Social bots – the tech­nol­o­gy behind fake news

What is a social bot? This term was used par­tic­u­lar­ly fre­quent­ly in con­nec­tion with online dis­cus­sions about Brexit as well as Donald Trump being elected as US president. Social bots are con­sid­ered opinion robots that influence dis­cus­sions in social networks. But is their influence really that great? What exactly is a social bot and how do they work?

Social bots simulate human users on social networks. If they do not indicate that they are machines then they are deemed fake accounts. Many users are misled, since they see be­liev­able in­for­ma­tion on a profile and assume that they are con­vers­ing with people. This type of bot is often used to spread opinions on social networks or to stimulate dis­cus­sion in the interest of their operators.

Companies sometimes use social bots for marketing purposes – for example, to fake how popular certain products are in order to encourage more people to buy them. They also sometimes write positive product reviews that are fake. They can also serve political purposes, since they influence public dis­cus­sions by rep­re­sent­ing the views of certain parties or politi­cians. To discredit a political opponent, social bots often spread in­for­ma­tion that isn’t true. It is therefore hardly sur­pris­ing that this tech­nol­o­gy is often discussed in con­junc­tion with emotive words such as 'fake news', 'hate speech', 'filter bubbles', or 'feedback loops'.

The po­lar­iz­ing effect of social bots affects jour­nal­ists and media companies as well as hu­man­i­ties and social sci­en­tists. Fur­ther­more, due to the in­creas­ing im­por­tance of social networks, a separate industry has now developed, which earns money by de­vel­op­ing and using social bots. Using social bots is legal, making social bot tech­nol­o­gy on the internet a lucrative business.

However, you shouldn’t turn your nose up at the tech­nol­o­gy so quickly. After all, it can be used in a mean­ing­ful way without nec­es­sar­i­ly spreading false in­for­ma­tion or ma­nip­u­lat­ing other people’s opinions – for example, as a chat program on company websites that answers customer questions.

To un­der­stand the dangers of social bots and how to recognize them, you need to know how bot tech­nol­o­gy works. This is followed by a brief outline of what a bot is and an ex­pla­na­tion of the bot types that exist.

A bot (shortened form of 'robot') is an automated program that is pro­grammed for certain actions and executes them either regularly or re­ac­tive­ly. The bot does this without needing human ac­ti­va­tion. It analyzes the en­vi­ron­ment and 'decides' which actions to take depending on the situation.

The different types of bots sometimes differ greatly when it comes to their technical com­plex­i­ty. The spectrum ranges from simple al­go­rithms to highly complex AI tech­nol­o­gy. If a bot works with ar­ti­fi­cial in­tel­li­gence, it is even capable of learning: it explores its sur­round­ings and adapts to them. Depending on the function, a bot isn’t usually rec­og­nized by humans and performs its tasks unnoticed in the back­ground, or appears as a human being (i.e. imitates human behavior). Some types of bots are described below in more detail:

• Web crawler: They belong to the category of bots that work com­plete­ly unnoticed. Usually they are used by search engines to browse the web, analyze websites, and then enter this in­for­ma­tion into search di­rec­to­ries. They usually act as 'in­no­cen­t' within generally accepted standards (such as the Robots Exclusion Standard). Other web crawlers operate beyond these standards and collect unau­tho­rized data on the net.

• Chat bots: In contrast to web crawlers, chat bots are reactive i.e. they react to human ac­tiv­i­ties and spe­cial­ize in re­spond­ing to other chat par­tic­i­pants in a mean­ing­ful way. In everyday life, chat bots are primarily en­coun­tered as digital as­sis­tants. For example, a website assistant can guide visitors through the website or answer questions about the website’s topic or what it offers. Language as­sis­tants such as Siri or OK Google or external language as­sis­tants such as Amazon Echo or Google Home are also based on this chat bot tech­nol­o­gy.

• Virtual computer players: Virtual computer games also require virtual actors who react flexibly to human beings. These op­er­a­tions are performed by bots. They are called non-player char­ac­ters (role-playing games), aim bots (action games), poker bots (online poker) and so on. These bots are reactive and work in­creas­ing­ly with ar­ti­fi­cial in­tel­li­gent tech­nol­o­gy. A good example is Google’s ar­ti­fi­cial in­tel­li­gence, which is called AlphaZero. It excelled in Chess as well as the board game, Go. In­tel­li­gent tech­nolo­gies like these are also used in computer games – in the form of bots.

• Social bots: These are the bots that are used secretly in social networks. They are both repet­i­tive and reactive: They like, comment, retweet, and try to provoke or involve others in con­ver­sa­tions and dis­cus­sions. They fake a human identity so that users react naturally to them.

The web crawler and the chat bot are con­sid­ered the two basic types of bot. Current bot types often combine their basic functions: the hidden data analysis from the web crawler and human com­mu­ni­ca­tion sim­u­la­tion from the chat bot. Social bots also rely on these two basic functions.

A social bot is an automatic program that simulates human behavior on social networks. Social bots take part in dis­cus­sions on Twitter or Facebook and act like human users. They spread content on a par­tic­u­lar topic on social media – mostly for the purpose of in­flu­enc­ing people’s opinions.

Social bots are usually used for marketing or political purposes. It is not uncommon for social bots to spread fake news. To sway public opinion, a social bot uses tech­niques that are typical of bots and are also used by other bot types. They search social networks for dis­cus­sions on pre-defined topics (data eval­u­a­tion just like web crawlers) and in­flu­ences them as a virtual con­ver­sa­tion par­tic­i­pant (sim­u­la­tion of con­ver­sa­tions like with a chat bot).

In terms of their function, social bots are very similar to chat bots or digital as­sis­tants – they are used to com­mu­ni­cate with people. But there is one crucial dif­fer­ence: while chat bots are usually like a con­sult­ing service, which helps the dialog partner, social bots are supposed to deceive and ma­nip­u­late. And while chat bots can fulfill their function even if they are perceived as technical programs, deceiving other net par­tic­i­pants is essential for social bots to influence public opinion.

There are many examples of the ma­nip­u­la­tive use of social bots. In 2016 and 2017 alone, the use of social bots was reg­is­tered in almost all major public elections. In par­tic­u­lar, there were extensive dis­cus­sions about social bots with regard to their influence on the Brexit vote, the pres­i­den­tial elections in the US, as well as the par­lia­men­tary elections in France and Germany.

• Brexit vote: In June 2016, the majority of British citizens decided to leave the EU. Prior to this, there were heated dis­cus­sions on social networks – and it was noted that many social bots were also involved. The In­de­pen­dent reported that social bots played an important strategic role, es­pe­cial­ly when it came to voting 'leave'.

• US pres­i­den­tial election: In November 2016, Donald Trump was elected the 58^th President of the United States. There was a lot of in­for­ma­tion on how much influence social bots had on his narrow election victory. According to Oxford Uni­ver­si­ty, automated pro-Trump bots over­whelmed pro-Clinton messages. Ap­par­ent­ly every third pro-Trump tweet was from a bot. There was also a fake news report that the Pope had rec­om­mend­ed Trump for election and this was shared almost a million times – including by social bots. But the use of pro-Clinton social bots was also reg­is­tered.

• German par­lia­men­tary elections: In light of what happened the previous year in the UK, many people were worried that social bots might influence the federal elections in 2017. As a result, all par­tic­i­pat­ing parties spoke out against their use in the election campaign, even though social bots are not illegal in Germany. Luckily there wasn’t much meddling on the part of social media bots. Due to the rel­a­tive­ly small number of Twitter users in Germany, however, their reach was also smaller, which is why fewer bots are used in Germany anyway.

The influence of social bots on the outcome of the elections was con­tro­ver­sial­ly discussed. In par­tic­u­lar, the Brexit vote and the sur­pris­ing election of Trump was written about for months in the press. Many people suspected that social bots were used as secret voting as­sis­tants before these elections took place. The effects of social bots and fake news are obvious: they lead to a great loss of trust in digital com­mu­ni­ca­tion.

A social bot usually posts using a fake account – with its own profile photo, posts, and a good amount of followers or 'friends'. The social bot uses this account to dis­trib­ute its marketing messages or political state­ments. This can be done via likes and retweets or in the form of posts or comments. Using a pro­gram­ming interface (API), a social bot can access social networks and receive and send data. Social bots usually operate at times when other users are more active. In addition, they usually post at varying intervals to give the idea of being human when in fact a machine is behind all the posts. A social bot can also send friend requests. If a request is confirmed by a human user, the social bot can then collect and analyze the data of the user. As early on as 2011, a Canadian study showed that social bots are able to collect data and analyze the account in­for­ma­tion from users that have accepted their friend request.

Many social bots are pro­grammed with simple al­go­rithms based on simple 'if…then…' state­ments: If a relevant topic has been iden­ti­fied, then the social bots will post the pre-pro­grammed content. To find relevant topics, social bots work with simple keyword searches and scan Twitter timelines or Facebook posts for specific wording and hashtags. Then they publish pre-written texts as state­ments or try to steer con­ver­sa­tions in a certain direction.

However, there are also social bots that are tech­ni­cal­ly much more complex. With the help of ar­ti­fi­cial in­tel­li­gence, com­pre­hen­sive data analysis, and text analysis, these in­tel­li­gent social bots succeed in con­stant­ly gen­er­at­ing new comments that differ from the previous ones. Sometimes these bots can even refer to the current day’s events. They usually assemble their posts from different online texts, which they simple rearrange. These more complex social bots are more difficult to expose.

However, robots do not really work ef­fi­cient­ly until they are connected with each other: If many bots co-ordinate with each other in a so-called botnet, they dis­trib­ute in­for­ma­tion even more ef­fec­tive­ly. For example, social bots can then like and share posts written by other social bots. The more accounts there are, the more their influence grows.

To develop a simple social bot, you don’t need any special technical skills: You can create your own social bots without any prior pro­gram­ming knowledge, as long as you use the right tools. It is just as easy to access fake user accounts, since they can be easily recreated using gen­er­a­tors available online – or the operator can simply buy existing fake accounts. Even the control software for these accounts can now be purchased online. Via a pro­gram­ming interface, the bot is then given access to Twitter or Facebook, where it reacts to pre-defined hashtags or keywords. The tech­nol­o­gy is easily available, which con­tributes sig­nif­i­cant­ly to the rapid spread of social bots.

However, dis­tri­b­u­tion is also made easier by social media itself, since Facebook and Twitter de­lib­er­ate­ly keep their pro­gram­ming in­ter­faces rel­a­tive­ly easy to access to encourage ap­pli­ca­tion de­vel­op­ers to continue working on new software for their platforms. But this also means that social bots have no trouble taking advantage of this. Twitter is es­pe­cial­ly easy to access, which is why most of the bots can be found there.

A social bot’s basic functions are always the same, but can be filtered into three cat­e­gories depending on their functions: the over­loader, the trend­set­ter, and the auto troll.

1. The over­loader: This is a bot that literally floods an online con­ver­sa­tion with its comments. It posts the same state­ments over and over again, causing other posts to get pushed out of sight. The over­loader is only really effective if it works with other bots. When a botnet overloads a post with likes and comments, human users quickly lose control of the dis­cus­sion. This makes it then im­pos­si­ble to exchange content this way.
   
   
2. The trend­set­ter: Trend­set­ters also work best in teams: If a large number of social bots seize a hashtag together, they can easily spread posts relating to the topic of the hashtag. If the topic ends up trending on Twitter or Facebook, it might be picked up by the press. This way, the trend­set­ter bot distorts the actual relevance of the chosen topic area: They ensure that marginal phenomena act like ground­break­ing trends, or small groups like large social movements.
   
   
3. The auto troll: The auto troll operates alone. It tries to distract users who express them­selves on a certain topic and tries to involve them in a con­ver­sa­tion. It usually does this with con­tra­dic­to­ry state­ments, which causes the user to retaliate. This distracts the user from the original con­ver­sa­tion, and the new con­ver­sa­tion becomes polemic and heated. With this method, bots can easily prevent users from ex­chang­ing content.

It is difficult to determine who is actually behind a social bot. So far, there is no method that can ac­cu­rate­ly identify fake accounts. Therefore, it’s even harder to track down the re­spon­si­ble operator. However, there are roughly four groups that can benefit from using social bots:

1. Social marketer/in­flu­encer: Smaller and larger companies can use social bots for covert marketing. As an in­flu­encer, you want to use these bots to start trends and influence them. Target group in­for­ma­tion can also be obtained via a social bot, because if you confirm a bot’s friend request, they will have com­pre­hen­sive access to the data stored in your profile.
   
   
2. Political figures: Lobby groups or political figures are also suspected of using social bots. For example, American secret services suspect that Russian hackers were behind many fake accounts and social bots during the US campaign. Whether the attack came from criminals or the Russian gov­ern­ment, however, is unclear.
   
   
3. People in­ter­est­ed in public opinion-forming: There are users that want to influence the opinion of others via social bots. These can be in­di­vid­u­als, groups, or­ga­ni­za­tions, or criminals. The third group – a col­lec­tion of people that are hard to identify – is probably the largest of the groups mentioned here. Those involved use bots to benefit a party or to raise awareness of certain issues – or simply to cause trouble. Spreading left-wing or right-wing extremist content is quite often the goal. Since this group is so het­ero­ge­neous, it’s not so easy to name one specific intention.
   
   
4. Those with no rec­og­niz­able interests: There are a number of similarly 'harm­less' social bots, which, for example, leave a crazy amount of likes on Star Wars comments. Bots of this kind do not serve any rec­og­niz­able political or economic purpose. Pre­sum­ably, those behind the bots just like to troll others.

However, there is dis­cus­sion about how well social bots carry out their tasks. After all, many experts agree that social bots are not doing their job very well and therefore have little influence on social media users. An article from UX Magazine explains why bots are overrated since they only have limited response ranges and provide rigid answers, which gives them away. There is a consensus that the work of social bots needs to be re­searched more thor­ough­ly. It can therefore be assumed that social research will provide more precise findings in coming years. For pre­ven­ta­tive reasons, a more detailed in­ves­ti­ga­tion of the technique would make sense: While many bots are currently still easy to unmask, it doesn’t mean tech­ni­cal­ly-advanced bots are. This would increase the potential of bots. It is therefore necessary to develop solution strate­gies at an early stage to be able to react to technical ad­vance­ments.