A framework for the analysis of AI creativity
Creativity is some of the distinctive and helpful traits of human beings. It permits us to supply authentic and significant artistic endeavors, science, and tradition. But can Artificial Intelligence (AI) ever match or surpass human creativity?
A current paper by Wang et al. (2024) presents a new framework for exploring and assessing AI creativity by introducing the ideas of Relative and Statistical Creativity. It additionally proposes a sensible measure of Statistical Creativity for autoregressive fashions, i.e. fashions that generate textual content with next-token prediction, as accomplished by many Large Language Models (LLMs).
In this text, I’ll briefly clarify the principle concepts and contributions of the paper, and illustrate why they’re essential for the way forward for AI analysis and improvement.
Traditionally, human creativity has been studied and outlined in varied methods by completely different disciplines, comparable to psychology, philosophy, and cognitive science. However, there isn’t any consensus on what constitutes creativity, and the right way to measure it objectively. Different standards, comparable to novelty, high quality, worth, and shock, have been proposed, however they’re usually subjective and context-dependent.
For instance, a tune that mixes genres, languages, or devices could also be seen as inventive by one group who like its range, fusion, or innovation, however others might discover it discordant, chaotic, or pretentious.
This lack of settlement on the definition of creativity poses a problem for evaluating the creativity of AI programs. How can we evaluate the inventive skills of machines and people, if we can’t agree on what creativity means?
To handle this difficulty, the paper proposes a brand new idea known as Relative Creativity. Instead of attempting to outline creativity universally, the paper shifts the main focus as to if AI can match the inventive skills of a hypothetical human. This perspective attracts inspiration from the Turing Test, which assesses the intelligence of machines by evaluating their habits with human responses. The thought of Relative Creativity is that an AI mannequin is deemed as inventive as a hypothetical human…
