TLDR: The current centralized model for AI conferences is facing a crisis due to unsustainable growth in publications, high carbon emissions, significant mental health strain on researchers, and venue capacity limitations. A new Community-Federated Conference (CFC) model is proposed, which separates peer review, presentation, and networking into globally coordinated digital platforms and locally organized regional hubs. This aims to create a more sustainable, inclusive, and resilient system for AI research dissemination and community building.
Artificial Intelligence (AI) conferences are vital for advancing research, sharing knowledge, and building academic communities. However, a recent paper highlights that their rapid expansion has made the current centralized model increasingly unsustainable, posing a significant threat to scientific dissemination, equity, and community well-being.
The paper, titled “Position: The Current AI Conference Model is Unsustainable! Diagnosing the Crisis of Centralized AI Conference,” authored by Nuo Chen, Moming Duan, Andre Huikai Lin, Qian Wang, Jiaying Wu, and Bingsheng He from the National University of Singapore, provides a data-driven diagnosis of this structural crisis. You can read the full paper here: Research Paper.
Four Key Areas of Strain
The researchers identify four major areas where the current conference model is struggling:
1. Scientific Strain: Over the past decade, the average per-author publication rate has more than doubled, now exceeding 4.5 papers annually. This intense pressure often leads to a focus on quantity over quality, fostering hyper-competition and potentially stifling genuine innovation. The annual conference cycle also struggles to keep pace with the rapidly shrinking lifecycle of AI research, meaning findings can be outdated by the time they are presented.
2. Environmental Strain: The carbon footprint generated by thousands of researchers traveling globally is substantial. For instance, the travel emissions for NeurIPS 2024 alone were estimated to exceed the daily emissions of its host city, Vancouver. This environmental impact contradicts global sustainability goals and creates economic and visa barriers for many researchers.
3. Psychological Strain: The high-stakes environment takes a significant human toll. Analysis of online community discussions revealed that over 71% of conference-related posts expressed negative sentiment, with 35% referencing mental health concerns like anxiety, burnout, and stress. This toxic atmosphere erodes trust, inhibits risk-taking, and diminishes the sense of belonging within the community.
4. Logistical Strain: Attendance at top conferences, such as NeurIPS 2024, is beginning to outpace venue capacity. This leads to measures like lottery systems for registration, which limit participation for students and early-career researchers who could benefit most from in-person engagement. Such constraints reduce opportunities for meaningful connections and compromise equitable participation.
Why Incremental Fixes Fall Short
The paper argues that current incremental adjustments, like limiting submissions per author or adopting multi-site conferences, are insufficient. Submission caps, for example, merely shift the pressure without addressing the underlying “publish-or-perish” culture. Multi-site events, while easing some travel, still concentrate the review burden and mental health anxiety into the same frantic annual cycles, potentially creating a two-tiered system of prestige.
Also Read:
- AI Agents Master Collaboration: A Hybrid Approach to Ad Hoc Teamwork
- Galaxy: A New Framework for Smarter, More Private AI Assistants
A Path Forward: The Community-Federated Conference (CFC) Model
To address these challenges, the researchers propose a fundamental shift to the Community-Federated Conference (CFC) model, guided by the principle of “Global Standards, Local Realization.” This model decouples the traditional functions of conferences into three interconnected layers:
Layer 1: Unified Global Peer Review and Publication: This involves a centralized, high-quality digital platform managed by academic societies. Submissions and reviews would occur on a rolling basis throughout the year, independent of physical meetings. This temporal decoupling aims to ease reviewer burden, allow for more thoughtful feedback, and ensure academic credit and visibility for accepted papers.
Layer 2: Federated Regional Hubs for Dissemination and Networking: Once papers are accepted, authors would present their work at regional hubs, organized by local universities, labs, or student groups. These smaller gatherings (typically 500 to 1,500 participants) would significantly reduce carbon emissions by encouraging regional travel, lower financial barriers, and foster more meaningful, less anonymous interactions, promoting greater diversity, equity, and inclusion.
Layer 3: Digital Synchronization and Collaboration: A unifying digital layer would connect all regional hubs. This includes a Global Plenary track live-streaming keynotes and award talks from a rotating anchor hub, permanent digital poster halls for asynchronous discussions, and thematic virtual channels to connect researchers globally. This ensures local participation remains deeply connected to global discourse without the need for physical convergence.
The CFC model is designed to be built from the ground up by the community, integrating digital participation as a primary component, unlike traditional hybrid models where remote involvement is often secondary. By decentralizing and distributing core functions, the CFC model offers a resilient and forward-looking architecture that aims to address the shortcomings of the current system and advance the core values of inclusivity, sustainability, and intellectual exchange in AI research.
