Tracking Political Manipulation Through Social Media - Samantha Bradshaw

Definition and Evolution of Bots

• A "bot" is defined as a script or piece of code performing automated tasks, ranging from beneficial functions (e.g., Google's web crawlers indexing the internet) to manipulative behaviors.
• Manipulative bots are designed to mimic human behavior on platforms like Twitter and Facebook to artificially amplify popularity through likes, shares, and follows.
• Sophisticated bots integrate natural language processing to interact with real users in comment threads, blurring the line between automated content and human dialogue.
• While media focus intensified during the 2016 US election, research indicates government agencies have used these techniques for social control and opinion shaping in authoritarian regimes since the early days of social media.
• Tactics are evolving from crude, high-volume posting (e.g., over 50 tweets daily) to sophisticated algorithmic gaming using specific keywords to trend topics or manipulate YouTube and Google search rankings.
• Content creation strategies have shifted from random "spray and pray" approaches to resource-intensive identification of societal "buttons" to generate targeted, high-impact media.

Platform-Specific Manipulation and Vulnerability

• Manipulation strategies adapt to local platform dominance: Twitter and Facebook are primary vectors in the US, whereas WhatsApp is the central focus in India due to its usage rates.
• WhatsApp presents unique research challenges as a closed platform, limiting academic ability to monitor private group communications.
• In Brazil, WhatsApp disinformation campaigns utilized memes and mobilizing images to evoke emotional responses, similar to Reddit's "Pepe" culture in the 2016 US election.
• Twitter facilitates bot infiltration through an open API, lack of real-name requirements, and tools like Hootsuite that allow easy automation.
• Facebook enforces stricter identity verification (potentially requiring government ID), making fake accounts harder to create but more powerful once established due to lower user skepticism.

Corporate Incentives and Regulatory Responses

• Platforms historically lacked incentives to remove fake accounts because a larger user base increased market valuation and advertising inventory, despite ad revenue being devalued by non-human engagement.
• Advertisers are now realizing the reality of bot traffic, creating a market shift where the value of user metrics is being questioned.
• Germany's 2017 network law (NetzDG) mandates the removal of illegal content within 24 hours, prompting immediate takedowns of both hate speech and legitimate criticism, leading to "collateral censorship."
• Governments using content-based laws to silence dissent are a significant risk, with authoritarian regimes adopting similar frameworks to target journalists and suppress opposition.
• Experts argue that enforcing transparency regarding platform algorithms and operations is more effective than direct content censorship, though complete algorithmic transparency could inadvertently aid bad actors in gaming the system.

Empirical Findings on Disinformation

• 2016 analysis of US Twitter and Facebook data revealed a 1:1 ratio of "junk news" (low-quality, fake, or non-journalistic content) to professionally produced information shared by users.
• The 2018 midterm election analysis showed an increase in this ratio to 1.2 or 1.3 to 1, indicating a rise in junk news sharing despite platform interventions.
• In 2016, "swing states" showed a higher concentration of junk news sharing compared to uncontested states, suggesting targeted botnet activity in competitive districts.
• International data from the UK, Germany, France, Sweden, and Mexico shows significantly lower ratios of junk news to professional news shares compared to the US, identifying the US as a "traumatic case."
• Trends suggest disinformation will likely increase leading up to the 2020 US election due to massive financial investment in campaign media strategies and experimental innovation in manipulation techniques.

Future Outlook and Societal Implications

• Deepfakes are expected to have limited global impact by 2020 due to existing detection technology developed by agencies like DARPA, though they may remain potent in low-literacy media environments.
• Social media consolidation (e.g., Facebook owning Instagram, WhatsApp, and Messenger) creates a locked-in ecosystem that amplifies disinformation risks by eliminating competition and forcing total reliance on a single model.
• Researchers advocate for "digital privacy" practices, such as restricting app permissions and limiting data collection, as a defense against targeted disinformation.
• Polarization in the US has widened over the last 20 years, prompting a call for users to practice "digital civility" and recognize shared humanity despite ideological differences.
• Optimism exists regarding growing government engagement with technology issues, exemplified by policymakers like Senator Elizabeth Warren and UK's Damian Collins who are actively researching proper regulatory frameworks.
• There is a concern that upcoming Canadian elections (2019) may see the importation of US-style anti-immigration rhetoric and populist narratives, suggesting no nation is immune.