Differences

This shows you the differences between two versions of the page.

--- public:t-709-aies-2025:aies-2025:ai_in_practice [2025/09/24 08:27] – created leonard
+++ public:t-709-aies-2025:aies-2025:ai_in_practice [2025/09/24 08:33] (current) – leonard
@@ Line 22: / Line 22: @@
 ===== What Can Go Wrong in Practice? =====
-|  Tupe of Issue  |  Description  |  Example  |
+|  **Type of Issue**  |  **Description**  |  **Example**  |
 |  Bias and Discrimination  | Biased training data leads to biased outputs. Replicates and reinforces social inequalities | Hiring tools, predictive policing |
 |  Lack of Transparency  | Users don’t understand or cannot challenge AI decisions. Systems act as “black boxes” | Credit scoring, medical diagnosis |
@@ Line 43: / Line 43: @@
+===== Categories of AI Risk =====
+|  **Risk Category**  |  **Description**  |  **Real-Worls Example**  |
+|  Epistemic risk  | The system is wrong or misleading | Self-driving car misclassifies child as inanimate object |
+|  Moral risk  | The system does harm or violates values | Hiring AI favors privileged groups |
+|  Political risk  | The system shifts power, often unequally | Surveillance AI targets protesters. Data sold to authoritarian regimes |
+|  Social risk  | The system changes relationships or behaviors | Recommender AI increases polarization and echo chambers |
+|  Legal risk  | The system conflicts with the law or lacks legal clarity | Chatbot offers health advice that violates medical regulations |
+===== How Can We Prevent These Failures? =====
+|  **Preventive Action**  |  **Goal**  |
+|  Bias auditing  | Detect and address fairness issues in data and model outcomes |
+|  Explainability features  | Help users understand and contest AI decisions |
+|  Ethical impact assessments  | Identify potential harms and affected groups before deployment |
+|  Stakeholder inclusion  | Bring real-world users into design discussions |
+|  Transparency reports  | Show how AI is trained, tested, and used |
+|  Independent testing  | Test systems in realistic, high-variance environments |
+|  Fail-safe design  | Build human override, logging, and rollback capabilities |
+===== Takeaways =====
+  * AI systems fail not only due to bugs, but also due to ethical blind spots.
+  * What works in lab conditions can be harmful in society.
+  * AI failures often reinforce existing social inequalities.
+  * Most failures are predictable and therefore preventable.
+  * Ethical foresight is part of responsible design, not idealism.