Human in the Loop
Human in the Loop
The Design Principles
Many Minds, Many Cursors:
Designing for the collective cognition of the human and the machine
The interface is changing.
We are entering a world where AI doesn’t just assist and respond, but acts as a collaborative partner in the innovation process. In this shift, intelligence becomes collective. Decisions, actions, and intentions are co-authored by human users and autonomous agents.
Goal:
To prototype and define principles for systems where humans and agents co-create, maintaining agency, trust, and legibility across the collaboration.
Human-in-the-loop (HITL) machine learning is a collaborative approach that integrates human input and expertise into the lifecycle of machine learning (ML) and artificial intelligence (AI) systems. This integration forms socio-technical ensembles of humans and machines, which aim to accomplish complex goals by combining human and artificial intelligence to collectively achieve superior results and continuously improve by learning from each other.
To promote optimal human-AI collaboration and performance, several design principles are crucial. These have been my guiding forces through the creation of this project.
Prioritizing Human Control
and Agency
This system is designed for high levels of human control AND high levels of computer automation to increase human performance. It supports human self-efficacy, mastery, creativity, and responsibility - ensuring people remain in charge. AI acts as a tool to amplify, augment, and empower, always in service of human intent.
Ensuring Transparency, Explainability, and Understandability
This system draws a clear line between seamless and deceptive design by making its actions and reasoning visible. Users are informed of what the system has done - and what it intends to do next. Interpretability is a foundation for trust, enabled through transparent algorithms, understandable models, and explainable outputs.
Managing Reliance and Trust (Promoting Cognitive Engagement)
This system is designed to reduce blind reliance on AI by encouraging active, analytical thinking. Rather than defaulting to AI suggestions, users are prompted to engage- through interventions like decision-first workflows, paced disclosures, or on-demand guidance. While this may increase cognitive effort, it builds more durable trust by supporting informed, intentional use.
Adaptive and Context-Aware Support
This system adapts its assistance to the task, the individual, and the moment. It leverages approaches like reinforcement learning to optimize for a range of human-centered goals - not just accuracy, but also skill development, confidence, and task satisfaction. Support is tailored dynamically based on the concept at hand, the user’s knowledge, and their cognitive needs.
Facilitating Human Input and Feedback Mechanisms
This system is built to learn with and from its users. It supports both explicit teaching - like labeling or correction - and implicit feedback, such as adapting to behavior and preferences over time. Human input is essential across the ML pipeline, and the design enables efficient, ongoing collaboration to refine outputs and improve performance.
Design Explorations
1. Multi-Agent Field + Cursor Behaviors
How might we visually distinguish between human and AI agency in shared digital spaces?
To be able to create this distinction, it is essential to first understand the difference in responsibilities that the human vs the agent would have in these spaces.
Research : I talked to 5 of my colleagues and friends to understand how they currently use AI in their work day. What are tasks that they would like to keep doing, vs what are tasks they would rather automate. I wanted to understand what kind of activities required active participation, vs what could be achieved through simple supervision.
Here’s what I learned.
Humans and AI have complementary strengths and weaknesses.
To promote optimal human-AI collaboration and performance, I explore creating a robust design strategy.
This starts with the understanding of novel design principles, and ends with a Case Study designed for optimum collaboration.