Learning Ethical AI

• In this repo are resources to learn how to implement ethical AI using machine learning.
• This repo mostly focuses on neural networks, it includes books, papers, talks, videos and tools
• Also, a major source of abuse is personal data harvesting.
  • The non-profit Center for Humane Technology was founded by ex-tech engineers to find ethical ways forward
  • They also have a list of actionable steps you can do to reduce this practice --> https://www.humanetech.com/take-control

Overview

What is Ethical AI (good overview) images shown above from this website --> https://devopedia.org/ethical-ai

• Related paper by Ajitesh Kumar --> https://dzone.com/articles/ethical-ai-lessons-from-google-ai-principles

Book / Papers

Listed (and some shown) below - also see the academic papers in /papers folder of this repo.

Algorithms

• 📖 Book: "Weapons of Math Destruction" by Cathy O'Neil - link
• 📖 Book: "The Alignment Problem: Machine Learning and Human Values" by Brian Christian- link (https://us.macmillan.com/books/9781250134769/theaddressbook)

Cognition

• 📖 Book: "Hello World - How to be human in the age of the machine" by Hannah Fry - link
• 📖 Book: "Calling Bullshit - The Art of Skepticism in a Data-Driven World" by by Bergstrom/West - link
• 📖 Book: "Black and White Thinking: The Burden of a Binary Brain in a Complex World" by Kevin Dutton - link
• 📖 Book: "Raising Heretics - Teaching Kids to Change the World" by Dr. Linda McIver - link

Bias

• 📖 Book: "Invisible Women" by Caroline Perez - link
• 📖 Book: "The Address Book: What Street Addresses Reveal About Identity, Race, Wealth, and Power" by Deirdre Mask - link
• 📖 Book: "Practical Fairness: Achieving Fair and Secure Data Models" by Aileen Nielsen - link
• 📚 Papers: Collection of Timnit Gebru's published papers - link

---

Guidance / Best Practices

Most major cloud vendors provide guidance and best practices for implementing ethical AI. I am most familiar with Google's guidance.

• ✨ Google's "People + AI Patterns" Guidebook - link
• ✨ Google's "Responsible AI Practices" - link
• ✨ AWS "Fairness and Explanability in AI" - link

In Healthcare

• 📚 Site: Coalition for Health AI (CHAI) - link

Talks / Videos

Authors of the books and papers listed above have also given talks on the focus of their writing. I prefer to read the book first, then watch the talk.

• 🗣️ Talk: "Weapons of Math Destruction" by Cathy O'Neil in 2016 / 58 min.- link
• 🗣️ Talk: "How I am fighting bias in AI" by Joy Buolamwini in 2017 / 9 min. - link
• 🗣️ Presentation: "Fairness and Explanability in Machine Learning" by AWS (shows SageMaker Clarify tool) in 2021 / 27 min. - link
• 📺 YouTube talk: "Ethical ML: Who's Afraid of the Black Box Models? • Prayson Daniel • GOTO 2021" / 38 min. - link
• 🎥 Documentary: "The Social Dilemma on Netflix in 2020 / 1 hour 30 - link

Groups / Open AI

ML Collective was born from Deep Collective, a research group founded by Jason Yosinski and Rosanne Liu at Uber AI Labs in 2017.

• They that group to foster open research collaboration and free sharing of ideas, and in 2020 we moved the group outside Uber and renamed it to MLC.
• Over the years they have aimed to build a culture of open, cross-institutional research collaboration among researchers of diverse and non-traditional backgrounds.
• Their weekly paper reading group, Deep Learning: Classics and Trends, has been running since 2018 and is open to the whole community.

ML Collective includes a 'Lab'. At the Lab, experienced researchers looking to dedicate time to mentor projects and give advice to starters should consider joining the lab, with a light commitment of joining our regular research meetings where research updates are presented.

• 🔬 More info about ML Collective Lab --> https://mlcollective.org/community/#lab
• 📺 YouTube channel for ML Collective --> https://www.youtube.com/c/MLCollective/videos

Tools for Data

Google has an extensive set of tools to evaluate bias in data used in models for AI. Many tools focus on data that will be used in TensorFlow models.

• 🔍 Google's Responsible AI - tools and practices - link
• 🔍 Data Card example - link
• ✏️ Datasheet Template - link
• 🔍 Know Your Data tool example (celebrity faces) -link
• 🔍 TensorFlow Data Validation tools (skew, drift, more...) - link
• 🔍 Pair Explorables, Measuring Diversity example -link
• 🔍 Pair Explorables, Hidden Bias example - link

Tools for Models

Google's model evaluation tools center around models built with TensorFlow. Other vendors also support open source tools for model evaluation and more.

• 🔎 Using MinDiff to do model remediation for TensorFlow - link
• 🔎 Model Card tool (provides context and transparency into a model's development and performance)- link
• 🔎 Example Model Card for face detection - link
• Open Source library 'InterpretML' to explain blackbox systems - link
• Open Source 'responsible AI toolbox' (from Microsoft) - link
• 🔎 Google's 'What If' tool for model understanding, faces examples - link - example image shown below.