What are the limits of artificial creativity?

[suchona.kani.z]

Creative artificial intelligences come very close to human creativity in some ways. However, what fundamentally distinguishes them from humans are physical aspects. Humans learn not only mental skills, but also physical ones that they use for their creativity. An AI that creates art has never felt the pressure of a brush on paper and cannot grasp the dynamics in the way that the human cognitive system can. These are experiences that the AI ​​cannot bring to the creative process at this point.

Furthermore, it is important to note that while AI can produce valuable works of art, it can also produce works that are either not novel or are perceived as meaningless by humans. The meaningfulness of generated artistic artifacts is often a controversial issue, even for works of art created by humans. Thus, AI, like humans, faces the challenge of finding the balance lithuania consumer email list between novelty and value. This is currently an even bigger problem for AI than for human artists.

In 1926, Graham Wallas divided the creative process into the following phases: preparation (gathering information), definition (definition of the problem), incubation (thinking about ideas), illumination (presentation of a solution) and verification (testing the quality of a solution). The capabilities of AI are currently limited to the process steps preparation and incubation. The other steps still require at least some human intervention.

Currently, the most powerful AI systems are still based on supervised learning, where data is already available that has been labeled as truthful by humans. The learning process consists of calculating an average error or loss over the known examples and making adjustments to minimize this loss. Unusual examples (outliers) are often given little influence to avoid distorting the overall error. However, the goal of generating novelty in the creative process is counteracted when such unusual data points are given little weight.