Amplifying Human Creativity and Problem Solving with AI Through Generative Collective Intelligence

Thomas P. Kehler, Scott E. Page, Alex Pentland, Martin Reeves, John Seely BrownarXiv 2025 · DOI · arXiv

Introduction

The rapid advancement of Large Language Models (LLMs) has sparked a debate centered on the "standalone agent" fallacy—the idea that AI will eventually replace or simply augment individual human productivity in isolation. However, this paper proposes a paradigm shift toward Generative Collective Intelligence (GCI). Rather than viewing AI as an independent super-intelligence, GCI reframes it as a cognitive bridge designed to facilitate and scale human collaboration. By synthesizing human causal reasoning and creative intuition with AI’s immense computational speed and pattern recognition, GCI offers a framework to address "wicked" problems—complex societal challenges like climate change and healthcare equity that remain intractable for either humans or machines alone.

The GCI framework acts as a cognitive bridge, linking human wisdom with AI capabilities through collective reasoning.

The Core Challenge: Bridging the Intelligence Gap

The authors identify a fundamental disconnect in current AI applications. Humans excel at reasoning based on causal theories and limited data, often relying on intuition and ethics. Conversely, AI operates on probabilistic patterns derived from massive datasets but lacks contextual understanding and "why" reasoning.

Furthermore, traditional human group decision-making is often hampered by "combinatorial complexity"—the difficulty of managing numerous subgroups—and social frictions such as groupthink or status hierarchies. GCI addresses these issues by providing a structured platform where these two distinct forms of intelligence can interact effectively, overcoming the limitations of individual human cognition and the lack of agency in standalone AI.

Methodological Pillars of GCI

The GCI framework is built upon two primary psychological and mathematical foundations:

  1. The Law of Comparative Judgment: Drawing on the work of L.L. Thurstone (1927), GCI avoids subjective absolute scoring. Instead, it utilizes pairwise comparisons (e.g., "Is Idea A more viable than Idea B?"). By applying maximum likelihood estimation, the system converts these ordinal rankings into precise cardinal scores, offering a more accurate reflection of group preference.
  2. The Minimum Regret Principle: Borrowed from reinforcement learning’s multi-armed bandit problems, this principle manages the "exploit-explore" dilemma. A "Collective Reasoning Agent" orchestrates collaboration by identifying disparate perspectives that share enough common ground to communicate, ensuring the group explores novel ideas without losing focus on proven concepts.

To further enhance deliberation, GCI employs Identity Masking and Structured Facilitation. Masking contributor identities ensures that ideas are evaluated on their intrinsic merit rather than the social status of the proposer, fostering an environment of "psychological safety."

Key Insights: AI as Social Technology

GCI redefines AI as a "social technology"—a platform that facilitates human-to-human interaction at an unprecedented scale. The goal is to catalyze both super-additivity (where a group outperforms its best individual member) and emergence (where a group performs tasks no individual could achieve alone, akin to a surgical team or a jazz ensemble).

A notable advantage of this approach is its energy efficiency. The human brain operates on approximately 15 watts. By coupling this low-power creative engine with the high-power processing of Generative AI, GCI creates a sustainable and potent problem-solving architecture. This synergy enables conceptual blending, allowing the integration of diverse mental spaces—such as combining indigenous ecological knowledge with modern climate modeling—to generate entirely new solutions.

Real-World Impact and Case Studies

The paper highlights several successful applications of the GCI framework:

  • Startup Viability: Early implementations via the CrowdSmart platform demonstrated high accuracy in predicting the survivability of startups by aggregating expert collective intelligence.
  • Biotech V2.0: In a collaborative initiative involving insurers, pharmaceutical companies, and patients, GCI shifted the perception of healthcare from an "expense" to an "asset class." This led to the innovative proposal of a healthcare derivatives market to hedge risks.
  • Civic Engagement: The framework aligns with platforms like Polis and MIT City Science, which use structured deliberation to find consensus in complex urban planning and policy-making scenarios.

Limitations and the Path Forward

Despite its potential, GCI faces several hurdles. It remains highly dependent on active, high-quality human participation. There is also a "logical disconnect" within many organizations that desire solutions to hard problems but are hesitant to let AI design the processes used to solve them.

To encourage broader adoption, the authors are transitioning the CrowdSmart platform to open-source code and an open API. Future success will depend on maintaining transparency and trust; the "Collective Reasoning Agent" must provide a clear audit trail and accurate attribution of ideas to ensure all participants feel valued and heard.

Summary updated

BibTeX
@misc{kehler2025amplifying,
  author = {Thomas P. Kehler and Scott E. Page and Alex Pentland and Martin Reeves and John Seely Brown},
  title = {{Amplifying Human Creativity and Problem Solving with AI Through Generative Collective Intelligence}},
  year = {2025},
  eprint = {2505.19167},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2505.19167},
}
Figure