Are Machine Learning, Big Data And Psychometrics Really Destroying Democracy?

A recent column claimed that it's doing just that


In a column on the Bloomberg website, Cathy O'Neil ripped into data, machine learning and psychometrics. In describing the way that Donald Trump and Hillary Clinton used data in their campaigns to target ads, O’Neill claimed:

‘The asymmetry of information represented by this new generation of political ads, tailored to the exact personality type and browsing history of each voter, is scary enough on its face. It's a threat to democracy that directly undermines the notion of an informed citizenry. It's just not specific to Trump.’

So is this the case? Is machine learning, data and psychometrics really a threat to democracy and accurate information?

To some extent, we see this whenever we look at the polarization of people online. This has largely happened due to machine learning, where people are often only exposed to material that they are interested in, which further entrenches views rather than questioning them. For instance, if you were to watch a video on Youtube, the next recommended video will be related to the first, so if you are researching a particular subject the Youtube algorithms will identify that you have an interest in that subject and continue to recommend similar videos.

It is unlikely that people liking videos of cats vs people liking videos of football plays is destroying democracy, however, the machine learning element of recommendation does not only suggest things that people may like - it also helps to profile them.

Traditional political targeting has taken place through geographies and broad demographics. Think about how political commentators refer to the ‘latino vote’ or ‘working class vote’, this is almost as broad as you can get and instead of sending accurate or insightful messaging, people simply get vague messages that don’t have much impact. However, this changed after Obama’s 2008 and 2012 campaigns, where he managed to effectively drill down and use big data and machine learning to send more clear and concise messaging to the people most likely to respond to it.

One of the likely reasons for O’Neill’s reaction to this is the way in which similar tactics have been used in this election, but going one step further than simple segmentation and moving into psychometrics. The idea behind psychometrics is to identify personality types, normally to sell to them in a specific way, but in the case of politics it is more related to specific messaging. Most of the ire surrounding how this was used in the 2016 election was due to how Cambridge Analytica, a British analytics and psychometric company, and the Trump team used the data - to suppress voters for their rival in specific areas and to specific types of people, rather than promote their own candidate.

With a winning margin of less than 100,000 in the three key swing states and a popular vote loss of 2.86 million votes, this seems to have certainly worked and given the feeling amongst most people in the US about Donald Trump, this explains the huge backlash against this approach.

However, blaming machine learning and the data teams that all modern political campaigns now use isn’t especially fair, accurate and it is certainly not ‘a threat to democracy that directly undermines the notion of an informed citizenry.’

Firstly, both sides had access to huge swathes of data, with the Democrat machine already having a huge advantage given that they had everything from two hugely successful previous campaigns and did more with it than the Trump team. They used it in a different, and some may argue more honest, way, but you cannot blame the data itself for the fact that they ultimately did not use it as effectively.

The claim is also forgetting the key reason why people create so much data that can then be used by these political campaigns - because they spend huge amounts of time online. To therefore argue that it ‘undermines the notion of an informed citizenry’ is simply not true as they have this data on the citizenry because they do so much research online. That people are not informed about wider issues is perhaps an indication of IQ prejudice, where smarter people believe that those with a lower IQ lack the ability to research beyond what is put directly in front of them. The internet is available to around 90% of the US according to Pew, so if they are uninformed it is not that they lack the capacity, but simply that they either lack the drive or propensity to do so.

Political messaging is always going to be divisive, especially during political campaigns as heated and bitter 2016’s. It is also important to note that Donald Trump was never going to be popular amongst the mainstream media and those who consume it, with only 1 major news outlet endorsing him compared to 40 who endorsed Hillary Clinton. However, blaming a shock result that, according to opinion polls, has seen more than 50% of the country view him unfavourably on machine learning seems like its just looking for something else to criticize. 

Big data hype small

Read next:

Is Big Data Still Overhyped?