Posted on October 7, 2022
For anyone who keeps (even vaguely) abreast of advances in computing, it is obvious that artificial intelligence has evolved a lot in 20 years. Recent news shows that we are now a long way from the pathetic attempts of the time, which painfully made it possible to obtain from these programs little more than approximate character recognition, slightly unbalanced expert systems or computer analysis. images that weren’t wrong half the time.
The last few years have seen the appearance of computing monsters whose number of variables (counted in tens of billions, as for GPT3 for example) require suitable computer hardware, essentially derived from graphics cards: since neural networks can be modeled by vectors and matrices, which graphics cards routinely handle, it was logical that specializations of these cards appear for the particular case of developments in artificial intelligence.
Incidentally, this explains the performance – particularly financial – of companies specializing in the production of these cards, such as NVidia, which advertise on a regular basis the new versions of its cards intended for the massively parallel processing that artificial intelligence uses.
The last few years have therefore logically been filled with significant advances in the field, even if some headlines in the mainstream press suggest that the results leave something to be desired; thus, in 2016, an artificial intelligence from Microsoft, released on social networks to maintain open discussions there, had quickly turned into a fiasco by developing a comic penchant for a particularly relaxed Nazism.
The field of understanding and manipulating text has thus progressed enormously to the point of allowing current engines to handle complex concepts and contexts. Coupled with analysis and image creation engines, we are now able to produce images at will that respond to more or less precise descriptions. The engines that have appeared in the last two or three years allow for more and more amazing feats.
Let us mention in particular Slab (name coming from a pun between the artist Dali and Wall-Ethe creation of Pixar), which we can understand how it works and see its performance in the following video:
By the way, if you are interested in the state of the art in terms of image processing and artificial intelligence, the chain Two Minute Papers is probably the benchmark to follow.
In principle and without going into detail, this type of engine builds an image by having previously established a library of concepts and corresponding images by indexing and analyzing thousands of images, sometimes categorized and described, sometimes not. The understanding of the concepts is quite deep since these engines generally know how to use the style, the colors, the main characteristics of an artistic reference.
As such, MidJourneyanother artificial intelligence engine proceeding globally in the same way and now available through Discord, thus offers to produce composite images of different subjects at the user’s choice, and possibly, “in the style” of a chosen artist.
To take an example among the hundreds available now, a web developer had so much fun to ask MidJourney the city of Lyon in the style of different known painters (Dali, Van Gogh, Monet, Giger) or according to different contexts (apocalyptic, futuristic, etc.) which gives a series of particularly illustrative illustrations of the capacities of those kinds of engines.
We quickly understand that once reprocessed by an artist, the raw result can lead very quickly to a finished work of professional quality. In any case, this is what Jason Allen recently proved in an artistic competition conducted in Colorado, competition he won…
As one might expect, a lively debate ensued: should such an achievement be rewarded when it is, essentially, the result of an algorithmic process and not a human artistic creation? The debate also focused on the future of illustrators and graphic designers who, faced with the increasingly efficient achievements of these engines, are seeing their added value rapidly melt away.
This is not a purely theoretical question: even more recently, it was a certain Greg Rutkowski, a fairly well-known illustrator, particularly in video games, who noted that his style has become a reference more in demand than Picasso in the engine. Stable Spread. Thousands of images are thus generated by users of this engine, in the style of this painter… whose original images are gradually drowned by the surge of automatic productions.
In addition to the notoriety problems that this entails, there is also the question of respect for the initial works insofar as, in order to be able to produce images “in the style” of an artist, it was first necessary to provide the productions of this artist to the engine for analysis.
In addition, how much is it inspiration or plagiarism, how much is due to the algorithmic process, how much depends on the user’s ideas, his inputs at the program prompt, and is- sufficient to characterize an original work that protects against possible legal proceedings? In fact, there is no shortage of legal issues as artificial intelligence technologies develop in this way.
And these legal questions will quickly be coupled with societal and psychological questions when these algorithmic processes will gradually influence ever more professions, artistic then intellectual: what is currently happening with these specialized engines is an excellent illustration of what should happen quickly in a increasing number of professions since what can be done with images, can be done with music; there are already production tools, such as Aiva and there is no doubt that the next few years will offer concerts written almost entirely by artificial intelligence engines, in “the style of” one or another well-known composer and which will be indistinguishable from the reference artist…
And if we move away from the art, we should know that these engines currently produce automatic texts for the weather, sports results, stock market actions, and those repetitive articles that we find everywhere in the newspapers. Institutions like the washington postthe BBC, or Bloomberg already use these tools daily.
Unsurprisingly, the following stages (law or medicine are frequently cited examples) promise cries of outcry from the professionals concerned, cries that will join those of artists and journalists gradually dispossessed of their added value.
We understand here that the current transformation will be disproportionate to what has happened in the recent history of human inventions. The previous ones (steam engines, internal combustion engine, radio, electronics, etc.) reduced the cost of physical labor and then made it possible to reduce or eliminate the cost of distances (whether for the transfer of goods, people, or of information) ; the next inventions, based on artificial intelligence, will not only lower the marginal value of works of the mind: they will make it possible to purely and simply free ourselves from the human link in a growing number of professions.
The ability of our societies to adapt to this drastic change will undoubtedly be put to the test.
Moreover, it is surprising that the intellectual profession with the lowest added value is not already on the bridge to purely and simply ban these facetious algorithms: but as so often, politicians are unable to see the change when it arises. .
On the Web
We would like to thank the author of this post for this incredible content
Artificial intelligence attacking copyright
Visit our social media accounts and also other related pageshttps://www.ai-magazine.com/related-pages/