We're sceptical about AI, political analysts tell Lords committee on polling
Polling analyses in the run up to this year’s UK general election came in for flak after failing to predict a hung parliament, but now two experts in gauging the opinion of the electorate have poured scorn on the idea that letting artificial intelligence crunch data from social media could result in increased accuracy.
Political science expert Professor Will Jennings and pollster Nick Moon today told members of a House of Lords Committee that they are sceptical about new-fangled polling methods based on supposedly cutting edge software that harnesses big data from posts on social media.
They suggested platforms like Twitter were too much of an echo chamber, with people merely amplifying the beliefs of like-minded types, but also that it was hard to trust people really meant what they said online.
Moon said forms of artificial intelligence were already widely deployed by technology firms to monitor what customers were saying about their products but that the technology often struggled to grasp the subtleties of sarcasm and irony and took wry musings far too literally.
When customers sarcastically tweeted “Great job Apple!” after complaining that their iPhone was not working, this was typically interpreted as praise for the computing giant, Moon said.
Baroness Couttie, a member of the Political Polling and Digital Media Committee, had asked whether analysing tweets en masse, while at the same time adjusting for bias, might be a better means of forecasting opinion than relying on traditional polling techniques.
Jennings told her: “I, at present, would be very sceptical about the possibility of using social media as an alternative to polls - simply because social media, just like polls but probably more so, has these fundamental issues around representativeness. Is the population that uses Facebook or Twitter representative of the general population? One would argue that they suffer even more from the sorts of biases that we see in polls.”
By way of an example, he said that in advance of the Stoke Central by-election earlier this year, pro-Brexit campaign group Leave.EU had published a poll showing the UK Independence Party (UKIP) candidate, Paul Nuttall, was far ahead of his Labour rival. Aaron Banks, the British businessman who co-founded Leave.EU, had promptly hailed this finding, saying it was “based on some sophisticated AI from US”, Jennings said, but it transpired that the poll was in fact deeply flawed, and Labour romped home to a relatively comfortable victory.
Political pundits “shouldn’t be completely dismissive” of information gleaned through the bulk analysis of social media posts, Jennings said, but he stressed technology alone had so far shown itself to be inadequate as an indicator of public opinion during election campaigns.
Moon went further still, telling Baroness Couttie: “I’m extremely sceptical. I’ve got no doubt at all that in the last two elections various people were doing exactly what you say because it’s virtually free, and if you’ve got the necessary computer skills you can access it [the relevant information from Twitter etc] very easily.
“People were doing it, and if it was any good they would have told us afterwards how good they were at it.”
He added: “It offers potential, but it’s how you sort the wheat from the chaff… The last time I looked the 80/20 rule still appliesd. Eighty per cent of all tweets come from 20 per cent of people tweeting. And the vast majority of tweets are just retweets from somebody else. Nowadays we know there are people in China paid by the hour to supply tweets about anything anyone wants them to tweet about. Those of us who watch Homeland will have seen it going on there. That’s the kind of reason I think we should be very dubious about this.”
As many as 48 accounts on Twitter alone could be being operated by internet bots, or robots, rather than people, according to one recent study in America. Currently robots are denied the vote.