FOSSMeet'24

The growth (and progress!) of AI research and application have been exponential. The recent successes of LLMs and diffusion models have been expanding the imaginations of the public, converting fiction to reality at rapid pace. But the accessibility (and equitable benefit) to the general public of both the present reapings and the bright future potential, is moot.

I propose a talk on what it’d mean to truly democratize AI, the importance of proliferation and avoiding concentration of power. And what it may take to achieve. I believe it’s of around equal importance to the popular “safety and alignment” branch of study.

Here’s a rough outline of things I’d be emphasizing on:

• the meaning of “true democratization”, and why it is needed
• factors and approaches I plan to expand upon:
  • open source AI and research
  • alternative paradigms
    • low-resource AI
    • federated/distributed learning
    • promising alt architectures
  • herd inference
  • the factor of hardware-availability and freedom for true equitable access
• some honorable mentions of other factors:
  • policy and regulation
  • community efforts, movements, organisations
  • spreading of awareness and activism
• the risks and “safety issues” cited around wide-scale proliferation, and making an attempt to separate the myths from the facts.

I also propose real-time discussions around the same topic. It should include proposal of further ideas, constructive scrutiny/reinforcement of ideas, and other talking points.
Comments and feedback are welcome!