The accuracy of the images generated depends hugely on how much noise it can cut off from the data

Artificial intelligence has come a long way in replacing human agency with automated mechanisms. The recent developments in the domain of art-making is only proof of its creative capabilities, so much so that it is being perceived as a threat to illustrator jobs. Illustrating is a creative vocation that in the commercial domain doesn't remain so, at least in certain areas. Well, it depends on what is the goal and the inputs used to generate the image. Earlier too, we have seen numerous AI image generators such as Anonimyzer, DeepDream, DeepAI, etc., but none have been even close to generating uncanny images like the ones generated by Open AI's Dalle E-2. Recently Google has released a new AI image generator that takes written commands to generate images. Imagen is considered the perfect solution for problems that put limits on imagination. Yes, digital designers understand pretty well what it is like to generate an image of two people sitting in front of a Paris tower. Does this mean it's time illustrators are obsolete and they should bid goodbye to their artistic pursuits? Not yet. There is still some time before this idea of doing away with illustrators becomes a reality. Reason: Limitations.

How do these models work?

Unlike the earlier image generating algorithms, these applications use Neural networks which churn images crunching huge sets of data relating to categorising text, and images and identifying the relation. The neural networks identify the pattern from the data fed, as to how an object should be manifested in a physical space. The accuracy of the images generated depends hugely on how much noise it can cut off from the data. Both DALL E2 and Imagen depend on diffusion models, which have proven to be effective to produce near-perfect images, semantic in their creation, wholly depending on textual inputs. So far so good. Considering the fact that we have been only able to have access to the images generated and not the applications, they need to be taken with an element of caution. As Dr. Hossein Malekmohamadi, the Senior Lecturer in Game Programming at De Montfort University, Leicester suspects, it is possible that the companies have released only the best of the pictures for public consumption.

AI models are only as good as the data they process

The models are trained with huge datasets like LAION-400M, an open data set filtered with OpenAI's CLIP, which scraps data from locations including unmoderated sites, a reason why the developers claim these algorithms are a bit off the intended purpose. Given the inherent bias the data carries within, the likely chance of it being exploited is immense. Precisely, because of this reason, as stated by Google, Imagen's access is restricted to developers. But it doesn't spell an end to the prospect of creating images out of mere text prompts. The biggies of the tech industry are taking prompt steps towards overcoming these hurdles. While OpenAI is planning to strictly implement a safety system by onboarding up to 1000 people every week, who should follow content moderation policies to access their application, Google intends to work on "vocabulary of potential harms". Besides, the artwork is always enriched by details, which DALL E-2 seems to fall short on. According to Mathematician Jeremy Kahn, the algorithm doesn't work for adding details like lighting or shadow, merging borders, or understanding binding attributes. This should sound like music to the designers bothered over AI-induced job insecurities!!

Final thoughts:

While the tech companies scratch their grey cells to find all viable means to overcome the challenges, it is utterly important to question if the AI-generated artwork be categorised as creative work at all. Creativity is very much a human predisposition and defines what it means to be human.  Is it possible for an algorithm to imagine a spiral staircase as whales' teeth? Perhaps not. Not until and even decades after the bias issues are resolved. If evolution is to be believed, then the way AI is integrating with human life has to be accepted. Therefore, it would be fair to say, AI art generators should be considered a source of inspiration and not an obstacle.