The Secret Service wants to know what you're talking about on Twitter and just how sarcastic you are. So be sure to limit your snark while Big Brother listens in.
Sounds scary, doesn't it? Well it is. Especially since sarcasm analysis and sentiment analysis are horribly imperfect technologies.
On June 2, the Secret Service posted a request for proposals for a "computer based annual social media analytics subscription."
Now, that in and of itself isn't cause for concern. Social media analytics is a growing field, and full disclosure, I write this as the CEO of one of the leading companies in the space.
Members of the Secret Service wait after President Barack Obama arrived at Atlanta Hartsfield Jackson International Airport March 16, 2012 in Atlanta, Georgia. President Obama is spending the day traveling to Chicago, Illinois and Atlanta, Georgia to attend private and public campaign events. AFP PHOTO/Brendan SMIALOWSKI
When used properly, this technology provides deep insights into voters, buyers, subscribers, and fans. It allows those who want to know what you enjoy and prefer to glean those insights with minimal intrusion into your online experience.
Political campaigns have found it incredibly useful in learning more about voters, and brands use it to hone their advertising and marketing spending.
But just what is "social media analytics?" Simply put, it's the ability to look at large amounts of online chatter and detect patterns within groups. It enables us to know what groups like, how they might vote, what music they tweet about, which movies they discuss online, and what's important to them.
I wrote a column here on TheBlaze in March explaining how social media analytics accurately predicted a special election primary in Florida.
Both the political and private sectors have found social media analytics are highly effective for them, while protecting the privacy and integrity of those engaging online.
Even law enforcement has found ways to use the technology to promote public safety while safeguarding privacy. For example, they might analyze tweets emanating from a school with known violent gang activity, but focus on precise key words that might indicate narcotics activity or a violent crime. In this case, no individual is the subject of unwarranted scrutiny, nor is a large group being cyber surveilled en masse.
Bethany Clarke/Getty Images
The Secret Service plans to change all of that, and not for the better. It appears that they seek to cast a very wide net searching for chatter considered threatening or indicative of illegal activity.
Predictably, this has privacy advocates and Americans who believe in small government very nervous. We tend to look at government watching and listening to us quite differently than we did in a pre-Edward Snowden world. What the Secret Service intends to deploy looks very much like blanket surveillance of all of our social media chatter.
They claim that they will critically filter with deployed sarcasm and sentiment analysis to avoid false positives. In other words, their technology will know you're being a snarky jerk, and not a real threat.
The only problem is, all of us in the social media analytics business know that those technologies do not work to any worthwhile degree of efficacy. In fact, at best they are about 20 percent accurate.
Now, I don't know about you, but I'm not comfortable with the Secret Service deploying blanket surveillance to find threats, when that system has an 80 percent chance of getting it wrong.
What Exactly Are They Buying?
Here is the Statement of Work from the request for proposal which outlines in detail the specific requirements. Most of it is run of the mill. Things like historical Twitter data, trend analysis, heat maps, user permissions, and modules of that nature.
What they are trying to procure is actually pretty simple, and the type of tools I deploy daily for clients. In short, they will build dashboards around a pipeline of Twitter data, drop some algorithms into the mix, crunch up the data, and generate reports and alerts which will then be reviewed to determine if further action is required.
Screenshot from Facebook.
Where this gets concerning, and takes a turn to a place that makes me very uncomfortable, is in the bullet point that requires the "Ability to detect sarcasm and false positives."
That simply cannot be done on any meaningful level. I have personally evaluated no less than a dozen vendors of this technology hoping to find one magic bullet to enhance my product offering. There are none. None that I have tested have ever demonstrated better than a 20 percent efficacy rate during test analyses.
It isn't at all surprising why the technology is flawed. One main reason is that here in the U.S. we have too many regional dialects. As a New Yorker transplanted to Miami, I order a hero when I want a sub. My friends in the south call soda a Coke. The examples go on and on. Therein lies the problem with sentiment analysis. It is based on key words. So while I might not think the deli I'm ordering in is the greatest in the world, the word "hero" will skew towards a positive sentiment when my online chatter is analyzed.
And then there is sarcasm, an even tougher safe to crack. It almost universally defaults to the opposite sentiment. "Damn that's a killer car" would easily return a negative sentiment and indicate a potential threat.
Factor in all of the other languages coming across social media and you get the picture.
Those selling this tech love to dazzle us with terms like "hierarchical etymology engines" and "advanced linguistics algorithms." But what they really have are white and black lists of key words they hope will work. And to date, they do not.
What's The Worst That Could Happen?
When one thinks about the current state of affairs, and the constant political chatter about the president and members of Congress, one needs to understand that a system like this could easily be abused and the results used to target selected groups. Something we know has occurred at the hands of the IRS.
A hypothetical abuse case might go as follows: A large amount of chatter critical about President Barack Obama, that also includes hashtags indicating support for the Second Amendment and conservative groups or causes is detected by the system. The sentiment analysis engine defaults all to negative, and the sarcasm engine is inconclusive despite the comments being obviously snarky.
An agent takes this information and begins to review the individual tweets based on leads generated by the system. He then focuses on a few that are highly critical of the president, written by those who are also firearms enthusiasts.
If that agent is reactionary, or there is an agency-wide agenda, what was normal harmless internet ranting is arbitrarily escalated to an actionable threat. It would not be difficult to find a judge to issue warrants.
I'm former law enforcement and have been on the other side of that conversation.
Words like "Your Honor, due to my training and experience over 'x' years as an agent with the United States Secret Service, and due to the information generated by our newly deployed social media monitoring system, I deem these threats real and credible and request a warrant for..." carry quite a bit of weight. Especially in this highly charged political environment.
Now of course that is a worst-case scenario, but Americans deserve better than to have that possibility hanging over their heads. Especially when it is based on technology that is 80 percent ineffective.
We would be wise to heed the words of Benjamin Franklin - “Those who would give up essential Liberty, to purchase a little temporary Safety, deserve neither Liberty nor Safety.”
TheBlaze contributor channel supports an open discourse on a range of views. The opinions expressed in this channel are solely those of each individual author.