๐Ÿ’ป Week 08 - Class Roadmap (90 min)

2024/25 Autumn Term

Author

The DS101A Team

Published

22 November 2024

Recommendation systems and their ethical challenges

Welcome to our week 08 seminar/lab class for DS101A.

We will discuss recommendation systems (RS) โ€“ and consider the ethical challenges raised by these systems, as one example of an application of unsupervised machine learning.

Preparation

To prepare for the class, you can watch a short clip from this video, where Lawrence Lessig (speaking at the LSE very recently) talks about the consensus built around television broadcasting, and how recent technology has transformed society into a โ€œpost-broadcast democracyโ€, and where the impact of recommendation systems has been to create islands of (shared) perspective across society.

He also talks about โ€œnews desertsโ€ โ€“ where โ€œnewsโ€ sources online (sometimes from adversaries) are able to fill-in information-gaps, due to the lack of any local regional newspapers. Recommendation systems are often used to provide news online.

It should take around 15 minutes of your time.

In the class we will discuss:

Part 0: Technical approach

  • What unsupervised learning is
  • What different techniques are used to build recommendation systems

Part 1: Big questions

  • What constitutes a โ€œgoodโ€ recommendation
  • Who gets to decide what a โ€œgoodโ€ recommendation
  • Whether algorithmically โ€œgoodโ€ recommendations are always moral

Part 2: Ethical considerations

  • The ethical issues we need to consider when building recommendation systems
  • The effect of different levels of abstraction

Part 3: Legal considerations

  • How recommendation systems are regulated in the UK, EU and US
  • Legal challenges

Step 00 - Unsupervised learning: recommendation systems (RS) (15m)

๐Ÿ‘จโ€๐Ÿซ Teaching moment

Your tutor will outline how recommendation systems work.

  • How does unsupervised learning differ from supervised learning?
  • How can unsupervised learning be used to build a recommendation system?
  • What forms do recommendation systems take?
    • How do you use a recommendation system?
    • What is their purpose?
  • What choices are available to the designers?

Step 01 - Big questions (25m)

Quality and evaluation criteria:

  • What constitutes a โ€œgoodโ€ recommendation?
    • How do we know when recommendation systems are working well?
  • Who gets to decide?
  • Are algorithmically โ€œgoodโ€ recommendations always moral?

Step 02 - Ethical considerations (25m)

  • What ethical issues do we need to consider when building recommendation systems?
  • At what level should these issues be considered?

๐Ÿต Break (~5 min)