even a plain text document), it will cost me negligible amounts to simply copy bit-for-bit of the digital ``territory'' than to re-create a ``map'' of it (through transcoding for example).

Other than that though, it is probably best to dissuade deception by making it clear when a work is made from these AI tools, at least until their use are so ubiquitous that it seems a bit silly to do so (no one ever declares their use of three-point perspective for example).

And when will that be? Who knows... I've written a fairly long and rambling post, and it is time to head back into Minecraft to prepare the perimeter for my traditional stone brick castle.

Till the next update then.