Scholarlink

<p>In this paper, I argue that contemporary AI systems can be meaningfully considered intelligent, albeit in a functional sense, challenging Floridi’s (2025) claim that AI is <i>Agency without Intelligence</i>. Using psychometric frameworks and empirical findings from mechanistic interpretability, I compare Large Language Models (LLMs) performance with human cognitive architectures, showing that current models already realise a non-trivial subset of broad cognitive abilities, despite structural limitations and different implementations. In an anti-anthropocentric perspective, I develop a substrate-neutral account of intelligence, according to which cognitive properties are defined by their operational roles rather than by biological occurrences. On this basis, LLMs exemplify a distinct form of <i>functional intelligence</i>: they lack intentionality, will, self-determination, or awareness, yet implement certain cognitive states at the operational level. The central question, therefore, is not whether AI can be considered a form of agency, but whether some forms of artificial agency can, under specific functional and empirical constraints, instantiate a limited but non-trivial kind of functional intelligence. I conclude that a unified framework combining Floridi’s <i>Artificial Agency</i> with <i>Functional Intelligence</i> offers a more adequate understanding of Generative AI, capable of grasping both the ethical implications of Floridi’s insights and the empirical commitment of my argument.&#xa0;</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

AI as Functional Intelligence: Cognitive Modeling in Generative AI

  • 期刊论文

  • 发表日期：

    2026年6月16日

查看全文
  • Simone Conversano

摘要

In this paper, I argue that contemporary AI systems can be meaningfully considered intelligent, albeit in a functional sense, challenging Floridi’s (2025) claim that AI is Agency without Intelligence. Using psychometric frameworks and empirical findings from mechanistic interpretability, I compare Large Language Models (LLMs) performance with human cognitive architectures, showing that current models already realise a non-trivial subset of broad cognitive abilities, despite structural limitations and different implementations. In an anti-anthropocentric perspective, I develop a substrate-neutral account of intelligence, according to which cognitive properties are defined by their operational roles rather than by biological occurrences. On this basis, LLMs exemplify a distinct form of functional intelligence: they lack intentionality, will, self-determination, or awareness, yet implement certain cognitive states at the operational level. The central question, therefore, is not whether AI can be considered a form of agency, but whether some forms of artificial agency can, under specific functional and empirical constraints, instantiate a limited but non-trivial kind of functional intelligence. I conclude that a unified framework combining Floridi’s Artificial Agency with Functional Intelligence offers a more adequate understanding of Generative AI, capable of grasping both the ethical implications of Floridi’s insights and the empirical commitment of my argument.