Gia
Designing a human–AI safety system for attention-critical environments.
Context
Motorcycling is a high cognitive load, safety-critical activity.
Riders continuously process speed, road conditions, navigation, and risk. Attention is fully occupied.
Most rider technologies fail in two ways:
- they increase visual load
- or they interrupt at the wrong moment
After experiencing a serious motorcycle accident, I reframed the problem:
Not how to add more information,
but how to intervene without increasing cognitive load.
Problem
This was not an interface problem. It was human–AI coordination under strict limits: attention already saturated, reaction time critical, interruptions risky, and no room to operate complex systems mid-ride.
The core question:
How do you design an AI system that improves safety without competing for attention?
In safety-critical systems, value comes from timing, not volume.
System principles
Gia was designed as a constrained system where attention is the primary limitation.
- Success = reduced cognitive load, not feature breadth
- Any interaction that pulls attention is failure
- Voice and haptics over visual interfaces
- AI supports judgment, not overrides it
- Learning happens through behavior, not configuration
System model
Gia is a real-time decision system that:
- Ingests environmental and behavioral signals
- Predicts moments where intervention improves safety
- Filters outputs to only high-value signals
- Communicates through voice and ambient cues
- Adapts to rider behavior over time
The system prioritizes what not to surface.
Interaction model
Gia operates on three rules:
- When to observe (continuous sensing)
- When to intervene (threshold-based triggers)
- When to stay silent (low-confidence or low-impact scenarios)
This ensures the system supports, rather than competes with, rider attention.
Design & prototyping
I focused on identifying when riders have cognitive availability versus overload.
This included:
- mapping high-risk vs low-attention moments
- testing voice timing under real riding constraints
- evaluating when intervention improves vs disrupts
Every decision was evaluated against:
Does this reduce cognitive burden in motion?
Voice
Voice was the primary interface.
When eyes and hands are occupied, spoken cues preserve situational awareness.
- short
- contextual
- timed to moments of availability
This was not conversational AI.
It was controlled intervention.
Key decisions
Haptics over visuals
- Decision
- We designed alerts to be primarily tactile rather than visual.
- Context
- Visual attention is already fully occupied while riding.
- Tradeoff
- Reduced information density.
- Impact
- Lower cognitive load and safer interaction model.
Selective intervention
- Decision
- We constrained when the system surfaces recommendations.
- Context
- Constant input creates distraction and distrust.
- Tradeoff
- Less “assistive” presence.
- Impact
- Higher signal-to-noise ratio and better rider trust.
Behavioral learning over configuration
- Decision
- We avoided heavy onboarding configuration.
- Context
- Riders won’t configure systems before riding.
- Tradeoff
- Slower initial system accuracy.
- Impact
- Lower friction and more natural adaptation over time.
Explicit failure states
- Decision
- We defined how the system behaves when data is incomplete or uncertain.
- Context
- AI systems are inherently probabilistic.
- Tradeoff
- Conservative outputs vs more “impressive” features.
- Impact
- Increased reliability and safety perception.
Outcomes
Gia demonstrated a viable model for AI systems in attention-constrained environments:
- reduced reliance on visual interfaces
- higher trust through predictable intervention
- clearer signal-to-noise ratio
- improved alignment between system behavior and user expectation
The project validated that:
AI effectiveness in safety-critical contexts depends on restraint and timing, not capability.
Learnings
- Attention is the primary constraint in safety systems
- AI value depends on knowing when not to act
- Trust requires supporting, not replacing, human judgment
- Adding features often reduces safety
Reflection
Gia clarified a shift in how I design AI systems. The challenge is not intelligence. It is control. Designing when a system should act, remain silent, or defer to human judgment requires defining boundaries, not features.
This project strengthened my ability to design systems where:
- attention is limited
- decisions are time-critical
- and failure has real consequences