Blog: The biggest Cambridge Analytica ‘scandal’ is that their methods don’t work
It is now over a year since the Cambridge Analytica scandal and today Facebook faces a fine of up to $5 billion as a result of its data-handling.
It was over two years ago the issue of Cambridge Analytica caught my attention, as part of writing my book, Outnumbered. While researching the book, I was the first person to get an on-the-record interview with Alex Kogan, the researcher who downloaded Facebook data and sold it to Cambridge Analytica.
What I found out was a very different story than the one presented by whistleblower Chris Wylie and that had been given earlier by Alexander Nix, CEO of Cambridge Analytica. They claimed that using Facebook data they could target voters personalities and influence the election in Trump’s favour. But as Kogan told me, almost two years ago now, “that shit doesn’t work”.
Kogan’s view squared with my own investigations. I used a similar data set to the one acquired by Cambridge Analytica and showed that, exactly as Kogan told me, the effectiveness of political personality targeting is negligible. Two of the Cambridge researchers, Michal Kosinski and David Stillwell, who carried out the original personality studies suggested that they could target instead extreme personalities. So I tested that too. While there was some effect, it remained weak. Too weak to influence an election.
As an academic, it has been interesting to me to watch the Facebook/Cambridge Analytica story unfold. In the UK parliament, Alexander Nix adopted my line: his methods were flawed, he claimed in defence of his actions. Chris Wylie continues to act as a digital warrior, despite several people alleging that he went even further than Nix or Kogan in downloading and using Facebook data. While Kogan realised the techniques were doomed, Wylie didn’t get it. Judging by an interview in the Guardian he still doesn’t get it.
I should admit where my own knowledge is limited. I don’t know the legal rights and wrongs of how Facebook handled their data. Maybe they do deserve a $5 billion fine? What I do know, however, based on research and an examination of the inner workings of the algorithm are two things.
1, Kogan is right, “that shit doesn’t work”.
2, Very few people seem to have understood that the shit doesn’t work.
This second point is important and is the key issue I look at in Outnumbered, not just in the context of Cambridge Analytica, but also in the context of sentencing algorithms, Fake news, echo chambers, Google language processing, Deep Mind’s Artificial Intelligence, Five-Thirty Eights election predictions, Amazon’s ‘also likes’, and many other examples.
If you want to find out which shit works and which shit doesn’t (and why) then buy the book! It’s really good.