THE KALEIDOSCOPE PROJECTThe Kaleidoscope Project

Community-Led
Public Interest AI

  • EVIDENCE-BASED DECISION-MAKING

  • CROSS-FUNCTIONAL COALITIONS

  • COMMUNITY AGENCY & BENEFIT

"Public Interest AI" is an all-encompassing term for work that advances, examines, or critiques the intersection of AI and society. Unfortunately, it has been dominated by a top-down push from Big Tech that spreads frontier models for "societal benefit" despite limited evidence of meaningful impact, public desire, or confidence that already-present harms won't escalate.

Only 9% of Americans believe AI will do more good than harm, as many have concerns around the mass, non-consensual collection of training data, environmental harms, and role in defense, surveillance, and other oppressive systems. Despite this, AI is increasingly embedding itself within society, as "trust based institutions" are facilitating Big Tech's market penetration that circumvents accountability as downstream communities absorb harm during this period of light regulation.

The Kaleidoscope Project is an attempt to counterbalance this top-down push by researching the use of community-owned contextual data to strengthen evidence-based decision making, build strategic cross-functional coalitions, and develop tools for democratic, transparent governance.

Our pilot project, Public Interest AI Matchmaking, is a survey open to anyone whose work fits into our 6 groups. We hope it can provide a good snapshot of whose working in Public Interest AI, and may even connect you to the funder, community, or other right partner to take your work further!

The Kaleidoscope Model

While there are many nuanced approaches to "AI for Good", they can generally be categorized into one of 6 groups depending on their structural power. Read more here [link] for more gaps, along with strengths and weaknesses in their general execution.

The current power distribution is primarily concentrated in AI labs, and secondarily in a subset of funders and governments. Builders and researchers sit below them, while end users have the least.

At Kaleidoscope, we advocate for a more equitable power distribution centered on community-owned, contextual-level deployment data that guides how AI is deployed and governed.

For an in-depth overview of how the Kaleidoscope works in practice, including examples, click here. The high-level operational rules include:

  1. Any stakeholder other than Big Tech can facilitate the collection of community-sourced data.

  2. Any data collected from the community should be owned by the community.

  3. Any stakeholder should be able to use the data, but only with the community's approval and, when needed, appropriate compensation for using their data.

Think that this work could be improved? We would love your feedback!