AI & Collective Intelligence
From deliberative democracy to corporate strategy: Can GenAI power collective decision-making?
This course is designed to deconstruct the "black box" of LLMs and explore their concrete application in group decision-making processes. It is neither a pure coding class nor a philosophy seminar. It is an engineering approach applied to social sciences: how do technical constraints (context window, hallucinations) influence the quality of democratic or strategic debate?
Syllabus
1. Introduction: the Taiwan story
Storytelling of digital democracy in Taiwan. From the Sunflower Movement to vTaiwan: how civic hackers and Audrey Tang used tech to resolve complex legislative deadlocks (Uber case).
2. Under the hood: transformers & probability
Technical demystification. From Next Token Prediction to the Transformer architecture (2017). Understanding why AI doesn't 'think' but computes probabilities.
3. Why we can't just 'ask the LLM'
Analysis of the limitations of simply feeding all opinions into an LLM: context window, hallucinations, loss of weak signals ('Insight gap'), and inherent political bias ('Fairness gap').
4. Is agreement the only goal?
Critique of models seeking 'Common Ground' (e.g., DeepMind's 'Habermas Machine'). Introduction of the 6 democratic values framework (Lazar et al.) to evaluate if AI strengthens or dilutes democracy.
5. A framework for augmented deliberation
Proposal of a 4-step protocol (Framing, Participation, Deliberation, Outcome). Concrete application: from citizen assemblies to corporate 'Open Strategy' (Vision 2035).
Presentation Deck
The full deck (62 slides) including technical diagrams, case studies, and bibliographic references is available for open access.
Download PDF