TLDR: A new research paper introduces Human-Layer AI (HL-AI), a user-owned, explainable AI framework designed to give individuals control over their social media experience. Implemented as a browser extension, HL-AI operates between platform algorithms and the user interface, offering features like a Context-Aware Post Rewriter, Post Integrity Meter, Granular Feed Curator, Micro-Withdrawal Agent, and Recovery Mode. This system aims to mitigate issues like misinformation, psychological distress, and loss of control by empowering users with real-time, transparent tools for safety, agency, and well-being, without requiring platform cooperation.
Social media platforms have become an integral part of our daily lives, connecting billions of people worldwide. However, their underlying design, often driven by engagement-first algorithms, can inadvertently lead to significant issues such as increased stress, the spread of misinformation, and a feeling of lost control among users. This fundamental problem stems from what researchers call ‘simplex communication,’ where platforms dictate information flow without meaningful user input.
Recent events highlight these concerns vividly. For instance, the July 2025 Milestone School helicopter crash in Bangladesh demonstrated how algorithmic feeds could repeatedly surface distressing content, intensifying psychological trauma for survivors and community members. Similarly, during Bangladesh’s July Revolution of 2024, social media became a double-edged sword, facilitating democratic mobilization while simultaneously acting as a vector for large-scale misinformation. Even Nepal’s September 2025 Gen Z protests, while showcasing the power of collective action, underscored the limitations of current platforms when the government banned major services, forcing protesters to migrate to more user-controlled environments like Discord.
These incidents, alongside cross-border misinformation campaigns, reveal a critical gap: mainstream platforms lack architectural support for user-owned, explainable AI intermediaries. This absence prevents real-time transparency, personalized control, and support for psychological well-being. Studies show that a significant percentage of daily users experience post-session regret, trust in platform content is declining, and many accidentally share misinformation due to inadequate decision-support mechanisms.
Also Read:
- AI Explainability: Integrating Large Language Models into Structured Decision Processes
- New AI System ED2D Uses Evidence-Based Debate to Combat Misinformation
Introducing Human-Layer AI (HL-AI)
A new research paper proposes a groundbreaking solution: Human-Layer AI (HL-AI). This innovative concept introduces user-owned, explainable AI intermediaries that operate directly within the user’s browser, sitting between the platform’s logic and the user interface. The core idea is to empower individuals with practical, moment-to-moment control over their social media experience without requiring direct cooperation from the platforms themselves. This approach aims to retrofit existing feeds with enhanced safety, agency, and well-being features.
The researchers have developed a working Chrome/Edge prototype that implements five key design patterns, offering a suite of humane controls:
- Context-Aware Post Rewriter: This pattern helps users reflect before posting potentially harmful content. It offers AI-generated rewrites for polarizing or biased phrasing, providing neutral or empathetic alternatives while always leaving the final decision to the user.
- Post Integrity Meter: Designed to combat misinformation, this feature provides real-time integrity signals for posts. It checks factual consistency, estimates the likelihood of AI-generation, and identifies political bias, offering transparent cues to help users judge credibility without restricting access.
- Granular Feed Curator: This gives users direct, actionable control over their feed composition and even ad categories. Through simple sliders and toggles, users can shape what content they see and how often, moving beyond opaque algorithmic defaults.
- Micro-Withdrawal Agent: To address compulsive scrolling, this pattern introduces brief, context-aware pauses. It detects patterns indicating a risk of compulsive continuation and offers lightweight reflective prompts or redirections, helping users disengage without imposing harsh blocks.
- Recovery Mode: This provides an immediate ‘shelter-in-feed’ option for users facing harassment, sudden online attacks, or emotional overload. It reduces harmful exposure, limits inbound interactions, and offers structured next steps for support.
The HL-AI framework is underpinned by a mathematical formulation that balances user utility, autonomy costs, and risk thresholds. This ensures that interventions provide meaningful benefits while avoiding excessive interference, especially when a user’s defined risk tolerance is exceeded. Every intervention is accompanied by a clear explanation and an option to override, preserving user control and transparency.
Implemented as a privacy-preserving browser extension, the HL-AI architecture dynamically modifies the social media interface without needing platform cooperation. It processes personal data on-device whenever possible, restricts cloud computation to anonymized aggregates, and ensures users maintain full control over data sharing and pattern activation.
This research offers a practical and human-centered path towards a more humane social media ecosystem, prioritizing user sovereignty, well-being, and safety. It invites further rigorous user evaluation across diverse cultures and contexts to refine and expand its impact. You can read the full research paper here: Towards a Humanized Social-Media Ecosystem.
