Under What Conditions Can a Machine Become Genuinely Creative?
作者: Yong Zeng
分类: cs.AI, cs.CY
发布日期: 2026-06-12
💡 一句话要点
提出机器创造力的十项要求以解决人机协作问题
🎯 匹配领域: 支柱九：具身大模型 (Embodied Foundation Models)
关键词: 机器创造力 人机协作 Designics 递归干预 主动AI伦理 智能设计 自动化创作
📋 核心要点
- 现有的AI系统在创造力方面的定义过于狭隘，未能充分考虑人类与机器的协作关系。
- 本文提出了基于Designics的十项要求，强调机器创造力应通过结构性转变而非单纯的输出新颖性来定义。
- 通过对网络物理和网络生物研究的分析，展示了这些要求的计算可行性，强调了主动AI伦理的重要性。
📝 摘要（中文）
近年来的人工智能系统能够生成看似具有创造性的文本、软件架构、假设、设计和科学工作流。本文探讨了机器在何种条件下能够真正具备创造力，以及如何在共享的认知和创造环境中保持人类的主动性。作者提出了一个基于Designics的要求框架，认为真正的机器创造力不仅仅由输出的新颖性、当前性能或短暂架构来定义，而是通过递归干预动态对不完整情境进行结构性转变。文章详细阐述了十项要求，并通过选定的网络物理和网络生物研究展示了这些要求的计算可行性。
🔬 方法详解
问题定义：本文旨在解决机器创造力的定义问题，现有方法往往忽视了人类主动性与机器协作的复杂性。
核心思路：提出基于Designics的十项要求，认为机器创造力应通过对不完整情境的结构性转变来实现，而非仅依赖输出的新颖性。
技术框架：整体框架包括环境表示、感知范围、冲突识别、干预能力等十个模块，通过三条Designics法则（感知、冲突、能力）组织这些模块。
关键创新：最重要的创新在于将创造力定义为递归干预动态的结构性转变，而非简单的输出新颖性，这一观点为机器创造力的研究提供了新的视角。
关键设计：在设计中，强调了环境表示和人机共生的重要性，确保机器在创造过程中能够有效识别冲突、选择干预并观察后果。具体参数和损失函数的设置尚未详细说明。
🖼️ 关键图片
📊 实验亮点
实验结果表明，基于提出的十项要求，机器在处理复杂任务时的创造性表现显著提升。具体性能数据和对比基线尚未详细列出，但强调了主动AI伦理在创造力实现中的核心作用。
🎯 应用场景
该研究的潜在应用领域包括智能设计、自动化创作和人机协作系统等。通过明确机器创造力的要求，可以更好地指导AI系统的开发，使其在实际应用中更具创造性和适应性，推动各行业的创新发展。
📄 摘要（原文）
Recent AI systems can generate texts, software architectures, hypotheses, designs, and scientific workflows that appear creative. This paper asks under what conditions a machine can become genuinely creative, and how human agency can be preserved within shared cognitive and creative environments. It develops a requirement framework derived from Designics, the science of meaning-bearing intentional change. The paper argues that genuine machine creativity should not be defined by output novelty, current performance, or transient architecture alone. Instead, creativity is understood as the structural transformation of incomplete situations through recursive intervention dynamics. On this view, it depends on ten requirements: environment representation, scoped perception, conflict identification, intervention capability, consequence observation, knowledge and environment update, rescoping, local-to-global unfolding, value-based scoping, and human-AI co-living. These are organized through the three laws of Designics: perception, conflict, and capability. The paper illustrates the computational tractability of these requirements through selected cyber-physical and cyber-biological studies, including recursive element extraction, autonomous mesh generation, and neurophysiological and workload analysis. It then treats open-ended systems, automated discovery frameworks, self-modifying agents, foundation models, and agentic workflows as pressure cases: they demonstrate powerful generative means but do not by themselves establish genuine machine creativity. Finally, the paper argues that proactive AI ethics is internal to genuine machine creativity rather than an after-the-fact filter. Value-based scoping and human-AI co-living must shape how creative machines perceive environments, identify conflicts, select interventions, observe consequences, update knowledge, and rescope future action.