Explainable Artificial Intelligence (XAI)

The growing use of artificial intelligence models in all industry sectors has led to the emergence of the Explainable Artificial Intelligence discipline, also known as XAI, which seeks to ensure the interpretability and transparency of artificial intelligence models. Regulation in this area is growing, and includes the AI Act in Europe, the AI Bill of Rights in the United States, and data protection regulations in several regions, giving citizens a "right to an explanation" if they are subjected to Artificial Intelligence-based decisions.

This transformational change in the way AI models are used has a profound impact on business and requires adapting models, systems, data and processes to ensure models are transparent, can be interpreted, and are used ethically.

Management Solutions has extensive experience in the implementation of interpretable artificial intelligence models (white boxes) and in the use of post-hoc interpretability techniques for difficult-to-interpret models (black boxes) in major banks and non-financial corporations in Europe and the America, having also worked closely with some banking supervisors to ensure these models can be interpreted.

Our value proposition is focused on supporting Data Science, Risk, Compliance and Technology areas in four lines of work necessary to ensure the interpretability and transparency of artificial intelligence models:

• Developing artificial intelligence models that are intrinsically explainable and interpretable, or implementing interpretation modules in existing models.
• Providing support in adapting the model risk framework, helping clients to adapt policies, inventories and validation frameworks to AI models.
• Transforming the operational aspects of XAI, such as incorporating XAI elements into the data framework, producing interpretation processes, and reviewing technical documentation templates.
• Finally, implementing ModelCraft as a proprietary AutoML system that incorporates an interpretability module, and Gamma™ as a model governance system.