As ML shapes more of daily life, the question is not just can it work, but does it work for people. That is what Human-Centered Machine Learning tackles.
Human-Centered Machine Learning (HCML), also called Human-Centered AI (HCAI), builds ML systems that prioritize the needs and well-being of the people affected by them. It rests on three principles: humans and AI collaborating (not AI replacing humans), responsible and user-friendly AI (fair, ethical, safe, responsive), and a human-centered design process. Its key pillars are Explainable AI (XAI), tackling ML bias, ML ethics, and ML governance.
Key takeaways
- HCML puts the people affected by an ML system at the center of its design.
- It is a subset of HCAI, which aims to complement human abilities, not replace them.
- Explainable AI (XAI) makes model decisions understandable, building trust.
- ML bias reproduces societal biases; fixing it needs diverse data and community engagement.
- ML ethics and ML governance keep systems aligned with societal values.
The three core principles
- Collaboration and co-creation: humans + AI outperform either alone; design for human-AI interaction.
- Responsible, user-friendly AI: fair and unbiased, ethical and safe, and responsive to real user needs.
- Human-centered design process: understand user needs and limits, and make systems easy and efficient to use.
The four pillars
| Pillar | What it means | Goal |
|---|---|---|
| Explainable AI (XAI) | The system provides understandable, interpretable explanations for its predictions and decisions. | Trust |
| ML bias | The tendency to perpetuate societal biases (race, gender, culture); addressed with diverse data and transparency. | Fairness |
| ML ethics | Moral and societal considerations: privacy, autonomy, accountability, and fair distribution of benefits and harms. | Responsibility |
| ML governance | The policies and institutions ensuring ML is built and used in line with societal values: regulation, accountability, participation. | Oversight |
The field is active across industry and academia. MIT and Google Design have run HCML research, and Stanford launched its Institute for Human-Centered AI in 2019. At Dorve, these ideas feed directly into Quantum UX, our data-and-AI-driven approach to designing experiences responsibly.
An approach to building ML systems that prioritizes the needs and well-being of the people affected by them, keeping humans in control and the loop.
HCML (Human-Centered Machine Learning) is a subset of HCAI (Human-Centered AI). HCAI is the broader goal of AI that complements rather than replaces human capabilities.
The ability of an ML system to give understandable, interpretable explanations for its predictions and decisions, which builds trust and helps detect bias.
The tendency of ML systems to reproduce and reinforce existing societal biases based on race, gender, culture and more, which can harm affected communities.
The practices, policies and institutions that ensure ML systems are developed and used in ways aligned with societal values, covering regulation, accountability and transparency.
Building with ML and want it human-centered and fair? Talk to Dorve.
Start a project