AI systems don’t fail silently, they fail creatively.

When data quality breaks down, models don’t just become inaccurate, they become unpredictable, biased, and operationally dangerous.

This session examines the fundamental effects of poor data quality on modern AI systems. It discusses how issues such as pipeline instability and missing context influence learning behavior during training, and why successful execution does not imply meaningful learning. Special attention is given to how data quality problems systematically amplify hallucinations, bias, and false confidence, particularly when AI systems are applied in business-critical contexts.

In the second part, the perspective shifts from AI as a victim of bad data to AI as a tool for improving data quality. The session shows how AI can be embedded into data engineering and data quality workflows to support profiling, semantic validation, and anomaly detection. We discuss how this changes the way pipelines, governance, and monitoring are designed when AI becomes part of the data engineering loop.

This talk focuses on control and trust in AI-driven systems. It helps practitioners understand where and why AI fails when data quality is insufficient, and how AI can be applied responsibly to improve data quality rather than undermine it.