On the “creativity”​ of AI — Preliminary critical remarks

Praise for the creative capabilities of recent developments in AI technologies is ubiquitous in the media and relevant blogs.

Although really astonishing (as an example, have a look here [and the examples in the article]), creating novel ideas by using AI seems to have intrinsic limitations.

To us humans, these results sometimes seem really “creative”, “novel”, or surprising. However this is mainly due to the limitations in our own imagination, which is simply not capable of processing such huge amounts of data.

In essence, AI’s creativity is the result of hyper-complex processes of learning and adaptation that is based on an almost endless ocean of data/”knowledge” (without meaning). This has several implications concerning the underlying premises of such an AI-driven understanding of creativity and bringing forth novelty:

  1. These systems are based almost exclusively on already existing knowledge. Hence, their learning algorithms apply a strategy of learning from the past.
  2. This leads to a form of creativity that is grounded in the idea of (re-)combining already existing concepts/things. This is an accepted and valid strategy well known from creativity and innovation research. However, we have to keep in mind, that the results will remain in the realm of the predictable, or, from a Kuhnian perspective, within the paradigm of what already exists.
  3. It is a purely “brain/mind-based” form of creativity that does not take into account the world and its potentials (e.g., affordances) as a possible source of novelty (e.g., by interacting and engaging with it).

If we are interested in really “ground breaking”, radical, or disruptive innovations, these strategies will not suffice. As we show in our research, we will have to follow a strategy of Emergent Innovation, “Learning from the future and future potentials as they emerge” as well as acquire futures skills and a perspective on innovation that is grounded in an enactive understanding of cognition.

Will AI be able to sense the future by learning from it, its affordances and potentials, and from interacting with and enacting its environment?