Ethical AI In Practice - Tapa blanda

Sinopsis

When the algorithm is wrong, who pays the price? Move beyond high-level principles to real-world, documentary-style case studies in responsible artificial intelligence.
AI systems arrive with polished confidence and clean dashboards. Yet behind the scenes, models hallucinate fake legal precedents, recruitment algorithms reproduce decades of workplace bias, and privacy breaks in the smallest everyday moments.
In Ethical AI In Practice , author Neha Rani cuts through high-level tech jargon to trace the exact evidence trail from system failure to practical fix. Using eight documentary-style case studies, this book breaks down the real mechanics of modern machine learning and provides clear, actionable frameworks for developers, tech leaders, data scientists, and policy decision-makers.
Inside this practical guide, you will explore:

• The Claim-Check Ladder: How to separate fluent language generation from verifiable fact and protect systems from plausible hallucinations.
• Hidden Data Bias & The Skew Radar: Techniques to uncover proxy variables, selection bias, and historical skew buried inside ordinary corporate spreadsheets.
• The Data Gravity Map: How everyday prompts and metadata cross privacy boundaries and transform casual text into permanent digital records.
• The Intent Firewall Protocol: Strategies to defend tool-using AI agents against prompt injection, jailbreaks, and indirect instruction attacks.
• The Evidence-to-Reason Bridge: How to evaluate machine explainability (LIME/SHAP) and avoid mistaking statistical correlation for real justification.
• Shared Accountability & The RACI for Risks: Defining exact liability across model developers, system deployers, and deploying organizations.
• The Safety-Utility Budget: Balancing latency, infrastructure costs, human oversight, and safety checks under real production loads.
• The Trust Continuity Loop: Maintaining long-term ethical performance across retraining cycles, data drift, and continuous software updates.

"Sinopsis" puede pertenecer a otra edición de este libro.