Policy : Use AI to Augment Human Capability, Not Replace Judgment

Commitment to AI as Human Augmentation The most valuable AI systems are those that make people more capable — not those that remove people from the loop. There is a seductive logic to full automation: AI systems do not tire, do not have bad days, and do not require salaries. But consequential decisions require judgment that integrates context, values, accountability, and the kind of tacit knowledge that humans accumulate through experience and that current AI systems cannot replicate. Our commitment is to design AI systems that amplify what humans are good at — handling novel situations, applying values, maintaining accountability, and exercising judgment in genuinely ambiguous conditions — rather than systems that automate away the parts of work that require exactly those capabilities.

What This Means Augmenting human capability means positioning AI as the tool and the human as the decision-maker. It means building systems that present AI analysis, recommendations, and predictions to humans in ways that enable better decisions — faster, with more information, with uncertainty surfaced clearly — while preserving human authority over the final decision. It means being deliberate about which decisions benefit from AI automation and which require human judgment, and not conflating technical capability to automate with wisdom about whether automation serves the organisation and the people it affects.

Our commitment to AI as human augmentation is built on:

• Decision Authority Mapping – For every AI system that informs decisions, we explicitly map which decisions are fully automated, which are AI-assisted with human sign-off, and which remain human-only. This mapping is informed by stakes, reversibility, regulatory requirements, and the availability of reliable signals — not purely by what is technically feasible to automate.
• Explainable Recommendations – AI systems that support human decision-making present their recommendations with enough context for the human to evaluate, challenge, and override them. Opaque scores without explanation undermine rather than enhance human judgment.
• Appropriate Friction for High-Stakes Decisions – Where decisions have material consequences, we design appropriate friction into AI-assisted workflows — confirmation steps, explanation requirements, and escalation paths that preserve human deliberateness rather than engineering it away.
• Override and Escalation Mechanisms – Every AI-assisted decision workflow has a clear, low-friction mechanism for the human to override the AI recommendation or escalate to a more experienced reviewer. Override rates are monitored as a signal of AI quality and human trust.
• Expertise Retention – We actively guard against AI systems eroding the human expertise they depend on. Where AI handles routine cases, we ensure humans remain engaged with sufficient case complexity to maintain their judgment and the institutional knowledge needed to supervise the AI effectively.
• Human Accountability Preservation – Accountability for consequential decisions remains with identifiable humans, not with AI systems. AI-assisted decisions must have a named accountable person — "the AI decided" is not an acceptable accountability structure for material outcomes.
• Augmentation Quality Measurement – We measure whether AI assistance actually improves human decision quality — not just speed. Systems that make humans faster but not more accurate, or that create the appearance of informed decision-making without the substance, do not meet the augmentation standard.

Why This Matters The assumption that automating human judgment always creates value is wrong. In domains with high stakes, genuine complexity, and significant edge case frequency, removing human judgment creates systems that are fast but brittle, efficient but unaccountable. The most durable AI value is delivered by systems that make the humans in the loop better — more informed, faster, less fatigued by routine processing — while keeping them engaged with the decisions that genuinely require human capability.

Our Expectation AI systems that affect consequential decisions are designed with explicit human oversight provisions appropriate to the stakes involved. Teams building AI that automates away human judgment without deliberate design authority to do so are not delivering efficiency — they are creating accountability voids. Using AI to augment human capability, not replace human judgment, is how we deliver Value that is sustainable, trusted, and genuinely useful to the people who use our systems.