The limits of AI’s creativity? Simple: us

An article in the journal Nature assesses a study by three Stanford University computer scientists who found that AI-generated ideas were judged as more novel than human expert ideas, while being judged slightly weaker on feasibility.

It’s now widely accepted that AI can manifest multiple forms of creativity, albeit in different ways to humans. AI is capable of generating original works of art, composing music, writing stories, designing new products and even, as the study proves, proposing scientific hypotheses. This creativity arises from basic processes such as pattern analysis and recognition, generative models capable of delivering results that remember but do not exactly replicate their training data, algorithmic variation and mutation processes, and style transfer processes that imitate, mix, and combine elements to result in something original.

Obviously, there are limits to the creativity of AI, because it lacks the intentionality, emotion, or personal experience that drive our imagination. AI creations are generated solely from data and training rules, not from personal perceptions or emotional expression, so they are limited by the quality, diversity, and breadth of the data used in their training: if an AI model lacks exposure to certain ideas or styles, it is not capable of inventing them on its own. Moreover, it lacks a sense of…