Explainable AI Crucial for EU Compliance, Says Telekom MMS' Kira Vinogradova
Telekom MMS' Kira Vinogradova, an Advanced Consultant for AI, Computer Vision, and GenAI, discusses the growing importance of Explainable AI (XAI) in today's increasingly complex AI landscape. As AI systems take over more speech- and text-based tasks, ensuring transparency and traceability is crucial, especially with the upcoming EU AI Act in 2026.
Vinogradova explains that implementing XAI involves several steps. First, existing AI systems must be analyzed. Then, XAI methods like LIME, which helps identify key words or phrases contributing to classification, are integrated. Ensuring user-friendly design and training employees are also vital. Continuous optimization and data protection, including protecting high-risk systems from manipulations, are essential. Companies must also document training data sources and identify AI-generated content.
The EU AI Act, set to be binding from 2026, defines risk categories and sets transparency requirements for high-risk systems. This regulation aims to address concerns about transparency and traceability in AI systems. For Large Language Models (LLMs) and chatbots, XAI approaches like Attention Visualization, counterfactual explanations, and the 'Chain-of-Thought' method help understand model decision-making processes, enhance trust, and aid error analysis.
With the increasing complexity of AI and the upcoming EU AI Act, the importance of Explainable AI will continue to grow. Kira Vinogradova emphasizes that XAI is crucial for making AI decisions understandable and traceable, fostering user trust, and facilitating error analysis. Companies must adapt their AI strategies to comply with these regulations and ensure transparency in their AI systems.
