Nonlinear Impact · Decision framework
The Responsible AI Compass
A design guide when you plan & build AI systems — and a diagnostic map when an AI deployment goes wrong and you need to find out where.
Before deployment
Name who this will affect, who has access, who is accountable — and whether the technology does what you intend it to do.
After deployment
Start with who was harmed or is at risk, then trace back: which decision, which governance gap, which design choice allowed it.
Purpose
What is the organisation for, and why does this problem need AI at all? The conversation most teams skip — and the one every failure traces back to.
- Mission alignment
- Problem definition
- Non-AI alternatives
- Success criteria
- Who asked for this
- Cost of doing nothing
Tech
Everything decided while the system is being made: what it learns from, how it fails, and whether anyone can explain it afterwards.
- Training data & quality
- Model choice & complexity
- Explainability
- Fairness & bias testing
- Safety & failure modes
- Privacy & security
Governance
Who is allowed to do what, who decides, who answers for it, and how someone harmed can push back.
- Access rights
- Deployment decisions
- Accountability & audit trails
- Transparency & disclosure
- Redress & contestability
- Ethical policy & blocklists
- Culture & AI leadership
People & Planet
Who this AI will affect, what it will cost the planet, and whether the people most at risk had any say in the decision.
- Beneficiary mapping
- Community consent
- Who owns the data
- Human displacement
- Do no harm
- Hidden costs of AI
- AI's ecological footprint