Algorithms won’t replace artists (yet): ‘AI seems intelligent, but it isn't’

This academic year, Toledo introduced a new background page, created by Herman Moens, an employee at KU Leuven. In a press release, Moens clarified that artificial intelligence (AI) created the work. However, the advance of AI in the art sector faces significant opposition from advocacy groups and artists.

The strike by the AWG also focused on the negative effects of AI on writers. It now counts as the longest post-war strike of writers and screenwriters. While some within the American TV industry worry about the pay and job security of screenwriters, others view AI as an innovative artistic medium that can open doors to a democratized art world.

AI training: practice makes perfect

‘In the visual arts, AI can, roughly speaking, be used by artists in two ways to create an original work’, says Jerry Galle, lecturer in Fine Arts at KASK and Howest. ‘Artists can either train an AI system themselves by creating or selecting input images, or they can use AI in the post-production phase.’

‘So artists themselves can train an AI system by selecting data that the system processes as an input source. The originality then lies in the artist developing the conceptual or thematic framework to give the work a certain meaning.’

For instance, Galle himself employed an AI system with memes to produce innovative images. He chose memes because they often consist of low-quality images that exhibit high variability, forming a common base that anyone can utilise. 

Musician Dorian van Heede was already experimenting with the possibilities of AI art. In 2022, he participated in the AI Song Festival with three other engineers under the banner of Beatroots AI. They presented Song of the Machines, a song partially created by AI. ‘We employed a diverse array of tools,’ he explains. ‘Training AI systems entirely from scratch demands substantial time, computational power, and financial resources. Lacking those, we creatively worked with our available resources. Jukebox, an AI model we occasionally use, permits us to infuse our unique sound into AI-generated music by employing our own audio files,’ details the AI musician. 

Galle recommends that artists wishing to work with AI utilize open-source models – AI whose source code is publicly available and therefore free to use and easily adaptable. This approach provides control over input data and the context of AI system training. 

‘After all, we cannot know how Big Tech companies acquire their data, and frequently, the circumstances under which those systems were trained remain unknown. Often, such training is outsourced, followed by unethical AI image screening in sweatshops', Galle points out.

Text Generation

Another application of AI in the art sector is text generation. ‘Text generation is a technique that individuals within Big Tech typically don’t fully comprehend. This AI operates by drawing from an extensive pool of words and phrases found in existing texts, which serve as the basis for its training.’

University students too are trained using existing knowledge and data, however, they do so to be able to apply critical and original thought, something which AI is not capable of.' ‘AI may seem intelligent, but it is not’, says Matthias de Ridder, President of the Vlaamse Auteursvereniging.

The lack of transparency regarding which input data has been consulted can be problematic in terms of copyright rights, according to de Ridder. ‘As an author, it is difficult to prove that an AI text is based on your original text. But we do know that programs like ChatGPT are trained with – and therefore use – copyrighted texts,’ says de Ridder. 

From copy-paste to copyright

De Ridder also believes that more transparency is needed. He is therefore a proponent of the recent proposal by the European Writers’ Council to work on a certification mark for publishers and authors, indicating whether a book was written by AI or not. 

Furthermore, de Ridder is cautious about granting copyright to AI-generated works: ‘It would be the wrong approach to assert copyright against authors for books written by an AI system that are largely based on their work.’

Van den Heede also acknowledges that AI in the music industry sometimes skates on thin ice in terms of copyright. ‘It is a bit of a grey area sometimes. AI can generate samples that might just be copies, but you could say that Nirvana did the same with a song by Killing Joke (in their song ‘Come As You Are’, red.).

‘Now suppose I use AI to make a Dr. Dre-like song in his style and with his voice. If you do that for profit, it’s not acceptable. If you however, divide the reproduction over thousands or millions of artists, every artist is only in it for one more.’ Regarding compensation for the use of authors’ works as training data, de Ridder sees potential in the system currently used in the academic world: ‘Universities pay a certain amount of coping compensation to Reprobel (a private institution responsible for the collection and distribution of some copyright fees).’

‘That compensation can indirectly benefit authors, allowing them to receive a certain percentage. Since Big Tech companies use large amounts of copyrighted material to train their AI, every author should share in a certain percentage of compensation paid by those companies,’ explains de Ridder.

Sector seeks AI, artists seek work?

Despite the rapid evolution of AI, de Ridder and Galle believe that human artists will not yet be fully replaced in the art sector. ‘AI still does not offer an adequate replacement for human writers; it cannot produce creative works of comparable quality.’

‘We do see that translation computers are being used more frequently for technical translations of manuals and instructions. The errors made by the computer must afterwards still be corrected manually. However, creative translations are not yet at risk,’ reassures de Ridder.

'I also don't see authors being replaced for scriptwriting just yet. On the other hand, there is more demand for spin-offs. For example, if Vikings becomes a big hit, Netflix might want a similar series, so there is more content on that theme that people will watch as well. It's a possible scenario that AI will write spin-offs, but that kind of content is merely "derivative of".'

‘Potentially, text- and scriptwriters could receive proposals for mere improvement tasks with the risk that their pay rates would decrease’, de Ridder considers. 

According to Galle, the rise of AI can also facilitate a shift in the visual arts in the sale and trade of artworks, including the use of non-fungible tokens (NFTs) – a way to link ownership to digital objects. ‘This is not about those NFT caricatures of primates, but about understanding the origin of the image and incorporating that into your contract,’ says Galle.

Fair AI art as a desirable future

‘The ego of an artist, which often takes centre stage, is now increasingly giving way to a more democratized art. Open AI can democratize art because people can create art themselves without relying on expensive software or institutional resources. One possible downside is that the aesthetics of artworks may be diluted, but I have confidence that artists can infuse enough inspiration into their AI art,’ says Galle. 

Furthermore, according to Galle, art schools can facilitate the transition to a renewed art sector: ‘I don’t think all art students will work with AI, but we can give them the history of AI and make them aware of how to make open AI models more unique by training data themselves.’

In conclusion, Van den Heede also sees the introduction of AI in the art world as ‘a technological evolution that we must embrace, but fairly and equitably.’ Similarly, De Ridder concludes: ‘The technology is unstoppable, and useful things can be done with it, but consideration must be given to the intellectual work done by people.’