How Learning to Lie with Data is Essential to Prevent AI being Sexist and Racist

Offering practical actionable support to data scientists who are making efforts to be responsible, while recognising why it is hard to do so.

The course materials are given on this GitHub page, with the course available here.

Course Summary

This course title "How learning to lie with data is essential to prevent AI from being sexist and racist." is intended to catch attention but also highlights the content of this course which intends to support data scientists looking to do responsible AI. The first part of the title comes from a book from 1954 titled "How to lie with statistics" which has been brought back into consciousness through another book "Rebooting AI". The first part of this course presents elements of how data can be misleading, while providing concrete tips to identify and address these data issues. The second part of the title refers to a series of recent scandals where it is argued that AI has not been used responsibly. These scandals, some of which are used as case studies in this course, are leading to the legislation coming in to ensure ethical uses of AI. The second half of this course is focussed on these ethical considerations needed for using AI responsibly. The course aims to support Data Scientists and their managers to increase their understanding of potential ethical challenges in the application of AI and provide concrete tips to support them to be responsible.

Who is it for?

This course is designed primarily for Data Scientists who are actively looking to be responsible in their work. Part of it is also intended to be appropriate for managers of data scientists or even their collaborators who may benefit from the broad discussions but skip some of the practical details.

Learning Objectives

By the end of this course, learners will have: * an awareness of some ethical considerations which are shaping the future of AI and why data scientists need to be responsible in their role. * been exposed to some common pitfalls where data mis-interpretation can arise and be presented with concrete advice to avoid them.

Learners may have: * gained practical experience working with data to draw correct conclusions in data containing complexities.

Course Details

1. Introduction

2. Data Considerations We have three approaches to consume the content in this section - A Content Approach, A Case Study Approach, A Practical Approach. All three approaches will cover the same case studies and content blocks, which are:

3. 2a- Content
   Module 1 - Definitions Matter
   Module 2 - Data Matters
   Module 3 - Variability Matters
   Module 4 - Interactions Matter
   

4. 2b- Case Studies
   i) COMPAS Case Study
   ii) Apple and Amazon Case Study
   iii) Ofqual Case Study
   iv) Protein Folding Case Study
   

5. 2c- Practical Approach
   Interactive example to consume the content using STACK.

6. Ethical Considerations

7. 3a- Introducing ethics in AI

8. 3b- Fairness and debiasing

9. 3c- AI ethics beyond debiasing

10. 3d- Accreditation

11. Conclusion

Behind the Course

This course was developed with The Alan Turing Institute and IDEMS International, in collaboration with partners from AI Ghana, Universitat Bonn, Center for Science and Thought, Zertifizierte KI, Lancaster University, and Caltech