CHAPTERS
From anti-vax research to spotting coordinated manipulation tactics
Renée DiResta explains how her initial interest in California’s anti-vaccine activism (as a new parent) led her to study how small groups can disproportionately amplify narratives online. She describes early observations of hashtag gaming, automation, and the way algorithms can be deliberately exploited.
The broader context: ISIS, propaganda, and the evolving information ecosystem
DiResta connects her early findings to parallel research on ISIS’s social-media propaganda operations. She outlines how information hops across platforms and how social systems evolved into an ecosystem vulnerable to influence campaigns.
How Russia entered the picture (2015) and why consolidation matters
The conversation shifts to the first major public reporting on the Internet Research Agency (IRA) and early rumblings about Russian manipulation. DiResta argues that consolidation of audiences onto a few dominant platforms made influence operations far easier to execute.
Bots, sock puppets, and “cyborg” accounts: what these operators really are
Joe and Renée dig into how suspicious accounts behave and why “bot” can be misleading. DiResta clarifies that many accounts were run by real people using semi-automation, posing as Americans and adopting trolling tactics to harass and provoke.
Inside the IRA: workplace structure, quotas, cultural training, and micro-targeting rules
DiResta describes whistleblower reporting and investigative findings about the IRA’s day-to-day operations. She explains how employees worked like a marketing agency with intelligence-style tactics, including quotas and highly granular cultural guidance to appear authentic.
The core strategy: build fake ‘tribes’ first, then activate influence later
DiResta explains that across platforms the IRA created identity-based communities (LGBT, Black community subgroups, far-right segments, etc.). Much of the content was non-political relationship-building, with occasional posts that reinforced in-group narratives and later nudged political behaviors.
Scale and optimization: follower counts, hit/flop pages, and repurposing ‘sleeper’ accounts
They discuss how some pages remained small while a few became huge, and how the IRA iterated like a growth team. DiResta shares examples of accounts that were rebranded and reused, potentially as sleeper assets that could be politicized later.
Did it swing the election? Measuring impact vs. missing context
Joe presses on influence and tone shifts; DiResta emphasizes the limits of what outside researchers can conclude. They can see engagement volumes and targeting behavior, but lack visibility into comment context, downstream persuasion, and behavior change (like voting).
Political objectives in the data: pro-Trump, anti-Clinton, and polarization mechanics
DiResta describes the political patterns found in platform-attributed datasets: sustained pro-Trump content on right-leaning pages, strong anti-Clinton messaging, and efforts to fracture moderates. On the left, content often aimed to suppress or redirect votes (e.g., to Jill Stein) and amplify resentment after primaries.
Beyond memes: real-world organizing, protests, and “spycraft” engagement
The discussion turns to the most unsettling element: IRA operatives directly contacted real activists and organized offline events. DiResta details Facebook events, funding offers, and stunts designed to provoke confrontation—showing influence operations can migrate from online manipulation to real-world unrest.
Platform response and accountability: from denial to hearings, notifications, and Section 230 tension
DiResta recounts the early resistance from platforms, followed by public pressure, investigative discoveries, and congressional hearings. They discuss notification efforts to users, the legal framework of Section 230, and the ongoing debate about what platforms should moderate—content, behavior, or coordinated inauthentic activity.
Moderation at scale and the ‘censorship’ feedback loop (Learn to Code, Warren ad, identity debates)
Joe and Renée examine how moderation errors and automation fuel distrust across the political spectrum. They cover keyword-based enforcement pitfalls, advertising policy confusion, privacy regulation (GDPR/California), and why real-identity requirements are not a simple fix.
The next wave: GAN faces, deepfakes, synthetic audio—and the risk of real-world escalation
DiResta warns that detection gets harder as AI generates realistic fake identities, voices, and videos. Joe expands on how synthetic media could trigger panic or irreversible actions, citing examples like false alerts; DiResta references earlier IRA hoaxes as proof of concept for crisis manipulation.
