Highlights from the Data Ethics Breakfast | by Luke O'Rafferty

Multiverse Roundtables bring together apprentices with shared experiences in an intimate setting to exchange success stories and share advice with peers who might be experiencing the same dilemmas they encounter.

At this Roundtable Breakfast that took place in October, we discussed the topic of Data Ethics. Led by a facilitator, each table discussed the prevalence of data and its implications using specific case studies.

One attendee, Luke O’Rafferty shares his reflections from the breakfast…

“What a great way to kick off the day, debating data ethics over breakfast with the Multiverse community. I completed my Data Fellowship Apprenticeship earlier this year and have been continuing to take advantage of events that I can attend as an alumnus. 

Towards the end of my apprenticeship, I was fortunate enough to successfully apply for an internal transfer to our central Data and Analytics team. As I have gotten involved with business data, it is easy to see how quickly data ethics becomes important. A discussion about improving gender balance quickly leads to thoughts of analysing how we improve the retention of staff. But if we can identify the primary factors that are likely to lead to somebody departing, it is not a large leap to think that you could take things so far to identify who is unlikely to leave even if you don’t reward them.
 
We talked about the importance of knowing your business principles and how they inform your culture and behaviour, but repeatedly at our table, we discussed the idea of transparency. This meant transparency both internally and externally around what data we have used, how we have used it and what shortcomings we are aware of. So for example we were asked; “If arrests are a proxy for crime, and police presence is allocated accordingly, is the efficiency gain in resource management worth the cost?”. Clearly here the problem is the use of that proxy, and if the leader of an organisation such as the Met Police admits to “systemic bias” then we need to find other proxies or data points to allocate what is a limited resource. If you were to allocate 100% of your police force to an area that was, for example, predominantly of one ethnicity, then nearly 100% of your arrest figures will clearly come from that population.

We had another challenging debate around healthcare data. If the measure of success is that the average life expectancy is increased by making a change in treatment, is that a good metric? The average there is the problem and could mask any number of deficiencies or actual harm caused by the treatment to some. Again in a discussion on improving healthcare, we need to be clear on what metrics are being used, who they consider, and the potential inherent biases that those metrics may be based upon. If a group has been marginalised or neglected in society it is likely that their health outcomes will be poorer than those who are well-supported. To treat them ethically we cannot ignore the difference, but we also cannot withhold a treatment only on the basis that the patient is from the former group without any other indicators.

The case studies presented gave us a variety of scenarios to consider, and it was a great opportunity to get the viewpoint of apprentices from a range of backgrounds, industries and stages of their apprenticeship. These are real-life issues that are being faced daily within the data community – indeed one keynote speaker at the recent BigDataLDN conference suggested we will need many more philosophers in the years to come as the application of Artificial Intelligence increases.

Thanks to Hannah Siaw for facilitating our table, best of luck to all of those apprentices who are just getting going and cheers to Multiverse for a lovely breakfast!”

Luke O’Rafferty is a Multiverse alumn from the data programme.