An AI Analyzed My Twitter Feed and Discovered I’m a Shithead
DeepSense says it can analyze tweets to see if someone is a good employee. I'm going to need to make better tweets.
It's fairly certain that, given the incredible amount of data about our habits and thoughts that are collected and analyzed and stored by our devices and accounts, The Cloud knows more about ourselves than we do. If our entire conception of self is distorted by the delusions and lies and beautiful memories we rely on to get through the unlimited hell that we occasionally experience, then who's to say the endless thoughtstreams we dump for free into Twitter isn't a more accurate representation of who we really are?
Computers, that's who. And a new computer program says that I'm a real asshole. Who knew?
Thanks to Wired's Tom Simonite, I have just been introduced to DeepSense, a tool that claims to analyze someone's personality and employment viability via their Twitter feed. A recent entrant in a long line of internet-based personality-analysis-and-data-collection tools, DeepSense's pitch is to help hiring managers figure out if potential employees are competent or not before they even come in for an interview.
As a person who occasionally hires people: great! And as a person who occasionally just vomits pointless nonsense into Twitter in a long attempt to obscure such personality analysis: shit! Might have shot myself in the foot on this one.
In any case, the idea that our Twitter feeds are an accurate representation of who we are is fun and terrifying. On the one hand, everything we post on socials is calibrated, subconsciously or otherwise, to present a version of ourselves honed by analytics of our every behavior online. On the other, perhaps there's some universal truth buried way deep inside the data.
Since you should probably think twice about plopping your accounts in for analysis by some anonymous tool—DeepSense seems to have been around since last year, but outside of a couple puffy pieces that might be nicely-disguised press releases, hasn't really gotten any scrutiny—I threw myself to the wolves. Here's what I learned!
Turns Out My Personality Kinda Sucks Ass
I always assumed that I am a fun, happy person from California who likes to ride bikes and celebrate friendship while saying things like "sick" too much, but turns out that's entirely incorrect. I highly doubt you want to read the hilariously-detailed block of text above, so here's me as a Zagat review:
"A quiet individual," Derek Mead is "not very thoughtful about the motivations and priorities of others." A journalist who "could almost be mistaken for a pessimist", he "can be very cynical and distrusting of what [he] comes across." As as a longtime, "very edgy" employee of VICE Media, Derek "could run out of patience with lack of achievement or direction." The frequent tweeter has "has a sharp goal focus" and "is generally a planned and organized individual."
I am 47% anxious.
I have no idea what the above graph means but I think it's roasting me.
I Think This Is Telling Me That My Ideal Career Is Either "Moody Assassin" or "The Guy Who Decides Which Trees Get Cut Down"
According to the computer, my potential for stability is "low", which I think is because I'm so action-oriented. I have no idea why this thing thinks I'm such a contrarian loner though, in real life I am a fantastic team player because I always agree with everyone.
I sent my DeepSense profile to my boss and she responded:
I only disagree with this: could almost be mistaken for a pessimist
I am optimistic this analysis will help me get a raise.
I Need to Be Nicer to the People I Like
I'd first like to point out that I am mostly active on Twitter in the evening because I am such a diligent employee and keep the personal stuff to after-hours. I am also not surprised that, with a score of 5.6/10, my social influence is "high." But what did hurt was to see a list of my favorite people (and Mike Isaac) on the right there with all this negative sentiment aimed at them. Forget my viability as an employee, what about my viability as a friend? This is kind of fucking me up, I'm not going to lie.
I Am Extremely Basic
As a bearded digital media innovator with a curated blend of high-end and shitty tattoos, I've always assumed that I am always abreast of the bleeding edge of cool. Nope! Turns out that if you put downloaded my Twitter feed into a flesh printer (maybe the next phase of DeepSense?) it would spit out a 30-year-old white guy with a bad haircut. And then it would print out an Instagram ad for the world's greatest white tee.
My Personal Brand Is "Thought Leader In Need of a Vacation"
As a professional thought-haver, I'm not surprised at the range of things I talk about. But I did not expect that so many of those thoughts to be negative. Am I sad because I've downranked positivity in my own mental algorithm due to the shitstorm our simulation has become, or am I sad because I've allowed my entire conception of reality to be subtly-but-notably shifted towards "fucked" after years of processing all information about the world via a platform whose core mechanism of virality is rooted in negativity? Imagine how much more positive the computer would know I am if I tweeted "taking a nap" more.
As someone who will "run out patience with lack of achievement or direction," I'm kinda tiring of this blog post, so let me try to get to some sort of point. The first question: Is it accurate? It depends on your definition of what accurate is. I don't think this is an entirely unreasonable approximation of my last few months on Twitter; I also hope to god that no one would ever assume that Twitter is entirely reflective of who I am as a human being. I wouldn't pass a ton of judgment on DeepSense itself as anything more than a detailed sentiment-analysis tool; Frrole, the company that created it, claims to have done a lot more impressive work in the social-analysis space than this tool. But I can guarantee this is the exact kind of buzzy (it's artificially intelligent, god damnit!) personality-analysis service that’s going to come into use eventually—and in other forms, already has.
Taking a risk on anyone—be it for work, or love, or just hopping in a cab—is fundamentally a pretty nerve-wracking endeavor, one that's so fundamental to the human experience that we still haven't adjusted fully to digital relationships. And we often compensate for risk in problematic ways, especially in the workplace. For example, it's impossible to fully know how someone will act at work before you hire them. As a result, managers tend to have a significant bias towards certain types of interviewees in the hiring process. In that case, isn't a tool that gives a clearer understanding of how an applicant thinks and works via a more freeform analysis of their personality as presented via socials a great idea? Sounds great, yeah.
The flip side is that trying to pull an accurate picture of someone out of the data streams flowing in and out of our brains and bodies is likely impossible due to the very biases that ebb and flow in our interactions with the systems that collect the data in the first place. Think about it:
- The thoughts you choose to present on Twitter are likely affected by what's currently part of "the discourse" on the platform.
- The shape, tone, and delivery of those thoughts is likely affected by the current sentiment of the tweets you're reading, as well as your past knowledge of what tweets are popular and which are not.
- While that persona is likely to evolve over time, analyzing it as a complete record adds significant noise, or at least presents a version of you that assumes time does not exist.
- Those biases, and many others, are also reflected in the people who are programming the algorithm analyzing your language and sentiment, and their assumptions about how our tweets connect to our personality will be shaped through a worldview and experiences that are assuredly different from yours.
All of that means it's really hard, at best, to accurately analyze the personality of a large swath of people based on thoughts that are skewed when put out and further distorted when absorbed by the machine. On a lighthearted level, that's how we end up with me having a 100% negative sentiment to a variety of people I adore so much that I feel comfortable talking shit with them. On a much more concerning level, this is how algorithmic-people-sorting has so far proven to further entrench disparities in socioeconomic power structures.
What's the solution? On a practical level, I guess tweet like your next boss is watching. Working in media means I read the tweets of any prospective hire, but there's a huge difference between that and having a hiring manager somewhere bulk-psychoanalyze applicants with an app they wandered across. And concerning as that is, it's going to happen.
But, again, we do find ourselves taking a lot of risks on other humans, especially in scenarios where our only interactions are online. This is why we've seen such big pushes for trust in the internet, which is already shaping how information is presented to us in our feeds, and which will continue to evolve to follow the path being blazed by China's Social Credit Score. (The American version is already arriving in suitably-corporate fashion.) The argument for these types of trust systems always boils down to a variation of "what have you got to hide from the machine?" But the real question is: why should I trust the machine to know me?