Scholarlink

<p>Current approaches to AI safety largely rely on reactive filtering that blocks undesirable outputs after generation. We argue that such post-hoc moderation introduces a structural ethical flaw: it creates a gap between internal generation processes and external outputs, producing what we term ethical hallucination—the appearance of alignment through surface-level filtering while the underlying architecture remains ethically unconstrained. This constitutes representational deception that violates stakeholder epistemic rights and reflects inadequate designer responsibility for process integrity. Building on Kantian and Aristotelian ethics alongside contemporary debates in philosophy of AI (Bostrom, 2014; Floridi et al., 2018; Gabriel, 2020; Coeckelbergh, 2020; Tasioulas, 2022; Vallor, 2016), and engaging recent alignment work (Christiano et al., 2017; Bai et al., 2022; OpenAI, 2023), we propose Proactive Moral Architecture (PMA)—a normative framework treating ethics as constitutive of generative processes rather than as external constraint. This paper reframes AI ethics from product-focused to process-focused evaluation, arguing that genuine alignment requires architectural integrity, not merely output compliance.</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

From reactive filtering to proactive moral architecture: rethinking ethical alignment in large language models

  • 期刊论文

  • 发表日期：

    2026年7月6日

查看全文
  • H. Mustafa Akyol

摘要

Current approaches to AI safety largely rely on reactive filtering that blocks undesirable outputs after generation. We argue that such post-hoc moderation introduces a structural ethical flaw: it creates a gap between internal generation processes and external outputs, producing what we term ethical hallucination—the appearance of alignment through surface-level filtering while the underlying architecture remains ethically unconstrained. This constitutes representational deception that violates stakeholder epistemic rights and reflects inadequate designer responsibility for process integrity. Building on Kantian and Aristotelian ethics alongside contemporary debates in philosophy of AI (Bostrom, 2014; Floridi et al., 2018; Gabriel, 2020; Coeckelbergh, 2020; Tasioulas, 2022; Vallor, 2016), and engaging recent alignment work (Christiano et al., 2017; Bai et al., 2022; OpenAI, 2023), we propose Proactive Moral Architecture (PMA)—a normative framework treating ethics as constitutive of generative processes rather than as external constraint. This paper reframes AI ethics from product-focused to process-focused evaluation, arguing that genuine alignment requires architectural integrity, not merely output compliance.