Social bot

A social bot (also: socialbot or socbot) or troll bot is an agent that communicates more or less autonomously on social media, often with the task of influencing the course of discussion and/or the opinions of its readers. It is related to chatbots but mostly only uses rather simple interactions or no reactivity at all. The messages (e.g. tweets) it distributes are mostly either very simple, or prefabricated (by humans), and it often operates in groups and various configurations of partial human control (hybrid). It usually targets advocating certain ideas, supporting campaigns, or aggregating other sources either by acting as a "follower" and/or gathering followers itself. In this very limited respect, social bots can be said to have passed the Turing test. If social media profiles are expected to be human, then social bots represent fake accounts. The automated creation and deployment of many social bots against a distributed system or community is one form of Sybil attack.

Social bots appear to have played a significant role in the 2016 United States presidential election and their history appears to go back at least to the United States midterm elections, 2010. It is estimated that 9-15% of active Twitter accounts may be social bots and that 15% of the total Twitter population active in the US presidential election discussion were bots. At least 400,000 bots were responsible for about 3.8 million tweets, roughly 19% of the total volume.

Twitterbots are already well-known examples, but corresponding autonomous agents on Facebook and elsewhere have also been observed. Nowadays, social bots are equipped with or can generate convincing internet personas that are well capable of influencing real people, although they are not always reliable.

Social bots, besides being able to (re-)produce or reuse messages autonomously, also share many traits with spambots with respect to their tendency to infiltrate large user groups.

Using social bots is against the terms of service of many platforms, especially Twitter and Instagram. However, a certain degree of automation is of course intended by making social media APIs available.

The topic of a legal regulation of social bots is becoming more urgent to policy makers in many countries, however due to the difficulty of recognizing social bots and separating them from "eligible" automation via social media APIs, it is currently unclear how that can be done and also if it can be enforced. In any case, social bots are expected to play a role in future shaping of public opinion by autonomously acting as incessant and never-tiring influencer.

Uses

Lutz Finger identifies 5 immediate uses for social bots:

• foster fame: having an arbitrary number of (unrevealed) bots as (fake) followers can help simulate real success
• spamming: having advertising bots in online chats is similar to email spam, but a lot more direct
• mischief: e.g. signing up an opponent with a lot of fake identities and spam the account or help others discover it to discreditize the opponent
• bias public opinion: influence trends by countless messages of similar content with different phrasings
• limit free speech: important messages can be pushed out of sight by a deluge of automated bot messages
• to fish passwords or other personal data

The effects of all points can be likened to and support methods of traditional psychological warfare and information warfare.

Detection

The first generation of bots could sometimes be distinguished from real users by their often superhuman capacities to post messages around the clock (and at massive rates). Later developments have succeeded in imprinting more "human" activity and behavioral patterns in the agent. To unambiguously detect social bots as what they are, a variety of criteria must be applied together using pattern detection techniques, some of which are:

• cartoon figures as user pictures
• sometimes also random real user pictures are captured (identity fraud)
• reposting rate
• temporal patterns
• sentiment expression
• followers-to-friends ratio
• length of user names
• variability in (re)posted messages
• engagement rate (like/followers rate)

Botometer (formerly BotOrNot) is a public Web service that checks the activity of a Twitter account and gives it a score based on how likely the account is to be a bot. The system leverages over a thousand features. An active method that worked well in detecting early spam bots was to set up honeypot accounts where obvious nonsensical content was posted and then dumbly reposted (retweeted) by bots. However, recent studies show that bots evolve quickly and detection methods have to be updated constantly, because otherwise they may get useless after a few years.

See also

• Astroturfing
• Crowd manipulation
• Fake news website
• Internet bot
• Marketing and artificial intelligence
• Messaging spam
• On the Internet, nobody knows you're a dog
• Post-truth politics
• Search engine manipulation effect
• Social spam
• Sockpuppet (Internet)
• Sybil attack
• Technoself studies
• Twitter bomb
• Votebots
• Whispering campaign

External links

• The Computational Propaganda Research Project University of Oxford
• What is a Social Media Bot? | Social Media Bot Definition Cloudflare