"AI's Lethal Weakness: Why Discrimination Remains an Unsolved Issue"

Recently, a study conducted by the Service Management Research Center (ZMM) at DHBW Stuttgart examined the ability of artificial intelligence (AI) to identify discriminatory content in images and advertisements. The study revealed impressive progress in this field, as well as significant limitations.

In this comprehensive study, AI was exposed to various images and advertisements and tasked with evaluating them. This included 60 advertisements recently criticized by the German Advertising Council. The results demonstrated AI's remarkable ability to identify discrimination in advertisements. AI evaluated most of the criticized advertisements as potentially discriminatory, while non-discriminatory advertisements rarely received such warnings.

Considering that just a decade ago, AI faced significant challenges in correctly classifying depicted objects in images, this progress is particularly impressive. Thanks to millions of images, AI has learned to accurately distinguish between dogs and cats. The rapid advancement of AI raises a series of questions: What are its current limitations? "We wanted to know to what extent AI can identify discriminatory behavior by simply showing it an advertisement and asking for evaluation," explained student Helen Beckers, who conducted the study.

Given the significant increase in algorithmic discrimination, this is particularly important, as algorithms may disadvantage certain individuals based on gender, religion, ideology, racism, or background. Algorithmic discrimination has become a serious issue affecting various areas such as job applications, credit allocation, healthcare, and predicting recidivism rates. Therefore, addressing this issue is becoming increasingly crucial.

The discovery that ChatGPT can detect gender bias and stereotypes is particularly revealing. AI is also capable of recognizing discrimination in reverse situations through advertisements that swap gender roles. "The different evaluations by ChatGPT in these two scenarios highlight its ability to identify discrimination even in reverse situations," shared student Sven Peter, who participated in the research.

However, AI reaches its limits in identifying other forms of discrimination, such as objectification, disrespect, and abuse of power. "The results of this study emphasize the need for further development of AI systems to effectively identify and prevent discrimination," reported student Marius Funk. The study raises important questions about how AI technology can be used in the future to combat discrimination and promote equality in various domains.