Jigsaw CEO Yasmin Green on AI and using technology for good

Jigsaw's Core Mandate and History

• Jigsaw is a semi-autonomous unit within Google established in 2010 by Yasmin Green and Jared Cohen to address the intersection of the internet and geopolitics.
• The unit was founded proactively before major events like the Arab Spring, the Snowden revelations, or Kremlin election meddling to study how governments might use the internet for surveillance and censorship.
• The name "Jigsaw" reflects the philosophy that technology is merely a single piece that can contribute to both good and bad outcomes.
• The group conducts interdisciplinary ethnographic research, including direct engagement with self-identified "trollers" and former terrorists to understand radicalization drivers.
• Key areas of focus include building technology to promote free expression, bypassing national firewalls, and securing free elections.

Product Innovations and Public Application

• Perspective API: A widely used free tool that analyzes billions of comments daily to help publishers identify and manage toxic content.
  • Recent iterations shift focus from simply removing toxicity to "upranking" comments based on constructive traits like curiosity, personal anecdotes, and well-explained arguments.
  • Pilot data from 146 publishers in the US and Canada indicates that ranking comments for these constructive attributes increases user engagement and subscriptions.
• Bowling Green Pilot: Jigsaw launched a "sense-making" AI tool in Bowling Green, Kentucky, a town of 70,000 residents.
  • The tool synthesizes large-scale public conversations into data visualizations and summaries for policymakers.
  • To prevent AI hallucinations, the system uses "grounding," linking every summary point back to specific source statements for auditability.
• Information Literacy Research: Studies with Gen Z reveal a shift from "information literacy" to "information sensibility," where users evaluate truth by triangulating headlines and comment sections rather than reading articles first.
  • The research notes that comment sections often reinforce polarization by highlighting the loudest voices or validating extreme in-group behaviors.

AI Trajectory and Technical Realities

• AGI vs. Superintelligence: Green defines Artificial General Intelligence (AGI) as an AI capable of human-level performance across all domains, while Superintelligence surpasses all humans combined.
  • She views AGI as achievable within the current trajectory but notes the "jagged frontier" problem, where AI excels at complex tasks (e.g., writing obscure sonnets) while failing at simple ones (e.g., counting strawberry hours).
• Reasoning over Retrieval: Advancements are currently driven by AI reasoning capabilities, where models deconstruct problems step-by-step rather than acting as "all-knowing oracles."
  • A demonstration using Google's Gemini model showed the system explicitly outlining its reasoning stages to generate relevant conversation topics for a specific host and audience.
• DeepSeek Analysis: Green characterizes DeepSeek's recent high-performance models not as a fundamental breakthrough in architecture, but as sophisticated optimizations built on top of Western frontier models and open weights.
  • She uses a "speedboat with a surfboard" analogy: the major Western labs provide the "speedboat" (innovation and infrastructure), while others can "surf" to achieve close performance without leading the innovation.

Workforce and Societal Impact

• Entry-Level Transformation: Green predicts that traditional entry-level jobs will vanish within five years as AI automates foundational tasks, forcing organizations to shift from "workforce-centric" to "workflow-centric" organizational structures.
• Interpersonal Dynamics: AI is increasingly being tested as a coaching tool to mediate workplace conflicts and improve human relationships rather than replacing human interaction.
  • A Cisco HR pilot used AI to diagnose communication loops between managers and reports, revealing mutual feelings of being unheard.
• Global Inequality Concerns: While AI boosts productivity across skill levels now, concerns exist (citing economist Daron Acemoglu) that the benefits may skew toward wealthy nations and corporations, potentially exacerbating global divides.

Future Risks and Strategic Focus

• Primary Concern: The most significant emerging anxiety is the public sentiment that AI is "happening to" people rather than "for" or "with" them.
  • This represents a shift from traditional content harms (trolling, radicalization) to a fear of human subservience to technology.
• Strategic Imperative: Jigsaw aims to counter this by prioritizing human connection and ensuring technology aligns with user needs, despite the competitive pressure to accelerate AI development regardless of societal readiness.