How to spot a bot in the comments section
Many people lead their lives becoming very upset at comments they read on the internet, and spend a good deal of time trying to understand what those comments mean about the state of society and humanity. Often, this is time wasted, because they're not comments that any human is genuinely making. They're bots.
One reason people don't expect many human-seeming comments to be bots is because the assumption is that bots cannot act like humans. Different types of bots are able to take on different types of human behavior and some are very lifelike. Let's take a moment to understand how they achieve this, and to understand how bots exist at all.
Types of Bots
This is not an exhaustive list, these are just bot tactics that are commonly employed in comment threads.
1. Mechanical Turks
A mechanical turk is a person who is hired to perform a routine action. The name "mechanical turk" comes from a story about a robot that would perform for the royal court in Turkey that was, in fact, a box with a person inside it who would essentially be the robot's brain and muscles. An army of modern mechanical turks can be hired from anywhere in the world now. A common task is for them to solve Captcha problems. Sometimes when a company says that they have an AI that can analyze images, they're really hiring people to just look at the images and sort them the old-fashioned way, then claiming they have a bot that does this. Mechanical turking is something that can be applied to any type of bot where programming isn't producing the behavior needed or the AI technology for it isn't available.
Also, this is the easiest way for a bot to pass a turning test--it's easy to think the other end isn't a computer when it really is a human that can make human decisions. The "bot" aspect is the scripted limitations that the person's interactions will be forced through to meet the standards of the job. For instance, holding strange personal views in a conversation, claiming illogical things, being involved in ludicrous conspiracy theory discussions--these are things a person can do for money without ever holding the view. You can often tell it's this type of bot when a person seems to be committed to arguing one way or to promoting a certain view, but their actual words and sentences don't hold a lot of internal logic, and they seem to easily lose track of the conversation. They're not personally invested in the conversation and it shows in their behavior. They'll often repeat boilerplate arguments that are designed to incite rage or other volatile emotions.
2. Grouped bots (balanced bot party)
A crowd of bots makes for more confusion than a single bot, because all of the sudden, there's threads of conversation between multiple people who all seem to agree with each other, or seem to always have the same "type" of conversation. Investment forum plants tend to have this type of dynamic where one bot pitches the perfect setup for another bot to really start up a discussion. Then you'll have a couple other bots who agree with whatever point is trying to be pushed, trying to project the image of crowd agreement and encourage human posters in the forum to join the artificially-created bandwagon. These are often tell-taled by their homogeny and genericness. They all sound the same, there's conversations in the comments repeating and always going the same way and repeating phrasing.
3. Cuckoo chicks
Some bots are operated in stages, and the early stage will usually be one where they blend in and behave like the rest of the chicks from the nest. They repeat popular community memes or ask very basic questions to look as if they regularly participate in discussions if one were to glance at their profile's post history. (This is tooled for forums that work like Reddit.) They do this for months or even years, depending on what the goal is and how long-term the operation is that they're part of. This "nesting" of profiles on strategic discussion forums is usually in preparation for a mass discussion event in the future where the person who hired the nested bots has the bots all engage in coordinated actions. This can be anything from flooding boards with spam to presenting the appearance that a community holds a different view than it really does by planting so many bots that the demographic appears shifted.
Two different ways to spot a bot
1. Judge the logic
Programmed AIs or hired actors both face a fundamental problem-they are forced into certain patterns of behavior by their roles. A clown cannot deliver a newscast. A scripted spambot can't convince you that it cares about what it's saying. And a person following a script can't decide to act in a different fashion on the fly, that's not their job. So if there's just plain illogical interactions, people acting like a whole conversation was there that never took place, entire phrases and words missing, typos that don't make sense for people to type and glyphs/characters that aren't common for people of the forum's dominant languages to write in, you're looking at a bot-intense discussion. Threads often become a mixture of fooled humans genuinely replying to scripted humans and bots. Usually bots are encouraged to sow confusion, so the incoherence of the resulting conversations would be considered a fringe benefit.
If someone is espousing a very extreme view uninvited and unprompted in a place that makes no sense to do so, then a programmer may be testing out a bot. Someone found some discussions in PornHub video comments sections of several users all talking about their favorite stocks and using Reddit stock board memes, and of course it had nothing to do with the video. Either a developer was testing their bots there or they left that tab open on accident while launching something. Small discussion forums for niche subjects are sometimes used as testing grounds for operations, as you can easily start an argument in a fandom to test out bot behavior and no one will think much of it.
Testing is often necessary for bots because AIs can use a vocabulary bank you can swap out, so you can switch your racist bots to transphobic bots for different operations. So the word-swap glitch may show up in speech, people using vocabulary that's strange to that discussion forum but otherwise speaking coherent sentences.
You don't want to test the logic of the arguments themselves, because humans can easily hold and argue for illogical opinion-based views. It's more about testing whether this is a person speaking with functional language, or whether they're making sentence-building mistakes that humans just don't make, even when learning a second language. Sentences may appear entirely constructed by copy-paste.
2. Question the motivation
I like to evaluate this by asking: "Why go through all the trouble of this?" Does the person look as if they're spending a lot of time arguing on a very specific subject, especially across all of the comments they make? Could you estimate that they spend a good deal of time every day on this, or even notice by the timestamps that they're doing it at odd times for their user demographic? Why would a person get up at 3AM to argue about veganism and anticapitalism with fifty different people on YouTube and then leave three hours later abruptly in the middle of all discussions? Why would this person be meticulously copy-pasting their garbled sentences all that time? And why are none of the videos they're commenting on actually mentioning the subjects they're ranting about?
With mechanical turks it's harder to spot language errors but easier to spot motivation errors. They're often overly focused on projecting a cover story and drop unnecessary details about what they're trying to build up cover on. They often repeat popular views and phrases and stick with the bandwagon when it comes to social interactions, but never seem to have a fundamental, core sense of why they're there having the conversation, because they're just being paid to be there and they can't forget that they're on the job and they don't actually care about this discussion. They can seem overly purpose-driven. Their lives and backstories tend to be laden with sympathetic details. (Sometimes you can even spot them living out a fantasy like saying they met a celebrity the forum respects.)
Sometimes you can simply analyze for tone and look at the context. Someone being irrationally angry in all their phrasing, for instance, may be a bot's script limitations. Looping back on logic repeatedly is often a result of script limitations. There's only so many ways that a bot, or even an actor, can be scripted to act. I find the conclusion of a discussion is the most telling--bots and mechanical turks either never stop even when it makes all rational sense to, because they either don't get tired or aren't actually feeling the opinions they're scripted to have so they're not nearly as taxed by a debate and can go on indefinitely. It can be easy to get worn out by this, but if you look at it from a different angle, it's really a tell.
Motivation is usually about profit. Who profits from this? Was the discussion about vapes? The meat industry? Vaccines? Factory workers' rights in China? Is it something a certain government or political party really cares about? If the dots are easy to connect, then that's another tell.
Bots are a part of the internet and always have been.
Hey, back in the 90s in the Sailor Moon fanpage webmaster community, people were spamming Guestbooks with fake discussions about the different webpage review groups they didn't like to try to stir up gossip in the community. That's basically a mechanical turk attack. Back then, it was pretty obvious that people were planting opinions, once you realized the greater context of the situation. And nowadays, it's still easy to spot planted opinions and bot discussions, once you know what's going on.
Who hires these? Who creates them? Well, there's a lot of contractor organizations who do this for politicians, governments, and corporations, and there's plenty of freelance programmers willing to take jobs on the grey market. Some tactics were developed by government espionage and security groups, some were developed on the open web in the wild by the current of social dynamics. I feel a lot of media attention that ought to be put on this sort of thing has been lacking because of a technical knowledge gap and a lack of understanding of basic espionage that the Sailor Moon fandom's teenagers sure seemed to grasp easily enough back in 1993.
If you're concerned, I suggest writing to your local representative about the issue. But most of all I hope you allow a little more hope for humanity. Some of those ugly discussions out there are just robots being paid to insult each other, which is sad, but it's a different sad than if they'd really meant it.
Why bother arguing with someone who isn't really there?
This is a long one, but is definitely worth the read. I have seen some of the examples provided here all over the place. Particularly "mechanical turks".
