Policy Synth is a Typescript class library for creating multi-scale AI agent logic flows, API's and state of the art, realtime, AI focused web applications.
Policy Synth aims to help governments and citizens make better decisions together by seamlessly integrating collective and artificial intelligence. Utilizing our top-rated community engagement solutions, over 50 x LLM agent types, and genetic algorithms. The mission and passion behind Policy Synth represents a distinctive effort to intertwine human insights with artificial intelligence in addressing complex policy dilemmas. We are currently developing a state-of-the-art platform where policymakers, citizens, and AI can partake in collective discourse.
Policy Synth is already having real life impact: https://www.fastcompany.com/91001497/ai-faith-in-democracy
This agent flow automates and scales GovLabs' crowdsourcing approach to channel expert collective intelligence into actionable policy solutions. It aims to streamline the generation of innovative solutions for complex policy challenges.
Policy Synth Engineer is an automated programming tool designed for Typescript projects, utilizing multi scale AI agents to streamline coding tasks such as feature development, bug fixes, and refactoring. It automates extensive web research for coding assistance and is optimized for Typescript due to its syntactical structure and popularity, significantly speeding up programming tasks compared to traditional methods.
The RAG Chatbot powers the Rebooting Democracy initiative by utilizing hundreds of documents to inform its responses. It serves to engage users effectively by providing well-sourced, context-rich information.
This agent flow uses AI to analyze extensive data sets, helping to identify and recommend changes to barriers in skills-first policies in US states. It focuses on enhancing legal frameworks to support skills-based education and employment initiatives.
Policy Synth utilizes "Your Priorities", a platform we've been refining since 2008. It's recognized as the world's leading citizen engagement tool.
- Rated #1 on PeoplePowered: 2024 Platform Ratings
- Top ranking in the 2024 Digital Democracy Report
- Top listing in OECD Guidelines for Citizen Participation Processes
Policy Synth makes use of All Our Ideas, Citizens Foundations open source, human, pairwise voting platform, to provide guidance and input into different AI agents.
Policy Synth is flexible and can work with many other citizen engagment tools, for example Decidim and Consul. As a class based library Policy Synth has no limitations regarding integration with other tools.
In the context of computational systems, AI agents are autonomous or semi-autonomous entities that interact with digital environments to achieve specific goals or tasks. These agents can range from simple rule-based algorithms to complex systems like Large Language Models (LLMs), capable of processing and generating text and performing simple reasoning. The essence of AI agents lies in their ability to perceive their environment through data, make decisions based on this data, and act upon these decisions to fulfill predefined objectives. In Policy Synth AI agents are used extensively for different tasks.
Fast (System 1) and Slow (System 2) thinking, terms popularized by psychologist Daniel Kahneman, offer a useful framework for understanding how AI, especially Large Language Models (LLMs), parallel System 1 thinking, characterized by swift, intuitive cognition without deliberate effort—much like a human's quick calculation of "what is 2+2?". While the best LLMs are also capable of limited form of System 2 thinking then this analogy suggests that LLMs excel in producing responses quickly, leveraging vast databases of information to generate answers that seem instinctive, much as System 1 thinking does for humans in certain contexts. System 2 thinking represents a more deliberate, analytical form of reasoning that humans employ for more complex tasks, such as solving "what is 17*24?" without the aid of calculators. This level of cognitive processing requires significant effort, attention, and mental manipulation of information.
In contrast to striving for fully autonomous System 2 capabilities with AI, Policy Synth approaches the challenge by manually constructing and integrating System 2 thinking processes through a multi-scale AI agent framework. This methodological choice allows for the deliberate orchestration of AI processes that implement a deeper System 2 thinking by combining specialized agents in a structured manner. Since GPT-4o Policy Synth has introduced Engineer that has limited scope of autonomy in the context of automatic programming.
Draft Policy Synth Developer Guide
The first test run is entirely automated, except for the problem statement provided by us. All sub-problems, entities, and solutions—complete with pros and cons—are generated using GPT-4 and GPT-3.5. These models run highly specialized agents that are, at the moment, very basic and therefore reliable, and should not be confused with ongoing, and exciting, LLM autonomous agent experiments. The context for solutions is obtained through curated web searches. This setup allows us to explore how human and AI-driven insights can work together to solve complex problems.
We're also experimenting with genetic algorithms to foster a co-evolution of solutions between AI and citizens. We use large language models to implement random mutations and crossover, creating a vibrant and diverse pool of potential policy solutions. Fitness ranking is performed by both humans and AI, with the human ranking weighting higher.
To further enrich our evolving solution space, we introduce fresh random immigration through solutions acquired via web searches. This process infuses new ideas and perspectives into our evolving policy generation, enhancing its capacity to tackle a wide range of issues.
For the real-world application of our platform, a series of expert and mass citizen crowdsourcing stages will be integrated. These stages are designed to accompany the AI-generated and ranked options. This multifaceted approach combines collective human insights, expert opinions, and machine intelligence. Our model leverages technology to harness collective wisdom, effectively facilitating the development and implementation of well-informed policy decisions.