Notes
Algorithmic Bias in Creative Artificial Intelligence?
110According to the machine heuristic, humans often perceive technology, like artificial intelligence (AI) to be objective and unbiased. In contrast, decisions of AI used in various contexts are prone to algorithmic bias: a systematic bias in outputs of algorithms, which leads to a benefit or disadvantage of groups or individuals. It occurs due to bias in society that results in biased data the AI is trained with. We focus on potential algorithmic bias in creative AI, a phenomenon, which has received a lot of attention in recent months. AI is now able to generate creative content like pictures and stories from short textual prompts given by users (e.g., ChatGPT, DALL-E2, or Midjourney). While critics highlight the potential danger of algorithmic bias in these frequently used creative tools, programmers could also take precautions to increase equality and diversity in AI-generated content. However, scientific reports, and systematic analyses on how AI-generated creative content reflects bias are scarce. Hence, we first address the question whether content generated by creative AI is impacted by an algorithmic bias. We conduct content analyses to detect effects of algorithmic bias in several AI systems and for different types of discrimination (e.g., gender discrimination, race discrimination). In addition, we consider reactions to algorithmic bias in creative AI. Users’ reaction to creative content is influenced by artist information and a lack of mental capacities attributed to AI. In other fields of application, the source (human or AI) of discrimination has been shown to influence perceived objectivity and moral outrage. In a second study, we examine the effect of artist information on users’ perception of creative content, and on their response, when they learn that the content is biased or diversified, in a 2 (artist information: human vs. AI) x 2 (content: biased vs. diverse) design.
111Artificial intelligence / algorithmic bias / creative AI / machine heuristic