I was surprised a few days ago following the vote on the Articles of Impeachment to find a Twitter user effectively calling me a “Trollbot.” It was alarming to see such a dangerous label slapped on my Twitter account. Even more alarming to see a website advocating the abuse of Twitter’s reporting system to attack my Twitter Account and by proxy my fair share of the public square so eloquently described by Jack Dorsey in September 2018.
It all started when I responded to a Tweet sent out by Chip Roy a member of the House of Representatives from my state, Texas. I follow his Twitter account, and I have my account set up to notify me when he tweets. I like to respond sometimes in support, others to call his Tweet into question or redirect, and still, others to add my opinion. You have to respond quickly or your Tweet will be missed.
All of my effort in the hope to influence a member of Congress from my state. There is nothing wrong with that, after all, that is one of the many things Twitter is good for. Our Representatives have a duty to listen. Thanks to people like Jack Dorsey and those who work alongside him at Twitter I have access to an ideal platform to do so.
At the bottom of that inflammatory and derogatory Tweet targeting me, the botsentinel website clearly advocates for users to “report this account to Twitter and avoid.” It does not say why to report it, just, report it. That is concerning when you take into account the number of users on Social Media who have lost their access and never really knew why. I know plenty of them personally.
I assume this was a huge problem for Twitter resulting in their update of the Twitter rules in 2017. A rules change that makes it easier for people to know what could get them banned. More importantly to respond to the general outcry for transparency following the 2016 elections, and the intense legislative scrutiny during the last three years. I should know I ran for Congress on a platform to dive deeper into the issues with Social Media.
Knowing the aforementioned past and with the hair on the back of my neck standing straight up, I decided to do some digging into this botsentinel website. What I found was more than just another bot or online troll, I found a site that is a bot in support of online trolls. Built by a Trump Hater often used by Trump Haters to attack Trump Supporters.
The website botsentinel.com is well built. My compliments to whoever came up with this easy to use website. It is indeed smooth, nicely laid out, and very, very convincing. For the unsuspecting online user, it would be easy to fall for the scam. A scam, in my opinion, that maliciously and slanderously labels Trump Supporters as “Trollbots,” or worse.
There is no escaping the evidence that clearly shows this site is designed to deceive.
Let’s start with the topic that anchors my entire post. Trump.
Thanks to the likes of CNN, MSNBC, The New York Times, and The Washington Post, President Trump is hated by many people. It has been my experience when you confront Trump Haters with facts they invariably end up calling you names. Little logic and almost no facts to support their arguments are ever given. I am accustomed to seeing in the gutter name-calling and sarcastic deflections.
For God’s sake, the last thing we need is a tool that makes their name-calling look professional, and worse advocates an inauthentic group response from Twitter users.
In case you were wondering why I have focused on the two extremes. Moderates who swing between support and opposition to the President rarely find comfort or a voice on Twitter. I am not the first to observe this and I’m sure I won’t be the last. Although I believe Twitter is furiously working to change that. Which is why I find myself, again, trying to figure out the best way to use Twitter. Extremes struggle with logic and it is in logic, I hope Twitter finds a better footing.
The illogical perspective of a Trump Hater is too complex to explain in this article, but let’s start with the anchor I mentioned earlier. How does this botsentinel website and the underlying bot rate the current sitting President of the United States of America?
The rating is curiously accurate relative to the general view of overall public opinion. The President does Tweet a lot, and over the years he has received some pushback from a lot of people some of them even his followers. So, a 42% rating seems fair, which might make the unsuspecting Trump Supporter begin their relationship with this Bot Sentinel website with some level of trust. For those who dig deeper, it becomes all too clear this is part of the deception.
I think you will agree as you continue to read that this rating is part of the scam. On one hand, it paints the President as someone who Tweets a lot and he doesn’t care much about how people feel about it. On the other hand, the rating is just low enough to keep users from reporting his Twitter account in mass.
After all, the builders of botsentinel wouldn’t want to clog the Twitter reporting system with an account they know will never be suspended. They just want to target their harassment to certain accounts. It might draw attention to their site. Reporting @realDonaldTrump might expose that botsentenil is clearly designed to easily be used to coordinate inauthentic reporting behaviors. They wouldn’t want Twitter to know they too are just another Trollbot, by their own definition.
Designed by a Trump Hater
Before I get into the evidence that shows how the site systematically discriminates against Trump Supporters, let’s have a look at an individual who claims it is all his idea.
The man behind the curtain.
You can find him on Twitter and Facebook. Neatly placed at the top of his Facebook Page you will find the following post, which I grabbed a screenshot of, just in case he deletes it.
I reached out to the man who claims he started this site in a Facebook message, and then an email. Finally, after reTweeting one of his Tweets he responded. You can see the response below. It is safe to say this man does not care for President Trump.
Thread: Thomas is a blogger who contacted me asking for a comment regarding Bot Sentinel. He claimed Bot Sentinel is biased against Conservatives and people who are being classified as bots are actually human. Thomas claims he has conducted extensive research, oh really?… https://t.co/Llv5yio6rc
— Christopher Bouzy (@cbouzy) December 23, 2019
The entire thread…
1. Bot Sentinel is able to classify more than just bots and that is made clear on the website. Being able to classify bots is fairly easy, it’s the inauthentic human controlled accounts and toxic troll account that are much harder to classify. Bots are only part of the problem…
Inauthentic accounts come in many forms, and automated accounts are easy to identify. Many inauthentic accounts use a combination of manual tweeting with some form of automation to make the tweet activity appear authentic. There are also toxic troll accounts who engage…
in targeted harassment and knowingly create and spread disinformation. Bot Sentinel is capable of classifying more than one category of inauthentic and toxic troll accounts. This is explained on the website but people still make the argument “I am not a bot, I am human.”…
2. Bot Sentinel doesn’t target Conservatives, the model wasn’t trained to classify accounts based on ideology…etc. Anyone who has researched Bot Sentinel would quickly notice there are thousands of Left leaning accounts with high scores that are tracked by the platform…
3. Bot Sentinel doesn’t suspend accounts. Our users report accounts with high scores and Twitter makes the ultimate decision if the account in question is violating their rules. If an account is suspended, it’s because Twitter determined that account was breaking their rules…
4. We make all of our data public including the accounts we are tracking. If a high number of Conservative accounts are being suspended, maybe the people behind those accounts should consider stop breaking Twitter’s rules.
5. I have talk to approximately two dozen journalists within the past year. I always go into detail on how Bot Sentinel works, our methodology, and in some cases details on the tech. I am always upfront about my political leanings, which have no bearing on how the platform works.
He was kind enough to share a screenshot of my Tweet, that I deleted and then corrected with a new Tweet to make it appear as though I was being disingenuous.
— Christopher Bouzy (@cbouzy) December 23, 2019
As you can see was quickly corrected only 2 minutes later.
Says the guy running a bot site masquerading as a tool to find bots. Laughable. https://t.co/WaANIC8LBO
— Thomas Dillingham (@tdilling) December 23, 2019
BTW: Someone once said I should've called it Troll Sentinel or something else because we just don't classify bots. Um… the name Bot Sentinel means Autonomous Guard. In other words, we are using a good "bot" to help guard against inauthentic and toxic accounts.
— Christopher Bouzy (@cbouzy) December 23, 2019
The actual bot in front of the man
The website sits between its creator and the users who engage with it. All automated, and based on what?
Let’s begin by taking a look at the word trollbot. A simple google search yields nothing about the word except links to the botsentinal website.
Now that we know this word is a word clearly coined by the creator(s) of botsentinel let’s break that word down and see what meaning an everyday online user might conclude.
When you do a simple google search for the word bot you are greeted with a definition. It is reasonable, then, to assume the definition provided would most likely represent the broad understanding of the word bot.
“an autonomous program on a network (especially the Internet) that can interact with computer systems or users, especially one designed to respond or behave like a player in an adventure game.”
Now that we understand what the vast majority of people understand bot to mean, let’s have a look at the word troll. Again, referring to the most used search engine in the world, let’s see what Google has to say about it. With this word, we find the dictionary word for a troll. Google is also kind enough to provide a side reference to a Wikipedia definition as well. Let’s focus on Wikipedia’s definition of an internet troll because that seems more relevant.
“In Internet slang, a troll is a person who starts quarrels or upsets people on the Internet to distract and sow discord by posting inflammatory and digressive, extraneous, or off-topic messages in an…”
So it is safe for me to say that a TrollBot is something or someone who interacts with the people on the internet to stir up crap often distracting people from the real issue or the facts. Ok, now we have a good idea of what the contemporary understanding of how the word trollbot would be interpreted. Now let’s have a look at what the botsentinel.com website says about trollbots.
We find the definition of this freshly coined word trollbot in the Frequently Asked Questions section of the botsentinel website.
Trollbot is a classification we created to describe human controlled accounts who exhibit toxic troll-like behavior. Some of these accounts frequently retweet known propaganda and fake news accounts, and they engage in repetitive bot-like activity. Other trollbot accounts target and harass specific Twitter accounts as an individual or as part of a coordinated harassment campaign. Ideology, political affiliation, religious beliefs, geographic location, or frequency of tweets are not factors when determining the classification of a Twitter account.
Now things start to get alarming. There is no harm in creating or coining a new word. In fact, it fits what the botsentinel self-proclaimed creator claims it is meant to do. However, the trouble starts when the site admits they are basing their decision on “known propaganda and fake news accounts.” They also describe accounts as “toxic.” There is no reference to a library or website of so-called propaganda or fake news account, and toxic is completely subjective. We are left to assume they make this decision on their own. So, are we to trust the people making that decision? The people who clearly hate President Trump?
They claim that “ideology, political affiliation, religious beliefs, geographic location, or frequency of tweets are not factors,” which I will explain later is an underlying flaw in their system. They admit they make no predilections about how people might Tweet based on what they know about them. That’s like saying I don’t care if a Republican opposes Trump it doesn’t matter. Or pointing out that a person who is clearly a Democrat now supports Trump.
Ignoring the underlying characteristics of users removes the ability to get at the core of what a user might be trying to convey when they Tweet, re-Tweet, or comment on a Tweet. After all, this website implies that it can detect the motives of Twitter users because they claim that those labeled as trollbot might intent to “harass” a specific user. They even go so far as to imply that the trollbots they point out may be part of a larger coordinated campaign.
How can you effectively do that, when you don’t know anything about the individual behind the account? I submit to you that you cannot. Actions speak louder than words, and you may be able to speculate, but you cannot definitively determine the motives of the account holder. Furthermore, how could you possibly detect or determine if the user behind the account is genuine? I mean we know the Russians have engaged in disinformation campaigns while pretending to be both Trump Supporters and Trump Haters. Which speaks to their mistake in refusing to look at the geographic location of the users.
Even Facebook rates its users as Liberal or Conservative for advertising purposes. How else could they target their advertising? I have no idea if Twitter maintains a field in their databases to label their users. On the topic of Twitter, the very platform the botsentinel website monitors, we know that even Conservative employees are afraid to share their views. In Twitter’s defense, they seem to be trying to fix that.
Speaking of Twitter
I will be quoting Jack Dorsey as we move forward, so it is important that you understand the context of his quote. Jack Dorsey appeared before Congress in September 2018. I found his opening statement to be sincere, honest, and frankly humble. The only thing about his appearance that I found disingenuous was his coat and a button-up shirt.
Mr. Dorsey has never struck me as the type of person who wants to be called Mister. Neither does Jack seem like he is comfortable giving his statement. He even admits he is usually shy, which explains his apprehension during the hearing. When I think of Jack Dorsey I think of a man wearing his very comfortable favorite t-shirt, sitting with extremely comfortable house shoes or sandals, while he deliberately and thoughtfully codes a platform that he fears will change the world in ways he cannot predict. All while knowing that if he doesn’t do it someone else will…
That said this is his 5-minute opening statement which is clearly meant for more than Congress. He was speaking to the people of the United States, the world, and the whole of the Twitter community and future Twitter users.
I interrupted my thoughts about botsentinel to inject Jack Dorsey and Twitter to make several points that will self-elucidate as we move forward. In these hearings, the man at the top of Twitter admitted they were struggling to identify bad actors who were abusing their platform. This gives rise to one serious question, what makes Christopher Bousy think he can do better than those armed with the vast resources, and direct access to the actual raw Twitter data?
…back to the bot in front of the man
The automated system creators admit to selecting 2500 trollbot accounts and 2500 regular accounts. This foundation of the entire automated system is based on an apparent bias and subjectiveness in the selection of 5000 accounts.
The first thing I learned about computers when I entered the business in 1994 was computers do what they are told, EXACTLY what they are told. All programs are written by someone. Even if that program is self-learning, it has to start somewhere. In this case, somewhere was 5000 accounts hand-picked by Trump haters.
Furthermore, the system stores those accounts for review each night. They make no mention of how the system is instructed to learn. To what degree are their 5,000,000 tweets graded, filtered, and scored. For that matter how do they arrive at a score? While I respect that their system must be something of a black box, they cannot expect us to believe it is fair when the original founding dataset was established by at least one person who has a clear and obvious bias.
Even the argument put forward by the website has a fault when examined against the original dataset used to form their model.
A blaring contradiction.
The site argues that there are more Conservative accounts being tracked on their website because there are more Liberals using the service to check Conservative accounts. How could they possibly know this if they do not track or classify accounts as Conservative or Liberal?
Finally, it is almost too easy to download a Troll Report to then quickly turn around and upload that report to Twitter. Why not use the Twitter APIs to have the official @botsentinel Twitter handle the reply in the thread? Could it be they know their tool is being used for targeted harassment? Could it be that Twitter finds it more difficult to analyze images? Of course, I’m sure they have gone out of their way not to offend key people at Twitter by skirting to close to the rules, or creating content Twitter might find objectionable.
So, here we are. We have an automated system, also known as a bot. That bot is conveniently set up to encourage targeted harassment and in fact, encourages other people to do so. This is a direct violation of Twitter rules clearly covered under their Abusive Behavior section. In fact, the first two sentences of that section reads:
You may not engage in the targeted harassment of someone, or incite other people to do so. We consider abusive behavior an attempt to harass, intimidate, or silence someone else’s voice.
Surely a bot designed to single out certain users and encourage other Twitter users to report those accounts falls under this category?
If the aforementioned rule isn’t being broken then surely this one is.
Encouraging or calling for others to harass an individual or group of people
We prohibit behavior that encourages others to harass or target specific individuals or groups with abusive behavior. This includes, but is not limited to; calls to target people with abuse or harassment online and behavior that urges offline action such as physical harassment.
They are calling on Twitter users to encourage Twitter to look into the targeted accounts. This could result in Twitter inadvertently banning someone because of the sheer volume of reports. To me, this all seems self-explanatory. We have a website that is tightly integrated with Twitter being used by Twitter users to target Trump Supporters on Twitter.
This is Part one of what has now become a multipart series. Next, I will dive into the accounts the botsentinel site advocates against in detail. I am curious to see what changes are made between now and then. I’ll be back with part two in a few days, how many, I don’t rightly know, but time will tell.
I uncovered more new evidence today, that will further delay my next post.
Finally, I will leave with this thought. If the system is self-learning, who is teaching it, what data set is it learning from? Without a broad set of input data from a broad audience, logic and statistics dictate the system will become more and more like the original data set the bot was provided. There is no way to log into their site using your Twitter account and provide human input to train their bot. So what humans are minding the automated slander system?