Data Ethics Club meeting 20-01-21, 13:00-14:00 GMT#


You’re welcome to join us for our next Data Ethics Club meeting on Wednesday the 20th January at 13:00-14:00 GMT. You don’t need to register, just drop in. This time we’re going to watch/read the Executive Summary of the Review into bias in algorithmic decision making by the Centre for Data Ethics and Innovation (CDEI), which is a recently published (end of November 2020) government report.

Natalie suggested this week’s content, and will be leading this week’s meeting.

Discussion points#

There will be time to talk about whatever we like (relating to the content), but here are some specific questions to think about while you’re reading.

  • Were you surprised by any of the examples of algorithmic decision making currently in use?

  • Which of the CDEI’s recommendations do you agree/disagree with?

  • Are there any recommendations that you think are missing?

Meeting notes#

What did we think?#

A summary of the discussion was that:

  • The report didn’t shy away from the potential harms which was good.

  • There is a sense of the tracks being laid in front of the train in terms of regulation. One group felt there is a need for more guidance for private companies, but who is responsible for implementing that?

  • Also, how do you regulate algorithms? One rule will not fit all, and it will be a difficult field to manage in this sense.

  • It’s likely that there aren’t enough people who understand enough about algorithms to assess whether they are a good idea for their use case. There are also times where algorithms are sold for a use case that should not be using algorithms at all. Pressures on funding, particuarly in local gov and police, make these seem more tempting though.

  • How can we centre the data subject? Do they know what their data are being used for, and the impacts of it?

  • Recommendations to collect more info about protected characteristics were open to question - who will volunteer this data, and will this serve the people who it needs to (who may be especially unlikely to provide this data in the first place).