19 Mar 2022
Thought Leader Keynote - TC22
Date
May 18 8:00am - 8:30am
Speaker
Hannah Fry
Professor, Centre for Advanced Spatial Analysis at UCL
Notes
I have been acquainted with the work conducted by Professor Fry for several years. Her work on understanding the spread of pandemics in the BBC Four documentary (2018), "Contagion: The BBC Four Pandemic" was highly thought-provoking. More recently, I have found her publications including "Hello World" and the DeepMinds podcast also extremely fascinating. Subsequently, I was extremely excited to see her presentation at TC22.
Data driven views vs Data informed views
Starting her talk with the example of making a decision based on an incomplete picture, Fry emphasises the need to ensure that individuals and organisations are making data informed views, rather than data driven views. For some, this may be a shift in views as the focus on data has led to data being placed at the centre of decision making.
As the talk highlighted, data driven views have major limitations. Firstly, collecting and modelling data effectively has a cost benefit. To accurately predict an outcome a large quality of data and even then, key information could be missed. However frequently this is not possible, and we need to make decisions with limited information. Therefore, in a data driven view, a less than optimal course of action can be favoured, in direct contrast to intuition and experience.
Furthermore, randomness cannot be omitted as a challenge to predict key markers in a longitudinal study of children has been revealed. This is further emphasised by studies that look at the cause of death and cancer which found that numerous instances where individuals passed away before cancer had a major impact on their lives. Consequently, in both these cases, a data driven decision would have led to an outcome that diverted from reality. This is particularly key as smart devices gain prominence. As Medlife Crisis (2022) highlights, too much emphasis on data can have a negative impact on individuals.
For companies and society in general to succeed, individuals must have a level of data literacy and be open to accept when the data contradicts their ideas. However, they should also understand the limitations of the data to ensure that the final action is undertaken in a robust manner.
Predicting imbalanced outcomes
Following the publication of a highly controversial study predicting the sexuality of individuals through images, the study has been referenced in many discussions about the issues surrounding data science and prediction and understanding the wider social impact. This study in particular has raised concerns as homosexuality is a criminal offence in many countries.
Others also discuss the accuracy of such models and so does Fry in her discussion. A high accuracy can be undermined if the sample or the overall populations actual result favours one outcome more than the other. For example, as Fry is quick to highlight, a model with a high degree of accuracy can be build even if the model simply assumes everyone in a given population is straight. Leaving aside this particular study, understanding the ratio of false positives and false negatives is extremely important. However, in a world where clickbait titles are prominent and individuals do not always have the relevant data literacy to critically examine a data, it can cause issues.
One key instance of this was during the pandemic and the results surrounding the accuracy of Innova lateral flow tests. According to one article published by the Imperial College London, (Duncan, 2022) there was a high rate of false negatives.
Analysis by a team including experts from Imperial College London found that the proportion of people with current COVID-19 infection missed by the Innova brand of lateral flow test (LFT) was “substantial enough to be of clinical importance”. ... It found that the devices would miss between 20 per cent and 81 per cent of positive cases in the different settings – 20 per cent at the Test and Trace centre, 29 per cent in the city-wide mass testing, and 81 per cent in the university screen testing.
These can be highly problematic when the key aspects such as the safety of a population is at risk. Therefore, encouraging data literacy is extremely important in the 21st century.
Conclusion
This was an extremely intriguing talk and if you are able to view it or see Fry's other talks and publications, I highly recommend it. Also she recommended a book "The Hidden Half: How the World Conceals Its Secrets" by Michael Blastland. I hope to check this book out in the near future.
Works Cited
BBC. “BBC Four - Contagion: The BBC Four Pandemic.” BBC, 2018, https://www.bbc.co.uk/programmes/p059y0p1. Accessed 19 May 2022.
Duncan, Conrad. “Lateral flow tests may be missing 'substantial' number of COVID-19 infections | Imperial News.” Imperial College London, 24 February 2022, https://www.imperial.ac.uk/news/234154/lateral-flow-tests-missing-substantial-number/. Accessed 19 May 2022.
Medlife Crisis. “The Problem With Silicon Valley Medicine.” YouTube, 13 April 2022, https://www.youtube.com/watch?v=rW3DGnHO2iY. Accessed 19 May 2022.