How bots can help fight online trolls
Loading...
The wonder of internet connectivity can turn into a horror show if the people who use online platforms decide that instead of connecting and communicating, they want to mock, insult, abuse, harass and even threaten each other. In online communities , this has been called 鈥.鈥 More recently it has been called . It happens on many different websites and social media systems. Users have been fighting back for a while, and now the owners and managers of those online services are joining in.
The most recent addition to this effort comes from , one of a few that allow gamers to play video games, stream their gameplay live online and type back and forth with people who want to watch them play. Players do this to show off their prowess (and in some cases ). Game fans do this for entertainment or to learn new tips and tricks that can improve their own play.
Large, diverse groups of people engaging with each other online can yield interesting cooperation. For example, in one video game I helped build, people watching a stream could make , like slowing down or attacking enemies. But of the thousands of people tuning in daily to watch gamer play, for instance, at least some try to overwhelm or hijack the chat away from the subject of the game itself. This can be a mere nuisance, but can also become a serious problem, with racism, sexism and other prejudices coming to the fore in toxic and abusive comment threads.
In an effort to help its users fight trolling, Twitch has developed bots 鈥 software programs that can run automatically on its platform 鈥 to monitor discussions in its chats. At present, Twitch鈥檚 bots alert the game鈥檚 host, called the streamer, that someone has posted an offensive word. The streamer can then decide what action to take, such as blocking the user from the channel.
Beyond just helping individual streamers manage their audiences鈥 behavior, this approach may be able to capitalize on the fact that , as my own research has documented. For instance, a bot could approach people using racist language, and suggest other forms of interaction to change how people interact with others.
In 2015 I was part of a team that created a system that uses Twitter bots to do the activist work of recruiting humans to do social good for their community. We called it .
We used Botivist in an experiment to find out whether bots could recruit people and make them contribute ideas about tackling corruption instead of just complaining about corruption. We set up the system to watch Twitter for people complaining about corruption in Latin America, identifying the keywords 鈥渃orrupcion鈥 and 鈥渋mpunidad,鈥 the Spanish words for 鈥渃orruption鈥 and 鈥渋mpunity.鈥
When it noticed relevant tweets, Botivist would tweet in reply, asking questions like 鈥淗ow do we fight corruption in our cities?鈥 and 鈥淲hat should we change personally to fight corruption?鈥 Then it waited to see if the people replied, and what they said. Of those who engaged, Botivist asked follow-up questions and asked them to volunteer to help fight the problem they were complaining about.
We found that Botivist was able to encourage people to go beyond simply complaining about corruption, pushing them to offer ideas and engage with others sharing their concerns. Bots could change people鈥檚 behavior! However, we also found that some individuals began debating whether 鈥 and how 鈥 bots should be involved in activism. But it nevertheless suggests that people who were comfortable engaging with bots online could be mobilized to work toward a solution, rather than just complaining about it.
Humans鈥 reactions to bots鈥 interventions matter, and inform how we design bots and what we tell them to do. In research at New York University in 2016, doctoral student Kevin Munger online. Calling out Twitter users for racist behavior ended up reducing those users鈥 racist communications over time 鈥 if the bot doing the chastising appeared to be a white man with a large number of followers, two factors that conferred social status and power. If the bot had relatively few followers or was a black man, its interventions were not measurably successful.
Bots鈥 abilities to affect how people act toward each other online bring up important issues our society needs to address. A key question is: What types of behaviors should bots encourage or discourage?
It鈥檚 relatively benign for bots to notify humans about specifically hateful or dangerous words 鈥 and let the humans decide what to do about it. Twitch lets streamers decide for themselves whether they want to use the bots, as well as what (if anything) to do if the bot alerts them to a problem. Users鈥 decisions not to use the bots include both technological factors and concerns about comments. In conversations I have seen among Twitch streamers, some have described disabling them for causing interference with browser add-ons they already use to manage their audience chat space. Other streamers have disabled the bots because they feel bots hinder audience participation.
But it could be alarming if we ask bots to influence people鈥檚 free expression of genuine feelings or thoughts. Should bots monitor language use on all online platforms? What should these 鈥渂ot police鈥 look out for? How should the bots 鈥 which is to say, how should the people who design the bots 鈥 handle those Twitch streamers who appear to enjoy engaging with trolls?
One Twitch streamer posted a :
... lmfao! Trolls make it interesting [...] I sometimes troll back if I鈥檓 in a really good mood [...] I get similar comments all of the time ... sometimes I laugh hysterically and lose focus because I鈥檓 tickled....
Other streamers even enjoy to trolls:
... My favorite was someone telling me in Rocket League "I hope every one of your followers unfollows you after that match.鈥 My response was 鈥淢y mom would never do that!鈥 Lol....
What about streamers who actually want to make racist or sexist comments to their audiences? What if their audiences respond positively to those remarks? Should a bot monitor a player鈥檚 behavior on his own channel against standards set by someone else, such as the platform鈥檚 administrators? And what language should the bots watch for 鈥 racism, perhaps, but what about ideas that are merely unpopular, rather than socially damaging?
At present, we don鈥檛 have ways of thinking about, talking about or deciding on these balancing acts of freedom of expression and association online. In the offline world, people are free to say racist things to willing audiences, but suffer social consequences if they do so around people who object. As bots become more able to participate in, and exert influence on, our human interactions, we鈥檒l need to decide who sets the standards and how, as well as who, enforces them in online communities.
鈥 is Assistant Professor of Computer Science, .
鈥 on .
