From the course: Learning XAI: Explainable Artificial Intelligence
Join today to access over 26,000 courses taught by industry experts.
Introduction to model constraints to control bias
From the course: Learning XAI: Explainable Artificial Intelligence
Introduction to model constraints to control bias
- Even with the best training approaches, models can still produce harmful or biased outputs in unexpected scenarios. This brings us to a crucial component of responsible AI, post-training constraints. Model constraints serve as guardrails that shape AI behaviors after training is complete. They represent a critical acknowledgement that we cannot anticipate every scenario during training, and that responsible AI requires ongoing governance throughout a system's operational life. So let's look at how these constraints connect to our broader ethical obligations in AI development. Post-training constraints generally fall into several categories, each addressing different aspects of ethical AI. To start us off are output filters. These analyze generated content before it reaches users, blocking or flagging potentially harmful material. These range from simple keyword block lists to sophisticated classifiers that detect subtle forms of bias or harmful content. While seemingly…
Download courses and learn on the go
Watch courses on your mobile device without an internet connection. Download courses using your iOS or Android LinkedIn Learning app.