From 'black box' to 'glass box': using Explainable Artificial Intelligence (XAI) to reduce opacity and address bias in algorithmic models

Resumo

Artificial intelligence (AI) has been extensively employed across various domains, with increasing social, ethical, and privacy implications. As their potential and applications expand, concerns arise about the reliability of AI systems, particularly those that use deep learning techniques that can make them true “black boxes”. Explainable artificial intelligence (XAI) aims to offer information that helps explain the predictive process of a given algorithmic model. This article examines the potential of XAI in elucidating algorithmic decisions and mitigating bias in AI systems. In the first stage of the work, the issue of AI fallibility and bias is discussed, emphasizing how opacity exacerbates these issues. The second part explores how XAI can enhance transparency, helping to combat algorithmic errors and biases. The article concludes that XAI can contribute to the identification of biases in algorithmic models, then it is suggested that the ability to “explain” should be a requirement for adopting AI systems in sensitive areas such as court decisions.