AI如何才能通过“终极考验”？让它重走人类来时的路

Core Idea - The article discusses the evolution of artificial intelligence (AI) and proposes a new testing framework called the "Nigiro Challenge" to evaluate AI's ability to understand and create language systems, drawing parallels with the historical development of writing and civilization [1][15][18]. Group 1: AI and Language Understanding - The Nigiro Challenge aims to assess whether AI can invent and systematize a writing system to record its civilization, similar to how humans developed writing [15][17]. - The challenge reflects on the limitations of the Turing Test, suggesting that it may not adequately measure true intelligence or understanding in AI [14][17]. - The article emphasizes the importance of social interaction in the development of intelligence, proposing that AI should demonstrate its capabilities through the creation of a unique writing system [15][18]. Group 2: Historical Context of Writing - The origins of writing are traced back to ancient practices such as tokens for counting and seals for confirming ownership, which laid the groundwork for the development of cuneiform writing [6][10]. - The emergence of cuneiform writing around 3500-3000 BCE is linked to the increasing complexity of society, necessitating a system for recording transactions and information [11][12]. - The article highlights that the development of writing was a collective human achievement, reflecting the growth of social complexity and the need for communication [11][18]. Group 3: AI Development and Challenges - The article discusses the evolution of tokenization in AI language models, from word-level to subword-level approaches, and the significance of the Transformer architecture in processing language [12][14]. - It raises philosophical questions about whether AI truly understands language or merely manipulates symbols based on statistical relationships [14][15]. - The Nigiro Challenge serves as a framework to explore the essence of intelligence, prompting a reevaluation of what constitutes understanding in both humans and AI [18].