Key for a Better Artificial System: From Gov to AI (Panel)

https://www.youtube.com/watch?v=14B6Gg0ik0Y&list=PL3C6eF-zu5AYohNL1ZgOBqlwwJ29x-lTO&index=29

1. panel overview

• speaker
  • Ryota Kanai (Araya, AI Startup / AI Alignment Network)
  • Jimmy Chan (fabd)
  • Cui Jia Wei (related to gov and Taiwan, V-Taiwan)
• theme
  • Toward Better Artificial Systems: from Governance to AI."
  • Democratic use and governance of AI, technological and social risks, and initiatives in each country and organization were discussed.

2. the V-Taiwan initiative (Mr. Cui Jia Wei)

• V-Taiwan Features
  • Emphasis on people-centered and citizen participation
  • Open and transparent proceedings and records (published on the web)
  • Progressive use of technology tools (both online and offline)
• Four Stages
  • Proposal stage: Citizens and government submit issues, stakeholders and issues
  • Online Opinion Gathering: Gathering a wide range of opinions from stakeholders
  • Face-to-face deliberations: Hybrid format (live streaming + face-to-face discussions)
  • Consensus building/implementation: recommendations to the government and legislation
• track record
  • Won OpenAI Grants and collaborated with Chatham House
  • Civic Assembly on AI with diverse participants (engineers, lawyers, journalists, indigenous peoples, gender groups, politicians, etc.)
  • Publish reports of consensus reached on citizen participation (clustering opinions collected with AI tools)
• Main Implications
  • Balancing expertise and plurality: the importance of argument design
  • People-centered or technology-centered: Need to reconsider, including means to close the digital gap
  • Wary of concentration of big business and government power: How citizens, non-profit organizations, DAOs, etc. can be empowered is key.

AI Alignment Network (Mr. Ryota Kanai)

• What is AI Alignment?
  • Research and activities to "enable AI to act in accordance with human values and objectives
  • Misaligned goal setting and evaluation metrics for large-scale AI (sophisticated models) may adversely affect society in unforeseen ways
  • Developers, researchers, and society must work together to build consensus
• issue
  • Risk of monopolization amid accelerated development competition (led by giant corporations and specific states)
  • Multinational and cross-cultural "differences in values" and "differences in ethical standards."
• Direction of Efforts
  • Promote research and discussion not only in Japan, but also in cooperation with Taiwan, Europe, and the United States
  • Addresses both technical issues (e.g., how to understand the inner workings of black-boxed models) and social design issues (democratic control)

4. perspective from fabd (Mr. Jimmy Chan)

• Sharing the concept of "[singularity
  • AI evolution is exponential, human thinking is linear → Irreversible balance of power shift can occur
• Importance of Objective Function
  • The setting of the "objective/goal" itself, which AI executes, determines its impact on society.
  • A consensus building process is needed to determine who will represent how objectives will be set
• Relationship with DAO and DCC (Decentralized Cooperative/Collective)
  • To avoid centralization of technology and power, it is important to have a system that allows more individuals and organizations to operate technology in a shared and decentralized manner
  • In addition to large-scale models, small-scale and specialized models can be treated as "public goods," and an approach to interconnecting them can be considered.

5. axis of conflict between large corporations/government and open source

• Meta's open source policy vs. closed OpenAI
  • Meta's policy is to open up large scale models and encourage community-driven improvements
  • OpenAI offers services for users while setting out to be closed to ensure security.
• European regulations (DMA, etc.)
  • Legal requirement for interoperability of platforms and AI models to prevent monopolies
  • Risk diversification and democratic control through "audit and disclosure requests" by the government
• What individuals can do
  • Lobbying (policy advocacy)
  • Organize and utilize their own specific data (e.g., creating glossaries of terminology in their field of expertise)
  • Trial of open source and P2P technologies

6. general summary

• Democratic and decentralized AI governance is key
  • The case study (V-Taiwan) is a useful model for involving diverse stakeholders and deepening discussions through both online and offline methods.
  • Efforts to balance technical and ethical aspects, such as the work of the AI Alignment Network, are important.
• Countering the risk of large corporate and government monopolies
  • Open sourcing, interoperability, and policy audits are leading approaches
  • Empowerment through new organizational forms such as DAOs and DCCs
• Objective Function Setting (Objective Function) / Shared Values
  • As AI becomes more sophisticated, human "consensus-building processes" and "ways of reflecting values" will become increasingly important.
  • Need to expand social and political consensus building beyond technology development

This page is auto-translated from [/nishio/Key for a Better Artificial System: From Gov to AI (Panel)](https://scrapbox.io/nishio/Key for a Better Artificial System: From Gov to AI (Panel)) using DeepL. If you looks something interesting but the auto-translated English is not good enough to understand it, feel free to let me know at @nishio_en. I'm very happy to spread my thought to non-Japanese readers.