Thinking Beyond Tokens: From Brain-Inspired Intelligence to Cognitive Foundations for Artificial General Intelligence and its Societal Impact
Thinking Beyond Tokens: From Brain-Inspired Intelligence to Cognitive Foundations for Artificial General Intelligence and its Societal Impact [31.6] 本稿では,人工知能,認知神経科学,心理学,生成モデル,エージェントベースシステムの学際的合成について述べる。 我々は汎用知能のアーキテクチャと認知の基礎を分析し、モジュラー推論、永続記憶、マルチエージェント協調の役割を強調した。 我々は、人工知能への道の鍵となる科学的、技術的、倫理的課題を特定します。 論文参考訳（メタデータ） (Tue, 01 Jul 2025 16:52:25 GMT)
AGIを目指すうえでの整理「Several challenges remains, such as the need for grounded world models, dynamic memory, causal reasoning, robust handling of aleatory and epistemic uncertainty, developing perception of emotional and social contexts and collective agent architectures. Significant advancements have been made, such as Large Concept Models, Large Reasoning Models and Mixture of Experts, which improve LLM performance beyond next-token prediction by incorporating biologically inspired behaviors into output generation.」と指摘。