Artificial intelligence (AI) is increasingly being used in critical decision-making processes across various sectors, including job selection, loan approvals, court sentencing, and medical diagnoses. However, concerns are growing that AI technology may perpetuate and even intensify social disparities, reflecting existing biases and inequalities present in the databases used to develop AI systems.
A recent case involving the testing of AI imagers by The Bulimia Project, an eating disorder awareness group, shed light on how preconceived data and values can contribute to social stratification and accentuate inequality. The results revealed that AI programs imagined a perfect physique predominantly featuring blonde women, women with brown eyes, and white skin. Similarly, the majority of perfect men had brown hair, brown eyes, and white skin. These idealized features often resembled those achieved through plastic surgery and fillers.
The problem lies in the development of AI systems by individuals who are unaware of the potential for inequality, as well as the use of these technologies by users who do not understand how they were created. For instance, algorithms used in job candidate selection are often trained based on previous resumes, which can perpetuate existing biases. If historically more men have been selected for senior positions while women are relegated to traditionally female-dominated roles, AI algorithms may automatically perpetuate and even intensify this disparity.
A well-known example of this occurred at Amazon, where an AI tool was implemented to assist the HR team in automating the candidate selection process. The system analyzed resumes and assigned each a rating based on past CV standards received by the company over a decade. However, as most resumes were from men, the tool discriminated against female candidates, reducing their chances of being considered. This was not due to the tool being inherently sexist but because it learned from biased data.
The repercussions of biased data extend beyond the hiring process. In the management of judicial sentences, AI can inadvertently contribute to racial injustices. The algorithms used to calculate incarceration time might treat black individuals more severely compared to white individuals. This racial bias can lead to the disproportionate incarceration of people of color.
Moreover, the biases present in AI systems often reflect the dominant viewpoint of society. For instance, chatbot programs may associate the invention of the airplane with the Wright brothers, even though other perspectives, such as those of Brazilians who credit Santos Dumont, exist. This uniformity of knowledge based on who develops the AI can suppress minority stories and conclusions, privileging the majority viewpoint.
Professors emphasize that programmers have an ethical and moral obligation when shaping AI systems with significant societal impact. Responsible AI development involves constantly evaluating, reevaluating, testing, and verifying models to identify and rectify errors. Understanding the data and its potential impact is crucial, as training AI is not solely about running algorithms but about developing a comprehensive comprehension of the implications.
In conclusion, as AI continues to rapidly expand, concerns regarding the perpetuation of prejudices and inequalities have arisen. The development and use of AI based on biased databases can contribute to the replication and even amplification of existing social disparities. Ethical considerations and responsible practices are essential for programmers and developers to ensure that AI technologies promote inclusivity and fairness rather than accentuate societal divisions.
