Democratising AI: Multiple Meanings, Goals and Methods

Last tended Apr 10, 2023 Edit Source

• Democracy
• Ai

Democratising AI: Multiple Meanings, Goals, and Methods

• Defines four modalities of “democratization”:
  • Use:
    • reducing the costs of acquiring and running AI tools
    • providing intuitive interfaces to facilitate human-AI interaction w/o extensive training or technical skills
  • Development:
    • model, code & data sharing
    • compute access
    • project support & coordination
    • education & upskilling
  • Profits:
    • avoid widening socioeconomic divide between leading & lagging nations
    • ease financial burden of job loss to automation
  • Governance:
    • decisions on usage, development and profit reflect interests and preferences of impacted peoples
    • decentralizing control away from big tech
    • normative cross-cultural questions
    • ensuring benefits and burdens distributed fairly
• May conflict (eg governance may restrict use or governance)
• democratization =/= dissemination
• positive value of AI democratization rooted in how well the effort reflects interests and values of impacted stakeholders