Typology of Risks of Generative Text-to-Image Models

Abstract

This paper investigates the direct risks and harms associated with modern text-to-image generative models, such as DALL-E and Midjourney, through a comprehensive literature review. While these models offer unprecedented capabilities for generating images, their development and use introduce new types of risk that require careful consideration. Our review reveals significant knowledge gaps concerning the understanding and treatment of these risks despite some already being addressed. We offer a taxonomy of risks across six key stakeholder groups, inclusive of unexplored issues, and suggest future research directions. We identify 22 distinct risk types, spanning issues from data bias to malicious use. The investigation presented here is intended to enhance the ongoing discourse on responsible model development and deployment. By highlighting previously overlooked risks and gaps, it aims to shape subsequent research and governance initiatives, guiding them toward the responsible, secure, and ethically conscious evolution of text-to-image models.CCS Concepts: • Human-centered computing → Human computer interaction (HCI); Text input; • Applied computing → Media arts; • Social and professional topics → User characteristics.

Ethical and methodological challenges in building morally informed AI systems

“…Fairness -Bias. Fairness is, by far, the most discussed issue in the literature, remaining a paramount concern especially in case of LLMs and text-to-image models 35,[45][46][47] . This is sparked by training data biases propagating into model outputs 48 , causing negative effects like stereotyping 33,35 , racism 49 , sexism 50 , ideological leanings 47 , or the marginalization of minorities 51 .…”

“…The emergence of generative AI raises issues regarding disruptions to existing copyright norms 25,52,80,93,142,146 . Frequently discussed in the literature are violations of copyright and intellectual property rights stemming from the unauthorized collection of text or image training data 45,53,77 . Another concern relates to generative models memorizing or plagiarizing copyrighted…”

How widespread use of generative AI for images and video can affect the environment and the science of ecology

“…There are direct environmental risks that are similar to those of text-based AI, in that AI-generated image and video will consume computing-related resources, leading to effects on energy and resource consumption and carbon emissions in addition to those of textbased models (Bender et al, 2021;Bird et al, Rillig et al, 2023), since they also make use of large language models as part of their model architecture.…”

“…Beyond the potential negative socio‐political implications of generative models (Bird et al., 2023; Weidinger et al., 2022) and legal concerns such as copyright infringement (Samuelson, 2023), the implications for environmental sciences and ecology are significant. Generative AI involves the use of machine learning approaches to generate new content (e.g., text, images, audio, or video) based on characteristics of training data and user input.…”

“…There are direct environmental risks that are similar to those of text‐based AI, in that AI‐generated image and video will consume computing‐related resources, leading to effects on energy and resource consumption and carbon emissions in addition to those of text‐based models (Bender et al., 2021; Bird et al., 2023; Rillig et al., 2023), since they also make use of large language models as part of their model architecture. Energy needs arise during model training, during inference (that is, queries made to the model), and also for server infrastructure, and corresponding carbon emissions depend on the carbon intensity of the energy source.…”