Process overview

# Goal

Our project yields three outcomes:

The first goal is to summarize the results of the deliberative process involving hundreds of Japanese people on the "AI-related topics" for humans. This is what the grant requires.

The second goal is to verify a new methodology of deliberation through this deliberation, based on science fiction prototyping. This must be applicable to more than AI-related topics. Members need it strongly from past civic tech activities.

Thirdly, we aim to utilize the information gathered from our deliberative process to create an AI that reflects the culture and thinking of the Japanese people. We intend to make this AI available worldwide, enabling people to converse with it in their own languages. This approach will foster cross-cultural understanding that transcends language barriers, a prerequisite for global deliberation.

# Process Overview: Iterative Science Fiction Prototyping

We stimulate participants' imagination and provide a foresight of the future through science fiction. This allows participants to envision suitable rule sets for AI.

Our process involves: initial story writing lectures, iterative story creation with LLMs and feedback collection. Stories and feedbacks are compiled into vector index.

# Sharing of Experience by Science Fiction

Deliberative discussions necessitate a shared knowledge base, which can be challenging if the subject is not of interest. We need to spark people's interest and curiosity to initiate the democratic process.

We use science fiction prototyping to foster imagination about the future. This approach mirrors clay prototyping in product design, but employs text to create science fiction narratives, immersing participants in potential social and technological contexts.

We craft diverse stories covering a spectrum from utopia to dystopia. This diversity dispels reader bias. Engaging with these stories allows participants to vicariously experience future scenarios, offering them a basis to contemplate and shape their own future. This spectrum from utopia to dystopia opens up avenues for individuals to deliberate on their present actions to better influence the future.

# Prototyping Support by LLMs

In this project, we are considering the creation of a system to support prototyping with LLMs: the creation of messy prototypes with LLMs stimulates subsequent improvement actions, which ultimately results in higher quality output at lower cost than if done by humans alone.

It is important to make this prototyping process something that can be done lightly and repeatedly with the assistance of LLMs. These stories will be improved in an agile cycle. Stories are the equivalent of the Minimum Viable Product in a Lean Startup. The story is the minimum viable product that can be released to the public and elicit a reaction from them.

LLMs can empower non-professional writers to create stories by supporting various stages of the writing process, including idea generation, structuring, and the actual writing. This inclusive nature of LLMs allows all participants, even readers, to share their worldviews.

Instead of having experts list complex facts, we use stories as prototypes. Experts highlight factual discrepancies in these stories, which are then corrected. After several iterations, we create numerous expert-supervised stories. By posing questions like "Having read this story, what rules do you feel should be made?" on Polis, we can solicit comments and facilitate voting.

# Iterating communication between writers and readers

After stories are written, they are published on the social platform. Visitors can participate in various ways: read, share, comment, vote(Polis like system), and write other stories.

People are diverse. There are active people who want to output by themselves and passive people who want to read and react to others' output. This diversity should not be ignored. We should not ask passive people to do a lot of output. We need different approaches for many preferences of participation.

The main focus of feedback is to gather sentiments. It is a startpoint of deliberation. Question using pol.is is an effective way. Is the future described in the story favorable? What are your interests and disadvantages in this future? What can you do to reach or avoid the future depicted? These questions are visualized as different perspectives and the participant’s position will be clear. It attracts participants to further discussion in a more clear and constructive way.

By gathering the emotions of a large number of people from diverse perspectives, a "future in which more people are happy" becomes clear. Productive discussions about how to reach that future can then take place. Ideas for a better future will be generated.

We take an iterative approach. Writers receive participants’ responses and discussions and write better stories for deliberative discussion. Improved stories can gather better responses.

# Belonging to a Larger Story

Contributors, whether they are prototypers, fact-checking experts, or readers providing feedback, receive NFTs as a proof of their contribution. This cultivates a sense of belonging to a larger story: "OpenAI Granted Project on the Future of AI".

# Shared External Brain: Vector Index. One of our outcomes is AI.

As a result of an iteration, we will have stories, expert facts, readers' feedback, chat logs, all compiled into a comprehensive vector index. This compilation not only serves as a dynamic resource, providing multiple viewpoints and a rich context for further iterations, but also promotes mutual understanding among the public. By making our collective knowledge easily accessible and searchable, it allows us to trace back to previous discussions, better understand the evolution of ideas, and synthesize more accurate and encompassing results in our AI-related deliberations. The process fosters a deeper understanding of each other's perspectives and contributes to a more collaborative and informed approach to problem-solving.

Also, we publish a vector index to make new AIs for facilitating mutual understanding. The vector index search will also be made accessible via APIs, allowing those who can write programs to contribute their own ideas to the project.

This vector index will be one of the deliverables of this project. Although this project will be conducted in Japanese, non-Japanese speakers around the world will be able to use ChatGPT connected to this vector index to converse with chatbots that incorporate Japanese culture and thinking.