As the media and pollsters begin their soul-searching to understand their biases and blindspots, Artificial Intelligence data experts have been taking a closer look as well. Sanjiv Rai of Indian start-up Genic.ai spoke about his MogIA to CNBC and revealed that AI may have been the shouting Cassandra in the 2016 Electoral Chaos.

We at Graphika, have also been reviewing the data that generated in the run-up to Election Day.You might remember that in the last week of the 2016 election, we released our  Political Media Barometer.

Social Media Data presents an unprecedented opportunity for students of large scale human social behavior.  This data contains digital traces of human activity: millions of interactions, posts and preferences recorded in a neat package, wrapped in an API layer, ready to collect and analyze. The scale of Social Media allows for studies with vanishingly small margins of error. Such studies claim to predict where a person will be, what they are looking to buy, or who they will vote for, with near-perfect accuracy at the scale of the entire human race.

For example:

If you took a quick glance at this graph . . . If you didn't know what the left or right meant . . . If you didn't know what the color codes meant . . . If you didn't know what the size of the dots meant . . . If you just saw THIS:

The Graphika Barometer - X axis obscured

Whose side would you think was winning?

Let's take a look at the left:

The left side of the barometer is filled with with the medium sized dots hovering solidly in the middle of the y axis.

The dots cluster in a maximum of six dots.  There are fewer clusters. Those clusters have fewer overlaps.

There is plenty of open space.

Would this strike you as the more engaged and active community?

Now look at the right side of the graph:

The dots are uniformly larger. They also stretch farther up the y axis.

The dots are heavily clustered. Those clusters overlap. Dots on top of dots on top of dots that are so dense that it's difficult to click on all the dots because you can't see where one begins and one ends.

You might think - THESE are good voters to have. They are united, engaged and excited. They are communicating with each other. They are on the same page.

And you would be right. You would be talking about the highly Conservative side of our Political Media Barometer, the side who just astonished the Mainstream Media with their electoral power.

But the data was there the WHOLE TIME . . .

SHARE AFTER SHARE AFTER SHARE . . .

DAY AFTER DAY . . . OR WAS IT?

There is reason to believe that a careful examination of the factors at play in a particular social media phenomenon can yield an accurate translation of the underlying human behavior.  Indeed, Graphika's maps provide an ecologically valid segmentation of social media that reveals the interest groups of real human actors around green energy, cybersecurity, or the US Political Environment. At the same time, this translation work requires a significant time and effort.  In such a high-stakes context as the recent US Presidential Election, it is not enough to merely record that one group has more Twitter activity than the other to predict victory at the polls. 

Our Science Director, Vladimir Barash,  explains the dangers of trying to predict human behavior from Social Media Data further below:

It is important to remember that a digital trace of voting is not the same act as filling out a ballot. It’s better to think of Social media as a shadow of human activity.

Much like a shadow, social media is neither perfectly aligned nor completely divorced from social behavior. Social media can be a near-perfect representation of said behavior when the data is complete and consistent.

It is, however, a fuzzy one when (as frequently) data is full of missing values and format errors. What’s worse, human actors can deliberately distort social media much as in a shadow play. Through trolling, spam and social engineering, a concerted group can create the appearance of consensus out of dissent or create the appearance of popularity out of thin air.”
— Vladimir Barash, Science Director, Graphika Inc.
Graphika's Political Polarization Map

In future posts, Graphika will identify the sources of activity in its political barometer and analyze their authenticity to confirm our initial conclusion. 

Graphika, and #BigData in general, must realize that as the behavior of the electorate changes and relies more on social media to communicate and energize  - that we have a huge responsibility to supply and interpret that data in a responsible and accurate way.  Let us hope we are up to the challenge.