Back
Democratizing AI Governance: Anthropic's Collective Constitutional AI Experiment
webCovers Anthropic's 'Collective Constitutional AI' experiment, which engaged ~1,000 Americans to crowdsource AI behavioral guidelines, representing a novel approach to democratic AI governance and public participation in AI alignment decisions.
Metadata
Importance: 52/100news articlenews
Summary
Anthropic partnered with the Collective Intelligence Project, Polis, and PureSpectrum to engage ~1,000 US adults in drafting a crowdsourced 'constitution' for Claude, testing whether democratic public input can shape AI behavioral guidelines. The experiment, called 'Collective Constitutional AI,' builds on Anthropic's earlier Constitutional AI work and aims to reduce the gap between AI developer values and broader public preferences. It represents a potential model for more inclusive AI governance beyond decisions made solely by AI lab employees.
Key Points
- •Anthropic engaged ~1,094 Americans to crowdsource AI behavioral principles, creating a 'Collective Constitutional AI' alternative to internally derived guidelines.
- •The experiment addresses a documented gap between public preferences (slower AI development, distrust of tech leaders) and current AI industry practices.
- •Builds on Anthropic's Constitutional AI framework, which trains LLMs using explicit principles inspired by documents like the UN Universal Declaration of Human Rights.
- •Conducted in collaboration with the Collective Intelligence Project, Polis, and PureSpectrum to ensure methodological rigor.
- •Policy chief Jack Clark framed the goal as developing AI constitutions by third parties rather than solely by lab insiders.
Cached Content Preview
HTTP 200Fetched Apr 10, 202615 KB
Policy & Regulation
Democratizing AI governance: an Anthropic experiment
October 19, 2023
Anthropic, the company that developed the AI chatbot Claude, recently presented a unique idea: why not let ordinary people help shape the rules that guide the behavior of AI?
Engaging around 1,000 Americans in their experiment, Anthropic explored a method that could define the landscape of future AI governance.
By providing members of the public with the ability to influence AI, can we move towards an industry that fairly represents people’s views rather than reflecting those of its creators?
While large language models (LLMs) like ChatGPT and Claude learn from their training data, which drives responses, a great deal of subjective interpretation is still involved, from selecting what data to include to engineering guardrails.
Democratizing AI control to the public is a tantalizing prospect, but does it work?
The gap between public opinion and the AI industry
The evolution of AI systems has opened a rift between the visions of tech bosses and the public.
Surveys this year revealed that the general public would prefer a slower pace of AI development, and many distrust the leaders of companies like Meta, OpenAI, Google, Microsoft, etc.
Simultaneously, AI is embedding itself in critical processes and decision-making systems across sectors like healthcare, education, and law enforcement . AI models engineered by a select few are already capable of making life-changing decisions for the masses.
This poses an essential question: should tech companies determine who should dictate the guidelines that high-powered AI systems adhere to? Should the public have a say? And if so, how?
Within this, some posit that AI governance should be directly handed to regulators and politicians. Another suggestion is to press for more open-source AI models, allowing users and developers to craft their own rulebooks – which is already happening.
Anthropic’s novel experiment poses an alternative that places AI governance into the hands of the public, dubbed “Collective Constitutional AI.”
This follows from the company’s earlier work on constitutional AI , which trains LLMs using a series of principles, ensuring the chatbot has clear directives on handling delicate topics, establishing boundaries, and aligning with human values.
Claude’s constitution draws inspiration from globally recognized documents like the United Nations Universal Declaration of Human Rights. The aim has always been to ensure Claude’s outputs are both “useful” and “harmless” to users.
The idea of collective constitutional AI, however, is to use members of the public to crowdsource rules rather than deriving them from external sources.
This could catalyze further AI governance experiments, encouraging more AI firms to consider outsider participation in their decision-making processes.
Jack Clark, Anthropic’s policy chief
... (truncated, 15 KB total)Resource ID:
f001f98d8a093f5b | Stable ID: sid_1PZZYzg4kM