SingularityNET (@SingularityNET): Dr. @bengoertzel, CEO of SingularityNET and @ASI_Alliance, delivered a keynote at the Decentralized Knowledge Graph Conference held by our partner @or…

Dr. @bengoertzel, CEO of SingularityNET and @ASI_Alliance, delivered a keynote at the Decentralized Knowledge Graph Conference held by our partner @origin_trail to discuss the combined values of LLMs and Knowledge Graphs and how they fit into the past and future of AI. Dr. Goertzel began by tracing the evolution of the AI field from its origins in the late 1950s, noting the foundational influence of Norbert Wiener's 1948 book "Cybernetics" which established principles for neural networks and cognitive algorithms in both biological and engineered systems. He highlighted that since AI's inception, two main approaches have coexisted: subsymbolic AI (neural networks) composed of tiny components without clear comprehensible semantics where meaning emerges from learned combinations, and symbolic AI which emerged in the mid-1960s, built with elements having clear semantic meaning, such as explicit representations using nodes and connections. The distinction, he explained, reflects both biological reality (neural networks resembling brain function) and human cognitive advancement through abstract symbol systems like science, language, and mathematics. The current AI landscape, as described by Dr. Goertzel, shows remarkable progress in subsymbolic AI, particularly with Transformer Neural Networks and Conventional Neural Networks achieving unprecedented success through increased computational power and data availability. However, he pointed out significant limitations in current neural systems, including difficulties in distinguishing imagination from reality, limited creativity beyond surface-level pattern combinations, and inability to make fundamental leaps beyond their training data. He illustrated this limitation by noting that a transformer neural net trained on music up to 1900 could never invent new genres like neoclassical metal or progressive jazz, instead only combining surface-level patterns from its training data. Looking toward the future, Dr. Goertzel suggested we are potentially just two to seven years away from achieving AGI, a term he popularized in the early 2000s. The path forward, he argued, lies in combining symbolic and subsymbolic approaches. This includes augmenting deep neural networks with logical reasoning capabilities, implementing Knowledge Graphs alongside deep neural networks, and developing more fundamental creative generation algorithms. He emphasized that Knowledge Graphs represent the most scalable and flexible approach to representing symbolic knowledge, making them crucial for bridging the gap between symbolic and subsymbolic systems. The next breakthrough, he predicted, will come from integrating these approaches rather than simply scaling up existing transformer models. Dr. Goertzel concluded by addressing the broader implications of AI development, highlighting the importance of developing AGI in a decentralized, beneficial, and ethical way. He warned against allowing a small elite group to control AGI development, advocating instead for a decentralized infrastructure to benefit humanity and other sentient beings. This ethical consideration, he noted, is as crucial as the technical challenges in advancing toward AGI and, shortly after, ASI.