01
Reveal the system
People should know when AI is involved, what role it is playing, and where human judgment still enters the process.
AI ethics + interaction design
Ethics becomes real in the interface.
Principles matter, but people experience AI through prompts, defaults, permissions, explanations, waiting states, errors, and decisions. Interaction design is where responsibility becomes visible.
Ethics interface
Human approval requiredDecision
Understandable
Controllable
Reversible
What happened
Explain
What can change
Control
What happens next
Recover
The core test
Can a person understand, question, and change what the AI is doing?
Ethical interaction design does not require people to understand the model architecture. It gives them enough context and control to make an informed decision.
- What is the system doing right now?
- Why did it produce this result?
- How certain is it?
- What data or permissions is it using?
- What can the person change, reject, or undo?
- Who is responsible when the system causes harm?
Six interaction principles
Responsible AI is a product behavior, not a policy page.
02
Calibrate confidence
The interface should communicate uncertainty without turning every answer into noise. Confidence must match the evidence.
03
Preserve agency
People need meaningful ways to edit, refuse, pause, override, and leave. A decorative cancel button is not control.
04
Make consent specific
Ask for permission at the moment it matters, explain the consequence, and avoid bundling unrelated data uses together.
05
Design recovery
Errors are inevitable. Ethical systems make correction, appeal, rollback, and human support easy to find and understand.
06
Expose accountability
The product should make ownership visible. People need to know which organization, team, or person can answer for a decision.
Ethical product delivery
Design responsibility across the whole interaction.
A trustworthy moment cannot compensate for a misleading onboarding flow, an opaque automated action, or a dead-end appeals process.
1
Before interaction
Expectation- Is AI involvement visible?
- Are permissions and data use understandable?
- Does the product promise more than it can reliably do?
2
During interaction
Control- Can the person inspect and challenge the output?
- Are uncertainty and consequences visible?
- Can automation pause before high-impact actions?
3
After interaction
Recovery- Can the result be corrected or reversed?
- Is there a clear appeal or escalation path?
- Can the team trace what happened and why?
Interaction anti-patterns
Harm often arrives disguised as convenience.
Hidden automation
The interface makes an automated decision look like neutral system output.
False certainty
Probabilistic output is styled as fact, recommendation, or authority.
Consent theatre
A person can technically agree, but cannot understand or meaningfully refuse.
Anthropomorphic pressure
Human-like language is used to create trust, guilt, intimacy, or obedience the system has not earned.
Opaque personalization
The product changes what people see without showing why or offering a way to reset it.
Irreversible action
The system acts faster than a person can inspect, stop, or recover from the consequence.
HAAM approach
Start with the decision, then design the model around it.
We map who is affected, what the system is allowed to decide, which actions require approval, how uncertainty appears, and what recovery looks like before polishing the interface.
The goal is not to make AI feel human. It is to make the product honest, legible, useful, and accountable.
Design AI people can question, correct, and trust.
HAAM connects AI governance to the screens, states, decisions, and recovery paths people actually experience.