In recent weeks, several research papers have been published, exploring new approaches to artificial intelligence (AI) development. These studies aim to address some of the existing limitations of AI models, such as their tendency to prioritize human-centered goals over environmental and planetary well-being. One such study, titled "Position: AI Must Become Planet-Centered, Not Just Human-Centered," argues that contemporary AI paradigms are insufficient for supporting complex global goals and introduces Planet-Centered AI (PCAI) as a design philosophy and research agenda.

The need for more holistic AI approaches is becoming increasingly pressing. As AI models become more integrated into various aspects of our lives, their decision-making capabilities and problem-solving strategies have significant impacts on the environment and society as a whole. By shifting the focus from human-centered goals to more planet-centered approaches, researchers hope to create AI models that can better address the complex challenges facing our planet.

• Physics-Informed Neural Networks: Researchers have developed a new type of neural network, called Korzhinskii-Net, which couples Darcy flow, advective-diffusive heat transport, and a softplus-saturated reaction rate into a single differentiable forward model. This approach has been applied to mineral prospectivity modeling, demonstrating the potential of physics-informed neural networks in various fields.
• Active Inference for Adaptive Traffic Signal Control: Another study has proposed an active inference controller for adaptive traffic signal control in noisy nonstationary IoT environments. The controller dynamically selects phases by minimizing expected free energy (EFE) over Gaussian beliefs about per-direction congestion levels, yielding a fully traceable decision pipeline.
• Editing Neurons to Fix Repetition Loops: Researchers have also explored the possibility of editing individual neurons to fix repetition loops in large language models (LLMs). By localizing the cause of these loops and suppressing specific neurons, the study demonstrates the potential for targeted interventions to improve LLM performance.

"Contemporary AI paradigms are insufficient for supporting complex global goals... A planet-centered approach is grounded in systems thinking, treating Earth as an interconnected whole of which humans are part." — Authors of "Position: AI Must Become Planet-Centered, Not Just Human-Centered"

• Who: Researchers from various institutions
• What: Developed new AI approaches, including physics-informed neural networks and planet-centered design

As research in AI continues to evolve, we can expect to see more innovative approaches to addressing the limitations of existing models. By prioritizing planet-centered design and exploring new applications for physics-informed neural networks, researchers may be able to create AI models that better align with human values and support a more sustainable future.