Learn more
Join Dexa
Published Sep 28, 2022

AI Today Podcast: Ethical & Responsible AI Series: Explainable and Interpretable AI

    Delve into the essence of explainable and interpretable AI, as hosts Kathleen Walch and Ron Schmelzer illuminate the vital techniques, ethical frameworks, and challenges necessary for enhancing transparency, trust, and ethical compliance in AI systems, particularly in crucial sectors like healthcare and finance.
    Episode Highlights
    AI Today Podcast logo

    Topics covered

    Popular Clips

    Episode Highlights

    • Challenges

      Explainable AI presents significant challenges, especially with black-box models like deep learning. Kathleen Walch emphasizes the need for transparency, questioning the acceptability of unexplained decisions, such as loan denials, by AI systems 1. Ron Schmelzer adds that without understanding these systems, trust is compromised, highlighting the importance of keeping humans in the loop 1. He notes that while algorithmic explainability is crucial, it doesn't solve all issues like bias or fairness 2.

      We humans want to know how things are done and how decisions are made.

      --- Ron Schmelzer

      Understanding these challenges is essential for developing trustworthy AI systems.

         

      Frameworks

      Frameworks for AI explainability focus on understandability and root cause explanations. Ron Schmelzer suggests that AI systems should provide human-understandable explanations for unexpected outcomes, even if algorithmic transparency is limited 3. He distinguishes between explanations and interpretations, stressing the need for interpretability when explainability is lacking 4.

      AI systems should use algorithms that provide a direct means to explain how outcomes were arrived from input data.

      --- Ron Schmelzer

      This approach ensures that AI actions are comprehensible to all stakeholders, enhancing trust and accountability.

         

      Ethics

      The ethical implications of AI explainability are profound, impacting trust and decision-making processes. Ron Schmelzer discusses the importance of ethical frameworks like CPMAI plus e, which ensure AI systems do not violate trust or cause harm 5. He illustrates the absurdity of unexplained AI decisions with a banking analogy, underscoring the necessity of transparency for trust 6.

      You don't want people not trusting your systems, you don't want people questioning things.

      --- Ron Schmelzer

      By focusing on ethical considerations, organizations can build AI systems that are both responsible and reliable.