Recently, the digital turn entered a new era as some technological tools proved to represent a breakthrough in virtual and real-life practices. Though the idea of Artificial Intelligence (AI) has been circulating since the late 1950s (McCarthy et al., 1955) it has only been part of everyone’s lives for a few years, since users had the possibility to create language and multimedia contents from scratch. Nowadays the most promising application is probably Generative AI (GenAI), which generates “synthetic data that closely resemble real-world data” (Bandi, 2023: 1). Trained upon LLMs (Large Language Models) developed by huge stakeholders (Dao, 2023), AI is now closer to the idea of “human reasoning” as it generates contents based on human-induced prompts. Ethical issues are paramount to the discussion of AI outputs (Dubber et al., 2020; Boddington, 2023). As a matter of fact, GenAI creates plausible outputs, though they are not always true since they rely upon the synthetic gathering of corpus-based data, thus leading to hallucinations (Ji et al., 2023). Another critical issue involves gender-related depictions, showing a disparity that leads to a proper machine-induced bias (Leavy, 2018; Foka, 2024). Against this background, this paper aims at providing evidence from specific AI-powered generative tools that create realistic images from textual prompts, thus carrying out an intersemiotic translation process (Dusi, 2015). In particular, providing different easy-to-use GenAI platforms (including the newly-introduced Grok featured in Twitter/X) with ‘neutral’ prompts, the analysis would assess the level of possible equality (or inequality) in terms of gender, thus assessing the level of potential bias that may influence the perception of users in terms of visual narratives (Chen et al., 2024).
Fulfilling GEN-der AIms: do image-generating tools discriminate? An on-field study
Francesco Meledandri
2025-01-01
Abstract
Recently, the digital turn entered a new era as some technological tools proved to represent a breakthrough in virtual and real-life practices. Though the idea of Artificial Intelligence (AI) has been circulating since the late 1950s (McCarthy et al., 1955) it has only been part of everyone’s lives for a few years, since users had the possibility to create language and multimedia contents from scratch. Nowadays the most promising application is probably Generative AI (GenAI), which generates “synthetic data that closely resemble real-world data” (Bandi, 2023: 1). Trained upon LLMs (Large Language Models) developed by huge stakeholders (Dao, 2023), AI is now closer to the idea of “human reasoning” as it generates contents based on human-induced prompts. Ethical issues are paramount to the discussion of AI outputs (Dubber et al., 2020; Boddington, 2023). As a matter of fact, GenAI creates plausible outputs, though they are not always true since they rely upon the synthetic gathering of corpus-based data, thus leading to hallucinations (Ji et al., 2023). Another critical issue involves gender-related depictions, showing a disparity that leads to a proper machine-induced bias (Leavy, 2018; Foka, 2024). Against this background, this paper aims at providing evidence from specific AI-powered generative tools that create realistic images from textual prompts, thus carrying out an intersemiotic translation process (Dusi, 2015). In particular, providing different easy-to-use GenAI platforms (including the newly-introduced Grok featured in Twitter/X) with ‘neutral’ prompts, the analysis would assess the level of possible equality (or inequality) in terms of gender, thus assessing the level of potential bias that may influence the perception of users in terms of visual narratives (Chen et al., 2024).I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.
