Algorithmic systems, particularly social media recommenders, have achieved remarkable success in predicting behavior. By optimizing for observable signals such as clicks, views, and engagement, these systems effectively capture user attention and guide interaction. Yet their widespread adoption has coincided with troubling outcomes, including rising mental health concerns, increasing polarization, and erosion of trust. This paper argues that these effects are consequences of a structural functional misalignment between what algorithms optimize - predictable behavior - and the human goals these predictions are intended to serve. We propose that this misalignment arises through three mechanisms: (1) a bias toward modeling fast, reactive behavioral signals over reflective judgment, (2) feedback loops that couple user behavior with algorithmic learning, and (3) emergent collective dynamics that amplify these effects at scale. Together, these mechanisms explain how accurate individual-level predictions can produce adverse societal outcomes. We present functional misalignment as a unifying framework and outline a research agenda for studying and mitigating its effects in human-AI interaction systems.

翻译：算法系统，特别是社交媒体推荐系统，在预测用户行为方面取得了显著成功。通过优化点击量、浏览量和参与度等可观测信号，这些系统有效捕获了用户注意力并引导交互行为。然而，这些系统的广泛应用却伴随着令人不安的后果，包括日益加剧的心理健康问题、持续扩大的观点极化以及信任的侵蚀。本文认为，这些后果源于算法优化目标（可预测的行为）与这些预测本应服务的人类目标之间的结构性功能失调。我们提出，这种失调通过三种机制产生：（1）偏向于对快速、反应性的行为信号进行建模，而非对反思性判断建模；（2）形成用户行为与算法学习相互耦合的反馈回路；（3）涌现出的群体动态在更大规模上放大这些效应。这三种机制共同解释了为何精准的个体层面预测可能导致有害的社会结果。我们将功能失调作为统一的理论框架，并提出了在人机交互系统中研究并缓解其影响的研究议程。