Proceedings from the Columbia Convening on Openness in Artificial Intelligence and AI Safety

[Submitted on 27 Jun 2025]
View the PDF file for the paper entitled “A different approach to the safety of Amnesty International
PDF view
a summary:The rapid height of the open foundation models and open source weight intensifies the commitment and reshape the opportunity to make artificial intelligence systems safe. This paper refers to results from Colombia that is held to open up to artificial intelligence and safety (San Francisco, November 19, 2024) and its preparatory program for a period of six weeks that included more than forty -five researchers, engineers and leaders of the academic, industry, civil society and government. Using a participatory process directed towards solutions, working groups (1) produced the search agenda at the intersection of safety and an artificial intelligence, an artificial intelligence; (2) Maps for current and required technical interventions and open sources tools to publish the basic models open safely and responsibly through the workflow developing artificial intelligence; And (3) maps of the content filtering system with a proposed road map for future research and development. We find that openness – concept as transparent weights, inter -operational tools, and general rule – can enhance safety by enabling independent scrutiny, decentralized mitigation and culturally combining. However, there are still large gaps: multimedia and multimedia standards, limited defenses against immediate injection and synthetic attacks in agent systems, and insufficient participatory mechanisms for societies most affected by artificial intelligence. The paper concludes with a road map of five priority research trends, focusing on participatory inputs, future content filters, safety infrastructure at the ecological system level, strict guarantee of the agent, and classification of expanded damage. These recommendations were informed of the French AI’s action in February 2025 and laying the foundation for the discipline of open, metal and accounting intelligence.
The application date
From: Camille Francois [view email]
[v1]
Friday, 27 June 2025 12:45:44 UTC (2086 KB)
Don’t miss more hot News like this! Click here to discover the latest in AI news!
2025-06-30 04:00:00