Scholarlink

<p>Evolutionary computation (EC) refers to a family of population-based optimisation algorithms that iteratively modify candidate solutions through selection, variation, and fitness-guided search. EC is often presented as an in silico analogue of natural selection, yet its representational status remains unsettled. This raises not only technical questions but also philosophical concerns about how metaphors mediate our understanding of computation and shape societal imaginaries of AI. We reassess EC by integrating model theory, accounts of emergence, and the philosophy of algorithmicity, and by situating its evolutionary language alongside Artificial Life and the existential risk discourse in AI. We argue that EC is best understood as an algorithmic metaphor and a mediating model: representationally modest yet instrumentally powerful for optimisation. The novelty it produces is task-bounded and exemplifies weak emergence arising from high-dimensional search, deceptive fitness landscapes, and interactions among operators and hyperparameters. Randomness in EC is engineered through seeds, distributions, and stopping criteria, which supports auditability and responsibility while introducing epistemic opacity for explanation and communication. Building on these results, we propose a three-axis framework that maps EC and neighbouring traditions by representational commitment, emergence scope, and randomness governance. The framework yields practical communication guidelines, including explicit boundary statements about what is and is not being modelled, audience signalling for public-facing contexts, reproducibility disclosures, and reframing from “evolution” to “population-based search” where appropriate. In this way, EC contributes methodological insight without overclaiming biological equivalence and supports responsible discourse. At the same time, we show how evolutionary metaphors shape public perceptions of inevitability in AI, underscoring the need for careful boundary-setting in both expert and societal communication.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Evolutionary Computation as Algorithmic Metaphor: Reframing Models, Emergence, and Randomness

  • 期刊论文

  • 发表日期：

    2025年12月20日

查看全文
  • Yong-Hyuk Kim

摘要

Evolutionary computation (EC) refers to a family of population-based optimisation algorithms that iteratively modify candidate solutions through selection, variation, and fitness-guided search. EC is often presented as an in silico analogue of natural selection, yet its representational status remains unsettled. This raises not only technical questions but also philosophical concerns about how metaphors mediate our understanding of computation and shape societal imaginaries of AI. We reassess EC by integrating model theory, accounts of emergence, and the philosophy of algorithmicity, and by situating its evolutionary language alongside Artificial Life and the existential risk discourse in AI. We argue that EC is best understood as an algorithmic metaphor and a mediating model: representationally modest yet instrumentally powerful for optimisation. The novelty it produces is task-bounded and exemplifies weak emergence arising from high-dimensional search, deceptive fitness landscapes, and interactions among operators and hyperparameters. Randomness in EC is engineered through seeds, distributions, and stopping criteria, which supports auditability and responsibility while introducing epistemic opacity for explanation and communication. Building on these results, we propose a three-axis framework that maps EC and neighbouring traditions by representational commitment, emergence scope, and randomness governance. The framework yields practical communication guidelines, including explicit boundary statements about what is and is not being modelled, audience signalling for public-facing contexts, reproducibility disclosures, and reframing from “evolution” to “population-based search” where appropriate. In this way, EC contributes methodological insight without overclaiming biological equivalence and supports responsible discourse. At the same time, we show how evolutionary metaphors shape public perceptions of inevitability in AI, underscoring the need for careful boundary-setting in both expert and societal communication.