Ethics in ML#

About this lecture#

You are part of it!#

This lecture will be delivered as a flipped-classroom.

What is a flipped-classroom?

A flipped-classroom is a pedagogical strategy consisting of reversing the traditional teaching scheme: instead of learning in class and practising at home, students are given material to study themselves beforehand and the in-class time is dedicated for debates, discussions and problem solving. This approach gives more autonomy to the students, enhances self-learning skills and promotes active participation and expression.

If you are curious and want to learn more, the Wikipedia article gives a good overview and describes some examples of flipped-classroom by field of study.

You will be asked to read some of the online articles listed in the Section Reading Material below. During the class we will focus on discussions and debates about each case study. Then we will extract the key terms and synthetize together to reach conclusive remarks about ethics in machine learning.

The lecture will end with a graded short essay.

More information is given in the following section.

Schedule#

Part One: What is ethics? What is ethics in machine learning?
At home, you will think of a definition of “ethics in machine learning” in your own words. We will start the lecture by sharing our findings and will agree on a definition together.

Part Two: Case Study
In groups, you will discuss a given case study according to the points provided in the Section Guidance below.

Part Three: Presentations
Each group will present the outcomes of their discussion on the case study in an oral presentation.

Part Four: Synthesis
As a class, we will extract together the key elements from the case studies and use them to draw conclusive remarks.

Part Five: Essay (graded)
After a deserved break, you will be presented three questions dealing with ethics in machine learning. Choose the one you prefer and write a short essay of minimum 300 words. In your essay you should clearly articulate your answer, bring a nuanced argumentation considering the different perspectives and illustrate as much as possible your arguments with examples.

Reading Material#

There is no reading material for part one (definition exercise) yet some hints. Wikipedia is also a good place to start. The UN agency UNESCO has written on the topic.

For the case studies: read at least the material for one. But if you have time, try to read another case study.

Bring Your Own Case Study

If you are already captivated by a tool employing machine learning and are curious to explore its ethical aspects and implications, contact the instructor as soon as possible (before the third week). If there is enough interest to form a group, an additional case study could be added to the debate.

Case Study: Timnit Gebru on societal implications of natural language processing (NLP) technology#

Timnit Gebru: The Computer Scientist Fighting for a Fairer World
Article from historyofdatascience.com

On the Dangers of Stochastic Parrots: Can Language Models Be Too Big? 🦜
Paper.

Behind the Paper That Led to a Google Researcher’s Firing
Article from www.wired.com

(Optional) What Really Happened When Google Ousted Timnit Gebru
(very long) article from www.wired.com

Case Study: Facebook#

How Facebook got addicted to spreading misinformation
Article on www.technologyreview.com

Shoshana Zuboff: Surveillance Capitalism and Democracy
YouTube Video.

Case Study: AI Weaponry#

UN fails to agree on ‘killer robot’ ban as nations pour billions into autonomous weapons research
Article from theconversation.com

What you need to know about autonomous weapons
Article from International Committee of the Red Cross (ICRC)

AI-Influenced Weapons Need Better Regulation
Article from Scientific American

Guidance#

These questions will help you extract the important information from the material and anticipate the discussion.

  • What are the different agents/actors? (think of professions and also societal entities)

  • How impartial can instances such as companies or organizations be?

A recommended framework to analyse each case study would be a table with:

  • the type(s) of machine learning implementation(s) and/or algorithms

  • the domain(s) (medical, language, internet, data science)

  • the recipients/beneficiaries

  • the good: how is it beneficial for the recipients/users

  • the harms: what are the ethical issues?

  • the solutions: how to tackle and address those issues?

More (extra, not needed for the class)#

If you are curious about ethics in AI, or even think of pursuing this topic professionally, find below some articles, organizations and companies that specialize in ethics in AI:

Courses#

Articles#

Organisations#

  • AI for Good: digital platform of the United Nations for learning, discussing, and connecting in order to identify practical AI solutions that can help advance the UN’s Sustainable Development Goals.

  • Center for Human-Compatible Artificial Intelligence (CHAI): multi-institution research group based at UC Berkeley, USA, focusing on advanced artificial intelligence safety methods.

  • Montreal AI Ethics Institute: international non-profit organization democratizing AI ethics literacy.

Companies#

  • AlgorithmWatch: non-profit research and advocacy organization that is committed to watch, unpack and analyze automated decision-making (ADM) systems and their impact on society.

  • DataKind. Global nonprofit that harnesses the power of data science and AI in the service of humanity.

  • Ethically Aligned AI: an example of a company providing training, consulting and public speaking about ethics in AI.