Development of AI from Search to Companion: A Literature Review

Abstract

The main problem of AI evolution (from Search Engine, Recommendation Systems, Generative AI, to AI Companion) is the existence of a consistent trade-off pattern in each phase of development: Potential Filter Bubble (limiting perspective), and triggering Polarization or Addiction (especially in the context of recommendation systems). This Study is important. AI has shifted from a functional tool (search) to a digital companion that prioritizes anthropomorphic (human-like) interactions. Artificial Intelligence (AI) has transformed from a PageRank-based search engine to a digital companion system that prioritizes anthropomorphic interactions. This study examines the evolution of AI through four phases: search engines, recommendation systems, generative AI, and AI companions, by analyzing the benefits, risks, and socio-technical implications at each stage. Using a systematic literature review of 45 publications (2014-2025), this study identifies a consistent trade-off pattern: increased personalization and system autonomy accompanied by the risks of filter bubbles, hallucinations, and artificial intimacy. The theoretical framework of the Media Equation and three-factor anthropomorphism explains why users tend to treat AI like humans, raising challenges of overtrust and cognitive manipulation. In this paper, findings highlight the need for ethical governance based on humans-in-the-loop, algorithmic transparency, and AI literacy for responsible adoption. Increasing personalization and system autonomy are always accompanied by increased risks of filter bubbles, hallucinations, and artificial intimacy.