## TL;DR
AI is redesigning how we interact with technology -- from interfaces that generate themselves from natural language descriptions to AI agents that simulate user behavior to test usability. Human-computer interaction, long a field of manual design and testing, is being transformed by generative AI and LLM-based user research.

## Core Explanation
AI+HCI convergence areas: (1) Generative UI -- LLMs translate design requirements ("a dashboard showing sales by region") into functioning interface code (React components, HTML/CSS). Tools: v0.dev (Vercel), Figma AI, Uizard. Text-to-UI: describe the interface, AI generates it; (2) AI usability testing -- LLM-based agents simulate user personas navigating prototypes, identifying usability issues (confusing navigation, unclear labels) and generating UX improvement suggestions. Cost: $0.50-5 per test run vs $100-500+ for human participant testing; (3) Adaptive interfaces -- AI personalizes UI based on user behavior (frequently used features surfaced, rarely used hidden), cognitive load (simplifying interface for tired/distracted users), and accessibility needs (auto-adjusting contrast, font size, navigation mode); (4) Multimodal interaction -- combining voice, gesture, gaze, and touch in unified interfaces.

## Detailed Analysis
CHI 2025 generative UI research: LLMs fine-tuned on UI component libraries (Material Design, Ant Design) generate layout code. Key challenge: generated UIs often look good but have functional gaps (missing error states, non-accessible). Solutions: constraint-based generation (design system rules as hard constraints) and iterative refinement (AI generates, human critiques, AI revises). AI usability testing: LLMs prompted with persona descriptions ("65-year-old user with mild vision impairment, first time using this app") simulate interaction. The agent describes what it sees, what it tries to click, and reports confusion. NN/g validation: AI testing catches 60-75% of critical usability issues at 5-10% cost. Benefits: speed (test overnight vs weeks), cost (democratized usability for startups), and iteration frequency (test every design iteration). Limitations: AI agents miss subtle emotional responses and cannot judge aesthetic preference. The ideal workflow: AI testing for rapid iteration (frequent, cheap) + human testing for final validation (infrequent, comprehensive).