香港大学马毅：人工智能应从“黑箱”走向“白箱”

Core Insights - The current development of artificial intelligence (AI) lacks a scientific understanding of its essence, necessitating a shift from "black box" systems to "white box" models based on mathematical principles and closed-loop feedback [1][4] Group 1: Evolution of Intelligence - The evolution of intelligence can be categorized into four stages: genetic intelligence represented by DNA, individual developmental intelligence through brain and perception systems, collective intelligence facilitated by language, and finally, true artificial intelligence [3] - Current AI, exemplified by large models, is still in the primitive "genetic intelligence" stage, relying on vast parameters and pre-training data, which leads to high resource consumption and inefficiency [3] Group 2: Core of Intelligence - The essence of intelligence lies in the ability for "self-verification and self-correction," allowing for critical examination of existing knowledge to identify and rectify errors [4] - Current large models function merely as static repositories of knowledge without true understanding, resulting in logical inconsistencies and "hallucination" issues [4] Group 3: Future Directions - Future research should treat intelligence as a rigorous scientific and mathematical subject, focusing on developing systems with individual memory and autonomous closed-loop capabilities [4] - The goal is to advance machine intelligence towards genuine "autonomous intelligence" within an explainable theoretical framework [4]