Michael Kreil

Over the last years I have been researching the widespread theories about social bots, scraped over a million Twitter accounts, and reproduced the main studies of the topic. None was able to withstand basic scientific verification.

8 Nobel laureates, 79 @Reuters journalists, half of the US Congress, @ggreenwald and @doctorow are bots: It is time to debunk the flawed #socialbot research.

The Computational Propaganda Project of the Oxford Internet Institute @polbots is the most widely used source of the claim "#socialbot manipulate voter opinion". The basis for their study? 50 tweets/day = highly automated Humans like @ggreenwald post more than 50 tweets/day.

The South California/Indiana @Botometer is based on a training data that doesn't contain #socialbot, but "content polluters" aka "spammers" & "malware disseminators". Their method misclassified 12% of Nobel Laureates as bots and 60.7% of the Botwiki bots as humans.

The Berkeley/Swansea team claims that during the 2016 US elections, #socialbot shifted the results by 3.23% towards Trump, based on a correlation between the amounts of tweets using various political hashtags & vote results. What they forgot: Correlation doesn’t mean causality!

We need to ask ourselves how the totally unfunded #socialbot theory spread so widely. And more generally, we need to ask ourselves how and why academic research teams spread wrong theories based on scientifically flawed clickbait research.