# Data Ethics Club meeting [31-05-23, 1pm UK time][timedate] ## Meeting info ### Quick links [Zoom link][zoom] Link to content: [A Twitter thread by Owen Jones about the classification of abuse online](https://twitter.com/OwenJones84/status/1590295662759153665) ### Description This week at Data Ethics Club we are discussing [a Twitter thread by Owen Jones](https://twitter.com/OwenJones84/status/1590295662759153665) about a study where machine learning was used to analyse three million tweets mentioning MPs for ‘toxicity’. You can [read more about the study in this BBC News Article](https://www.bbc.co.uk/news/uk-63330885). The basic premise of this sort of sentiment analysis is that you isolate occurrences of certain comments, for example “you are/you’re a disgrace” “you are/you’re a liar” and evaluate the sentiment associated with them. In the case of this study, the sentiment is classification with an associated probability: you classify how likely is something to be toxic. The BBC's Shared Data Unit used [Perspective](https://perspectiveapi.com/), a tool that uses artificial intelligence to spot toxic comments online. Developed by [Jigsaw](https://jigsaw.google.com/), a research unit within Google, it defines a toxic comment as one which is "rude, disrespectful or unreasonable" and "likely to make someone leave a conversation". The team analysed all tweets mentioning MPs from March to Mid-April. The article goes on to discuss the effect of toxic tweets on MPs. Owen Jones’ twitter thread portrays this study in a different light, taking issues with the conflation of being called a “disgrace” or a “liar” with racist or sexist abuse. For example, “you are a hypocrite” was classified as abuse. The thread goes on to point out that the most negative pushback in a tweet towards an MP was classified as toxic. Where is the line? Should AI be drawing the line? One Twitter user tested Perspective on phrases associated with Nazi views: “We must secure the existence of our people and a future for white children” which was deemed 34.33% to be toxic, [“I have 14 words for you”](https://en.wikipedia.org/wiki/Fourteen_Words) which was deemed to be 9.48% toxic and finally “You are a poo poo head” which was deemed to be 76.52% toxic. ### Discussion points 1. The BBC News Article states: “Machine learning algorithms allow researchers and journalists to measure a phenomenon at a scale which would otherwise not be feasible with other methods.” Do you agree with this? Why/why not? 2. How can we decide if something is toxic? Who/what should be the ones to decide this? 3. Should the line of what counts as toxic be different if you’re someone in the public eye? --- [timedate]: https://www.timeanddate.com/worldclock/fixedtime.html?msg=Data+Ethics+Club&iso=20230531T13&p1=299&ah=1 [content]: https://twitter.com/OwenJones84/status/1590295662759153665 [zoom]: https://bristol-ac-uk.zoom.us/j/94475153265