How can we take into account the indirect environmental impacts of generative AI and fulfill environmental responsibilities in an era of climate crisis? At present, even identifying the environmental costs is difficult, as information such as carbon emissions generated during the development and deployment of generative AI is rarely disclosed, often under the pretext of corporate confidentiality. One starting point, therefore, is to demand greater transparency and the disclosure of such information. Beyond this, there is also a need for broader structural discussions about reinvesting the benefits generated by AI—within the AI industry and across society more generally—into efforts to address the climate crisis. Discrimination and Bias Even before the rise of generative AI, various AI and automated systems have reproduced existing biases in opaque ways. Generative AI models, which are trained on historical data, likewise tend to reproduce biases that reflect existing social power structures. For example, social biases that associate certain occupations or cultural contexts with particular genders, races, or social classes may appear in AI-generated content, potentially leading to unfair outcomes in areas such as hiring, content recommendation, or legal decision-making. In principle, generative AI should not be used in high-stakes decisions that have significant impacts on people’s lives, such as hiring or judicial rulings. The generation of hateful or stereotypical content that objectifies marginalized groups also constitutes a serious risk. 36 37