Armchair Architects: What Is Responsible AI?

Responsible AI is a term that refers to the ethical and accountable use of artificial intelligence (AI) systems, such as large language models (LLMs), that can generate natural language responses to user queries. Responsible AI aims to ensure that AI systems are fair, transparent, reliable, and respectful of human values and rights.
In this episode, our #ArmchairArchitects discuss what responsible AI means in the context of large language models (LLMs) and how to avoid unintended consequences and harms, then suggest some concrete techniques and tools to control the inputs and outputs of LLMs, such as content safety filtering, JSON schema, and TypeChat.
You’ll hear the architects emphasize the importance of defining constraints and parameters for the LLMs and tracking the confidence intervals of the response, and how to fine tune the LLMs based on their own data sets and libraries and how to avoid bias and PII in the data.
Before watching this episode, you should first listen to The Danger Zone (Part 1) and The Danger Zone (Part 2) for context.
Resources
• Read the blog: Armchair Architects: What Is Responsible AI? techcommunity.microsoft.com/t...
• What is Azure OpenAI Service? learn.microsoft.com/azure/ai-...
• What is Responsible AI? learn.microsoft.com/azure/mac...
• Responsible and trusted AI learn.microsoft.com/azure/clo...
• Fundamentals of Responsible Generative AI learn.microsoft.com/training/...
• Introducing TypeChat microsoft.github.io/TypeChat/...
Related Episodes
• The Danger Zone (Part 1) aka.ms/azenable/146
• The Danger Zone (Part 2) aka.ms/azenable/147
• Watch more episodes in the Armchair Architects Series aka.ms/azenable/ArmchairArchi...
• Watch more episodes in the Well-Architected Series aka.ms/azenable/yt/wa-playlist
• Check out the Azure Enablement Show aka.ms/AzureEnablementShow
Chapters
0:00 Introduction
1:07 Genesis of Responsible AI
1:54 Increased Importance
2:34 Beware unintended consequences
4:26 Moving goalposts
5:12 Responsible AI is mandatory
5:45 What you can’t control
7:15 What you can control
8:15 Ways to use prompt engineering
8:42 From Explainable AI to Observable AI
9:49 The architect’s influence
11:31 Prefiltering and moderation
11:56 Control API access
12:33 Fine-tuning and customization