From Symbolic Models to Collective Intelligence: A Physics and Neuroscience-Inspired AI Framework for a Sustainable Future -

This paper describes the evolution of a next-generation AI architecture that not only reduces risk and incorporates human cognition but also holds the potential to revolutionize the field. The new architecture incorporates fundamental principles from the natural sciences, specifically physics, computational biology, and neuroscience. The first iteration of this architecture combines the science of collective intelligence with an adaptive learning approach that builds knowledge models from the collaborative work of1human and AI agents working together. Early work on the application of developing causal models of investment decisions was published in a paper for NIPS 2018.1. That work was extended to a general AI architecture for decision intelligence discussed here.The work is set in the larger context of AI over the past five decades. The first wave of AI focused on creating symbolic models of human reasoning and knowledge representation. These models, based on logic and network models of semantics, were explanatory and contextualized. However, they were limited in their capacity for self-extension—they did not learn from data.The current generation of AI, which applies statistical methods, has resulted in significant advances in deep learning and pre-trained transformer models. These models show a significant ability to scale and demonstrate impressive capabilities. However, because they are learning from complex aggregated statistical correlations, they lack explanatory capabilities and thus open the door to distrust and concerns about future impact on humanity. In addition, they need a clearer method of managing context. Training massive models with a wide context window is energetically challenging. They threaten struggling climate change initiatives, highlighting the urgent need for a new approach.Over the past decade, new distributed AI models have emerged that promise a path forward that addresses explanatory power, context management, and energy efficiency issues. These models are grounded in physics and biology. Furthermore, the essence of the new AI architecture is to amplify the co-creative of collective intelligence. It is collective intelligence, institutionalized in the scientific method of knowledge discovery, that has led to all humanity's progress to date.Key aspects of how the third wave of AI will be presented include a survey of results in applying the neuroscience-based work associated with the Free Energy Principle and Active Inference. In addition, results will be presented on integrating the science of collective intelligence with new AI methods and the implications of these methods for multi-agent models that incorporate human collective intelligence with LLM-based AI agents.