Being the Director of AI/ML Technology at Arcurve, I am in an interesting position. While I live in the world of AI every day, I am also the parent of a daughter, Tori Morley, who is a budding musician. And because of this I know that artists, Tori included, are watching the rapid rise of generative AI – technology that produces or “generates” text, images, music, speech, code or video – with dismay.
The technology used in generative AI is incredibly useful and can be applied to solve many interesting and complex challenges that will benefit society. It gives the ability to create art in new and powerful ways that opens the creative process up to many more people. Despite the powerful new capabilities generative AI systems offer, I also don’t consider these technologies ‘intelligent’ – they are a new type of generative database. Unlike traditional databases, which are populated with structured or semi-structured data, these are trained with tremendous volumes of art in the medium of the desired output through annotation of the art with text. Thus, while these new types of generative AI are indeed impressive tools, they are still just tools. The AI has no more idea about its purpose or what it is being used for than a hammer does. It also doesn’t care what it is being applied to, nor does it care about results.
So, why then are these new generative AI tools generating so much consternation and controversy?
The issue, I believe, is the fact that the data used to train the models that are the foundation of these powerful systems have been derived from the creative works of tens of millions of artists, generally without their permission or consent. This is further exacerbated by the fact that while the accolades and associated Venture Capital funding are now being showered on the firms and organizations that created the generative AI systems, the artists whose life’s work made these systems possible are facing the very real fact that their livelihood is being threatened.
Naturally, the response from those who work in the technology field and those who providing funding for tech is to wave this off as just the most recent example of luddites sabotaging automated looms. While the comparison is apt in some ways with respect to the fact that new jobs emerge when old jobs are eliminated by technology, the existential threat that arises from generative AI is both more subtle and potentially insidious. And this is simply because of the way these new AI systems are trained to. Use existing artwork as a foundation for their models. The models would not work nearly as well as they do without the rich diversity of the human creative works, they have been trained with. By using people's creative works without compensating the artists, these models are reducing the opportunity for artists to grow their creative talents and ultimately succeed.
The irony, therefore, is that there is a very real risk to not only artists, but also to the AI models which are dependent on a continuous supply of creative work for their training. As opportunities for artists dwindle, so too will the rich ecosystem of new training data as human artists become extinct. Starved of new content, the models will start to collapse back in on themselves, resulting in a 'cultural grey goo of AI sameness'... and there will be no artists left to save us from the dull sameness of machine generated art…
Tori’s song, RageEhLittleBit, expresses one artist’s feelings on the subject of AI generative art and the situation facing not just artists, but humanity as a whole. Take a listen here.