Source-linked AI summary
Social Bots: Human-Like by Means of Human Control?
Christian Grimme, Mike Preuss, Lena Adam, Heike Trautmann
TL;DR
The paper addresses divergent definitions and uncertain capabilities of social bots in public communication. It proposes a broad definition, reviews bot operation and limitations, and discusses human integration as a way to extend and control bot capabilities for human interaction.
Problem
Social bots are influential in public discourse, yet the term lacks a common definition across scientific disciplines.
Method
The paper proposes a balanced definition, reviews how social bots operate and their technical limitations, and discusses integrating humans into bot processes.
Results
The paper concludes that hybrid social bots can bypass important limitations and that human integration is a promising way to realize effective interaction with humans.
Takeaways & Limitations
Social bots should be analyzed as a broad, context-dependent category whose capabilities and effects depend on communication, purpose, and human control.
Takeaways & Limitations
Chat bots remain limited by their scripts and databases, while the proposed social-bot definition is intentionally a high-level concept.
Abstract
from arXiv · showhide
Social bots are currently regarded an influential but also somewhat mysterious factor in public discourse and opinion making. They are considered to be capable of massively distributing propaganda in social and online media and their application is even suspected to be partly responsible for recent election results. Astonishingly, the term `Social Bot' is not well defined and different scientific disciplines use divergent definitions. This work starts with a balanced definition attempt, before providing an overview of how social bots actually work (taking the example of Twitter) and what their current technical limitations are. Despite recent research progress in Deep Learning and Big Data, there are many activities bots cannot handle well. We then discuss how bot capabilities can be extended and controlled by integrating humans into the process and reason that this is currently the most promising way to go in order to realize effective interactions with other humans.
Definition and Taxonomy of Social Bots
The paper addresses divergent definitions of social bots by proposing a high-level concept covering different semi-automatic agents, communication modes, purposes, and online environments.
- Definition and Taxonomy of Social Bots: Social bots are inconsistently defined across technical and social-science perspectives, with definitions variously emphasizing automation, human imitation, communication, or political influence.These differences can contradict each other or exclude bot classes included by other definitions.
- Definition and Taxonomy of Social Bots: The paper defines Social Bot as a superordinate concept for different types of semi-automatic agents designed for specific purposes through one- or many-sided communication.The definition includes agent behavior and communication rather than restricting social bots to a single technical form.
- Definition and Taxonomy of Social Bots: The proposed concept allows fully automated and partly human-steered action, autonomous agent-like behavior, goal orientation, multiple communication modes, and all online media.It retains elements of earlier definitions while extending the scope beyond social media alone.
- Definition and Taxonomy of Social Bots: The paper distinguishes social bots from content, game, and other non-social bots by requiring communication or interaction with humans.The taxonomy is illustrated through examples of included and excluded bot types.
Social Bots
Social bots range from scripted chat bots and assistants to spam and political bots, with capabilities determined by their scripts, data, goals, and operator control.
- Social Bots: Chat bots communicate one-to-one in natural language but are limited by their scripts and databases, so they are often restricted to specific topics.They are used in customer service, help desks, education, business, and e-commerce.
- Social Bots: Spam bots use one-to-many communication to distribute information, advertisements, or phishing links without involving recipients in interaction.Their messages may be sent on behalf of companies, groups, or individuals.
- Social Bots: Political bots spread political content, participate in online discussions, boost ideas or people, and may intervene in conversations using keywords and propagandistic content.Their conversational or non-reactive behavior depends on the operator’s aim, technical skill, and code.
- Social Bots: Nearly 19 percent of election-related Twitter posts during the 2016 U.S. presidential election were made by bots.The passage presents this as evidence of the scale of bot activity in political communication.
- Social Bots: Mobile phone assistants translate natural-language voice commands into phone actions or search results through voice recognition and keyword identification.They represent human-to-machine communication rather than autonomous public discussion.
Bots Not Regarded as Social Bots
The paper excludes bots that substitute users, manage content, or handle machine-to-machine tasks without pronounced human communication or interaction.
- Bots Not Regarded as Social Bots: Game bots automate gameplay tasks such as navigation, training, farming, cheating, or standing in for unavailable players, focusing on user substitution rather than interaction.Their functions vary from assistance to completing repetitive duties.
- Bots Not Regarded as Social Bots: Curator bots collect and manage content for presentation to humans but work silently with content rather than communicating with users.Wikipedia bots that correct typos, generate links, or remove whitespace exemplify this class.
- Bots Not Regarded as Social Bots: Content, game, and service-level-agreement bots are excluded when they substitute users or communicate machine-to-machine rather than interacting socially with humans.The defining boundary is the absence of a human communication or interaction aspect.
Discussion
The discussion treats social bots as a broad category of automated social interaction whose diverse purposes require precise, context-sensitive analysis.
- Discussion: Social bots automate social interaction through publicly visible posts, chats, comments, direct messages, and other online communication media.Any online medium supporting human communication can provide a connection point for social bots.
- Discussion: Social bots should be understood as a high-level concept because they differ according to the reasons for which they were built.Their uses extend beyond influencing people and must be discussed in their specific contexts.
- Discussion: Social bots’ diverse tasks create different technical and economic considerations, so influence-oriented examples should not define the entire category.The discussion links classification to the specific purposes and contexts of bot deployment.
- Discussion: The growing availability of group bots and bot repositories suggests that social bots may become increasingly pervasive in internet experience.The paper therefore calls for more precise discussion of their possibilities for influencing users and societies.
Automation using Social Bots
Social bots serve purposes ranging from advertisement to propaganda, creating technical and economic trade-offs. The paper presents a low-cost progression from a simple implementation toward moderately complex, human-like Twitter behavior and evaluates 30 bots.
- Social-bot applications span advertisement to propaganda, creating both technical challenges and economic considerations.
- The paper develops a simple social-bot realization and extends it toward moderately complex, human-like behavior on Twitter while keeping costs low.
- An experiment using 30 Twitter bots connects the implementation to economic analysis.
A simple reactive Twitter Bot example
A reactive Twitter bot listens to the public Stream and uses the REST API to reply to applicable posts. Because Twitter exposes only a subset of activity, the bot may receive roughly 1% to 40% of actual traffic.
- The bot registers a Twitter Stream Listener and triggers an actuator that replies through the Twitter REST API.Figure 1 presents the components and their data flow.
- Twitter bandwidth management means only a subset of posts reaches the listener, typically about 1% to 40% of actual traffic.The passage notes that higher coverage may occur only in very restricted cases.
- The presented social bot can react directly to Twitter posts from its listening pipeline.
Functionality.
The bot consumes Twitter posts matching selected hashtags or topics and can trigger standardized responses. Examples include greetings, weather retrieval, question answering, and fixed political statements.
- The bot filters the Twitter Stream by hashtags or topics, allowing adaptation to a specified context or domain.
- Its actuator can be extended from greeting users to standardized services such as weather forecasts and topic-specific question answering.Question answering can use the Google Knowledge Graph as an external knowledge source.
- In political contexts, the bot can respond to topics with fixed political statements, usually independently of the posted content.
- A simple implementation of no more than 30 lines of code demonstrates substantial potential for social-bot applications.
- The implementation examples illustrate how a reactive bot can support demonstrations, greetings, information services, and political messaging.
Costs.
A medium-experienced developer can implement the simple service bot in under an hour, with most effort devoted to account and API setup. However, its instant, continuous, restricted behavior makes automation readily detectable.
- Implementation time is lower than one hour for a medium-experienced developer, excluding the provided error-handling code.
- Most development effort concerns creating a Twitter account and obtaining API access, including mobile-number connection for developer access.
- No human recipient would consider a message from this bot to have been sent by another human.
- The bot’s instant reactions, permanent activity, and restricted content analysis expose it as automated.
A Social Bot with human-like behavior
Human-like social bots must produce credible content, leave human-like traces, and build balanced follower networks while varying their timing and activity. The paper presents temporal and behavioral imitation as a partial solution, but credible intelligent content remains an open challenge.
- Credible and “intelligent” content accepted by humans remains the central open challenge for social bots.
- Human-like bots should imitate day-night cycles, rest periods, realistic activity levels, and variable timing rather than act continuously.
- Larger follower networks increase the reach of an account’s distributed content, making network growth a core bot objective.
- A follow-for-follow strategy grows a bot’s network by following users, retaining reciprocal connections, and dissolving nonreciprocal ones after a time window.
- Strongly imbalanced follower ratios can reveal prominent or advertising accounts and provide signals for bot detection.
Extending bot functionality.
The simple social bot can be extended through independent Twitter actuators coordinated by a profile that encodes behavioral guidelines and randomized action timing. These components collect candidates, manage reciprocal following, and publish tweets, retweets, pictures, and comments.
- The simple social bot can be extended with independent actuators that perform specific Twitter actions and address network-growth and activity challenges.
- CollectionActuator listens to the Twitter stream and selects follow candidates using follower ratios, activity, and tweet properties.
- BotProfile stores behavioral constraints and guidelines, including rest periods, posting, retweeting, and following, while random noise varies the next action time.
- FollowActuator manages follow-for-follow behavior, adds users who follow back, and cancels or blacklists nonresponders after 24 hours.
- PostActuator and PictureActuator publish tweets, retweets, pictures, and matching comments from stored databases according to BotProfile settings.
Experimental evaluation.
The experiment tested hybrid Twitter bots across follower growth, coordinated content distribution, and identity disclosure, combining automated behavior with human-generated content. The bots grew networks and deceived users, but failed to establish a significant trend; BotOrNot also struggled to identify them reliably.
- Follower-network growth: About 1350 followers were acquired in eight productive days through an automated follow-for-follow strategy.The follower network increased continuously during the evaluation period.
- Coordinated content distribution: The bots briefly placed a promoted hashtag in the German top 100 trending topics but did not establish a significant trend.Human actors devised the content, while the bots coordinated its distribution.
- Human responses: Many human followers were deceived by the fake identities and actions, producing reactions ranging from disappointment and anger to amusement and disbelief.The experiment revealed that fairly realistic bot behavior could mislead human users.
- Implications and limitations: The authors characterize the experiment as a snapshot and identify adaptive, real-time, human-inclusive detection as a future requirement for hybrid bots.They argue that detection should consider content, background strategies, and distributed narratives rather than relying solely on automatic signals.
- Automation and human control: The experiment found that tedious network-building, posting, and retweeting could be automated without exposure as bots, while content remained human-provided.The framework deployed the same code across thirty accounts with individualized profiles; 27 bots continuously followed users automatically without human intervention.