header_1_crop

What does an A.I. think of Clinton, Trump and the RNC?

We’ve heard plenty from pollsters, pundits and politicos on what people think about Hillary Clinton and Donald Trump, but what does an AI (artificial intelligence) think of them?

But some background first. Why am I doing this? AI is an emerging technology that’s on the verge of being deeply incorporated into much of our daily lives. Its impact will be widespread and profound — and potentially invisible. I’m interested in understanding and influencing the design of new and innovative ways that it can be used — enabling all people to understand its impact, and help give voice to its development. I am also interested in understanding how technology may embed existing cultural biases — either through the development of the technology, or in its uses — and finding ways to address this.

IBM’s Watson, one of the best known AI platforms, has a really interesting  tool that gives personalty insights based on its analysis of a person’s Twitter postings. Given a Twitter account name it creates a narrative personality profile, a visualization of the personality data, and the raw personality data. I’ve found it’s remarkably accurate — but try it on yourself and see what you think.

Sample output from Watson Personality Insights for Twitter

Sample output from Watson Personality Insights for Twitter

I was curious to see what emerges when Watson’s intelligence and analytics look at the current US election and the twitter feeds of Hillary Clinton and Donald Trump. Here are the two personality profiles that it created:

@DonaldTrump:
You are shrewd.

You are compromising: you are comfortable using every trick in the book to get what you want. You are confident: you are hard to embarrass and are self-confident most of the time. And you are carefree: you do what you want, disregarding rules and obligations.

You are relatively unconcerned with taking pleasure in life: you prefer activities with a purpose greater than just personal enjoyment. You consider helping others to guide a large part of what you do: you think it is important to take care of the people around you.


@Hillary Clinton:
You are confident and heartfelt.

You are confident: you are hard to embarrass and are self-confident most of the time. You are proud: you hold yourself in high regard, satisfied with who you are. And you are assertive: you tend to speak up and take charge of situations, and you are comfortable leading groups.

You consider helping others to guide a large part of what you do: you think it is important to take care of the people around you. You are relatively unconcerned with taking pleasure in life: you prefer activities with a purpose greater than just personal enjoyment.

Pretty interesting, right? I don’t know the candidates personally (obviously), but these seem quite accurate to me.

Can we learn more by going deeper into the personality data? If we compare the two, what new insights or differences might emerge?

(And this seems like the place to insert a disclaimer… I’m not a statistician, data scientist, or anything else that gives me expertise in this. So there are probably huge flaws in my process, logic, language, and reasoning. Also, keep in mind that the Twitter feeds may be managed by other people, and so only partly represent the candidates personalities. And Watson’s data values include a sampling error, that I’m not showing here, which can range up to ± 10%. )

Ok – so first I started by doing a simple comparison between Clinton and Trump:


(View this chart as an image or a larger interactive)

The first 5 values (Openness through Emotional Range) are called “The Big 5” and each are a summaries of five other attributes. The remaining five values (Conservation through Self-Transcendence) are categorized as “Values.” I chose to visualize Values because it seems that so much of the current political debate is around this topic.

At first look, the two candidates seem fairly similar. But there are differences. Clinton has noticeably higher scores for Extraversion and Agreeableness, while Trump has a higher score for Self-Enhancement.

If we expand Extraversion and Agreeableness to see their sub-values this is what we get:


(View this chart as an image or a larger interactive.)


(View this chart as an image or a larger interactive.)

Again, there are scores that are similar, but there are a couple that appear quite different. For Extraversion, Clinton is more Assertive, Cheerful, Outgoing, and Gregarious — although neither candidates score very highly for cheer. As for Agreeableness, Clinton is higher on Altruism and interestingly (considering the tone of the campaign) Trust. It’s somewhat surprising to note that Trump is higher on Modesty, although neither candidates scores very highly there.

Hashtags (or: The Wisdom of Crowds?)

What about the movements that have emerged surrounding the candidates? Can we learn anything about their characteristics or behaviors? Luckily Watson also lets us do personality analytics based on hashtags. I took a look at #NeverTrump and #NeverHillary (and also #BernieOrBust).

Starting with the personality profiles…

#NeverTrump:
You are a bit inconsiderate, somewhat compulsive and excitable.

You are unconcerned with art: you are less concerned with artistic or creative activities than most people who participated in our surveys. You are self-doubting: you frequently doubt your ability to achieve your goals. And you are contrary: you do not shy away from contradicting others.

You are relatively unconcerned with tradition: you care more about making your own path than following what others have done. You consider independence to guide a large part of what you do: you like to set your own goals to decide how to best achieve them.


#NeverHillary:
You are a bit inconsiderate, somewhat compulsive and guarded.

You are unconcerned with art: you are less concerned with artistic or creative activities than most people who participated in our surveys. You are self-doubting: you frequently doubt your ability to achieve your goals. And you are reserved: you are a private person and don’t let many people in.

You are relatively unconcerned with tradition: you care more about making your own path than following what others have done. You consider independence to guide a large part of what you do: you like to set your own goals to decide how to best achieve them.

That’s not a copy/paste error. The two profiles are almost identical! The only difference is the one sentence that I emphasized. What does this mean? Is there something about how the voices of so many different people using the same hash tag tends to normalize the group voice? Is there something about campaigns that are built as “never,” in opposition to another, that share a common approach?

Still, that single sentence difference does seem to nicely summarize the differences between them.

Note that the Bernie Sanders crowd’s profile is a bit different…

#BernieOrBust:
You are shrewd and inner-directed.

You are unconcerned with art: you are less concerned with artistic or creative activities than most people who participated in our surveys. You are compromising: you are comfortable using every trick in the book to get what you want. And you are hardened: you think that people should generally rely more on themselves than on other people.

You consider independence to guide a large part of what you do: you like to set your own goals to decide how to best achieve them. You are relatively unconcerned with tradition: you care more about making your own path than following what others have done.

Does the #BernieOrBust read differently because it isn’t based on the “Never” premise? To me, it reads closer to Trump than to Clinton’s. (We’ll save that comparison for another day.)

If we look at the graphed data from the three hashtags…


(View this chart as an image or a larger interactive.)

It’s fascinating to see how there are places with noticeable differences in the scores between #NeverTrump and #NeverHillary, yet the personality profiles of the two are almost identical. I don’t pretend to know much about how Watson works, but would love to learn more about how such different data can lead to similar profiles.

The Republican National Convention

Lastly, I thought I’d take a look at the voices that are part of the Republican National Convention, happening this week in Cleveland. I used the Twitter profiles of the “headline” speakers (as published on July 17 by the GOP ). (The only speaker that didn’t have a Twitter account was Peter Thiel, so he isn’t included here.) This is also where my data analysis skills start to get a little wobbly — so I share this in the spirit of co-exploration.

Because there are 16 headline speakers the graph gets a bit busy…


(View this chart as an image or a larger interactive.)

It looks like a fairly broad spread of personality types. They all tend, more than not, to be Open and Conscientious, and with a lower Emotional Range. If you look at where Donald Trump (marked by a star) falls, he’s much less Conscientious and Agreeable than the others, but has a higher Emotional Range than everyone (except for Donald Trump, Jr.). I wonder if this raises the likelihood that his speech, the last of the convention, will end things on a high note.

I was also curious to break out the speakers by sex and by race. I’m not very sure that these indicate — do they illustrate characteristics of how the GOP chose speakers, broader differences between sexes or races, or biases in how Watson interprets different voices? I share the charts, but leave the analysis to another, smarter, project.


(View this chart as an image or a larger interactive.)


(View this chart as an image or a larger interactive.)

What’s Next?

I found Watson’s analysis of the candidates to be remarkably accurate, and examining the voices of crowds via hashtags gives a new and unique “personality” insight. As a follow-up to this, I’m eager to do a similar analysis of the speakers at the Democratic Convention later this month; as well as diving deeper into this data. If you have any suggestions or ideas, please let me know.

Methodology
Data was taken from Twitter accounts on July 20, 2016. Although Watson can do analysis from other data sources, such as Facebook pages and bodies of text, but I chose Twitter feeds because, unlike the Facebook analysis, it can be done without the account owner signing in; it also is more representative of the person’s voice over time, that if I were to select an arbitrary body of text to represent the individual. I used http://tweetmewatson.mybluemix.net/ to get the data values.

 

header_1_crop