Javier Hernandez The AI-focused startup Anthropic has garnered insights from approximately 1,000 U.S. citizens concerning the foundational principles that should guide artificial intelligence. In partnership with the Collective Intelligence Project (CIP), a nonprofit entity dedicated to aligning technology with the greater good, this endeavor has laid the groundwork for a “AI constitution.” This initiative aims to investigate the extent to which democratic participation can impact the trajectory of AI development.
Table Of Contents
Anthropic and CIP Collaborate to Develop AI Constitution Through Citizen Engagement
Anthropic, the firm behind the Claude chatbot, collaborated with the Collective Intelligence Project to harness the views of roughly 1,000 American individuals for the construction of an AI governance document. Claude was initially governed by a set of principles formulated by Anthropic’s staff, using their proprietary Constitutional AI (CAI) method, which ensures that large language models operate according to overarching ethical guidelines. These internal guidelines drew inspiration from seminal texts, such as the United Nations’ Universal Declaration of Human Rights.
Earlier this week, Anthropic released a blog post detailing the constitution developed through public consultation. The post also described the results of training a novel AI system in accordance with these publicly-determined guidelines, using the CAI approach. Backed by Amazon, Anthropic elaborated on the venture’s motivation:
The goal was to probe the potential impact of democratic mechanisms on AI evolution. Through this exercise, we identified points of agreement and divergence between our internally-developed constitution and the public’s perspective.
Utilizing Polis, a data analytics platform designed to gauge public sentiment, both Anthropic and CIP invited a diverse sample of American citizens to propose or vote on normative principles that should govern large language model-based chat agents.
The project revealed that approximately half of the principles in the publicly-generated constitution were consistent with Anthropic’s internal version. Notable differences in the public’s suggestions included the emphasis on balanced and objective information, as well as accessibility and adaptability for persons with disabilities.
Anthropic further noted public propositions that were excluded from the final constitution due to a lack of consensus. These include contrasting opinions on whether the AI should favor collective well-being over individual rights or prioritize individual liberty and personal responsibility over communal interests.
The Collective Intelligence Project concluded its report on the experiment by stating, “The resultant public model exhibited fewer biases across multiple stereotypes and showed comparable performance to the baseline model in metrics such as mathematical ability, natural language comprehension, and overall utility and safety.” CIP further stressed the critical importance of including public viewpoints in the formation of AI conduct, particularly as AI becomes increasingly integrated into daily life and communication.
What are your opinions on the role of public engagement in training AI systems? We welcome your perspectives in the comments section below.
Frequently Asked Questions (FAQs) about AI constitution
What is the main purpose of Anthropic’s AI constitution project?
The primary aim of Anthropic’s AI constitution project is to investigate how democratic participation can influence the development and ethical guidelines of artificial intelligence. Through collaboration with the Collective Intelligence Project, Anthropic has sought public input to draft a constitution that would govern the behavior of AI systems.
Who are the key stakeholders involved in this initiative?
The key stakeholders in this project are Anthropic, an AI startup, and the Collective Intelligence Project (CIP), a nonprofit organization. Additionally, around 1,000 American citizens participated in the initiative, contributing their views on the principles that should guide AI.
What method does Anthropic use to ensure AI follows ethical guidelines?
Anthropic uses a proprietary method called Constitutional AI (CAI) to ensure that large language models operate according to overarching ethical guidelines. This method was initially used to formulate a set of internal principles that governed the behavior of their Claude chatbot.
What platform was used to gather public input?
Polis, a data analytics platform designed to collect and analyze large-scale public sentiment, was used to gather the views of approximately 1,000 American citizens. Participants could propose or vote on normative principles for AI behavior.
What were some notable differences between the public’s and Anthropic’s constitutions?
One significant difference was the public’s emphasis on the provision of balanced and objective information. Another was a focus on making AI systems accessible and adaptable for individuals with disabilities.
Were there any conflicting principles that were excluded due to lack of consensus?
Yes, certain conflicting viewpoints did not make it into the final public constitution. These include opinions on whether the AI should prioritize collective welfare over individual rights and preferences, or vice versa.
What was the Collective Intelligence Project’s final assessment of the public model?
The Collective Intelligence Project concluded that the public model showed fewer biases across a range of stereotypes and performed comparably to the baseline model in evaluations focused on mathematical ability, natural language comprehension, and overall utility and safety.
Why is public input considered crucial in shaping AI behavior?
Public input is considered crucial because AI is becoming increasingly integrated into various aspects of daily life, work, and communication. Including diverse viewpoints in its governance can lead to more ethical and universally acceptable AI systems.
More about AI constitution
- Anthropic Official Website
- Collective Intelligence Project
- United Nations’ Universal Declaration of Human Rights
- Polis Platform
- Amazon Corporate
- AI Ethics Guidelines
- Democratic Processes in Technology
8 comments
Loving the public’s focus on accessibility for ppl w/ disabilities. Often overlooked but super important.
What’s this Constitutional AI (CAI) method? Sounds intriguing but I couldn’t find much info on it. Anyone got more deets?
Really interesting read! It’s about time someone involved the public in AI development. I mean, we’re the ones gonna be using it, right?
so were these 1000 people really representative of the whole population? Hope it’s not just tech-savvy folks who got a say.
Great start, but this is just the tip of the iceberg. AI’s gonna change everything, so better we have a say now than regret it later.
Not sure I like the idea of AI prioritizing ‘collective good.’ Sounds too Orwellian to me.
They talk abt ‘less biased’ AI, but can AI ever be completely unbiased? Makes you wonder.
Anthropic’s onto something here. Glad to see they are considering ethics and not just diving headfirst into AI development. Props to them.