Notice: Polkadot has migrated to AssetHub. Balances, data, referenda, and other on-chain activity has moved to AssetHub.Learn more
ChatGPT integration with Polkadot proposals to improve governance
Below is a summary of our planned treasury proposal for a ChatGPT integration with Polkadot proposals to improve governance. Your feedback is much appreciated.
Requested Amount: $115,200
Short description: This integration of ChatGPT in Polkadot's governance aims to enhance engagement by training the AI model to summarize proposals, answer questions, provide contextual information, and summarize conversations.
Detailed Proposal: https://docs.google.com/document/d/1HthB4yB7eGsbdiDm638Ag1R-HG1n8_tE9Vq-ATgQU_k/edit?usp=sharing
Anyone participating in decentralized governance knows keeping the participants engaged and informed can be challenging. The amount of information and context in each proposal is difficult to digest, long conversations occur in many places, and governance rules can be complex and easy to forget.
This ChatGPT integration aims to improve governance and engagement in Polkadot and serve as a POC for AI in decentralized governance participation by training OpenAI's ChatGPT model to summarize proposals, respond to specific questions, and provide contextual information like voting status, trends and practical info on how to participate. In addition, it would summarize conversations and provide context. Finally, we would enable opt-in reminders that keep DOT holders engaged and informed about governance and participation opportunities.
Here is a video showing how the proposal integration could work:
** In Polkadot direction, Wei Tang asked us to look into LLaMA and Alpaca LLM; we are working on it and will share the findings here.
** Payments would be against delivered milestones.
Use cases
Polkadot's decentralized governance model is arguably the best of any blockchain, but like any democratic system, it presents challenges for users regarding understanding, engagement, and staying informed. Some of these challenges include
**The complexity of proposals: ** Treasury proposals can be intricate and long. Users may struggle to comprehend the nuances and implications, discouraging them from participating in discussions or voting.
Information overload: There is a high volume of information surrounding each proposal, as well as ongoing discussions across multiple platforms (e.g., Element, Polkassembly), can be overwhelming for users. Keeping up with the constant flow of information is time-consuming, which can lead to disengagement or uninformed participation.
Governance rules: The rules and processes involved in Polkadot's decentralized governance model can be complex and difficult to grasp. Understanding the mechanics of voting, proposal submission, and the roles of various actors can be a barrier to participation. The model is not static, as we move from Gov1 to Gov2 having tools to educate stakeholders will be important.
Onboarding and support: Newcomers to the Polkadot ecosystem may find navigating the decentralized governance landscape difficult. Providing adequate resources, guidance, and support for new users is crucial for maintaining an engaged and growing community.
Notification and reminders: Keeping the community informed about ongoing proposals, voting deadlines, and other governance events is essential for fostering engagement. However, managing notifications and reminders across various platforms can be challenging, leading to missed opportunities for participation.
Multilingual and global collaboration: Polkadot's community is distributed across different languages and regions. Tools to drive adoption must bridge the language gap
Who does this solution help?
This proposal benefits the entire ecosystem by providing a low friction and low complexity tool for onboarding and informed participation in decentralized governance. It’ll be like having an expert you can ask questions instead of having to look them up.
DAOs and other digital native organizations with decentralized governance have the same need to improve informed engagement, simplify complexity, and improve onboarding, training, and participation. The lessons learned from this implementation will benefit the entire ecosystem and its projects.
Comments (2)
I would rather look into integrating ChatTensor, the LLM created on Bittensor's decentralized network. Bittensor is built on substrate which would likely make integration easier, while at the same time popularizing a Substrate use-case.
Thanks for your feedback. Do you have any hands-on experience with ChatTensor you could share? We are open to other LLMs. For context: we want this solution to be immediately useful to users, and ChatGPT seems to be the most robust solution. Governance participation is a real challenge, and we believe this can be a part of the solution.
Why is Hashed team the right team to be implementing this? Where would these be hosted? Discussion happens on Polkassembly & Subsquare, do we expect teams to go to another website and see this? What questions would be answered per proposal? As seen in the video and overall usage of chatgpt is more interactive. One keeps asking questions to get into further depth of the proposal. I am not sure how this implementation will help. Teams should rather be given a template to fill in the details in a particular format and the proposal creator should go through to summarie the proposal. There is a team working on proposal audits and I believe them using chatgpt for summarizing proposals would make more sense than spending 100k+ on this implementation.
Thanks for your comment. Why are we the right team? I've been doing Analytics and Data Science for 14 years. The lead architect Sebastian Montero too and spent a year doing ML research at Tokyo University. The rest of the devs are also familiar with ML. Here is the rest of our blockchain experience: https://hashed.io/ Where should it be hosted? I see your point, and if you ask me, it should be widely available as a chat window in Polkassembly, Subsquare, and others. Summarization is valuable, but as you note, the back and forth is where it comes alive. That is what we would implement, not only summarization. Does that make sense?