Mauritius Needs an Ai Decision-Rights Map Before it Scales

Forum

By Dr Gleb Tsipursky

In his August 31 interview with Mauritius Times, IT Minister Avinash Ramtohul made the central challenge of the country’s AI push unusually clear. Mauritius can invest in infrastructure and tools, but it still has to build capability inside government, universities, businesses, and communities. He also stressed that AI cannot substitute for leadership or administrative responsibility.

That distinction should shape the next stage of adoption. Technical skills matter. Yet people also need explicit rules for when AI can act, when it can only recommend, when a human must verify its output, and who carries responsibility when the answer matters.

Mauritius should turn those rules into a simple decision-rights map for AI-enabled work.

A decision-rights map begins with the work itself. Every organisation can divide recurring tasks into four categories: AI may execute with routine monitoring; AI may recommend but a person approves; AI output requires independent verification before action; and the decision remains human because accountability, context, or consequences demand it.

This classification prevents two common operational problems. Without clear boundaries, employees may trust polished AI output more than they should. Others may respond by reviewing everything so heavily that automation saves little time. A shared map gives teams a middle path based on the consequences of error.

The second part should define verification standards. “Human in the loop” sounds reassuring, but it says little about what the human should actually do. For a low-risk internal draft, a quick check may suffice. For a procurement recommendation, benefits decision, financial action, hiring judgment, or public communication, the reviewer may need to compare the AI output against independent records, policy, or professional standards before approving it.

That matters because review quality depends on whether the reviewer retains enough knowledge to recognize a problem. If people gradually stop doing the underlying work, they can become supervisors in name while losing the competence needed to challenge the system. Organizations should therefore identify which skills employees must continue to practise without AI and test those skills periodically.

The third part should name escalation and override authority. Employees need to know who can stop an AI-enabled workflow, who resolves disagreement between an automated recommendation and professional judgment, and what evidence should trigger escalation. These rules become especially important when AI moves from drafting or search into workflows that affect citizens, customers, employees, money, or safety.

The fourth part should create a feedback loop. When an employee catches an AI error, overrides a recommendation, or discovers a useful new application, the organisation should record what happened and update the workflow. The goal is not a large bureaucracy. A short exception log can show where models repeatedly fail, where review consumes too much time, and where guardrails can safely loosen because performance has become reliable.

Mauritius already has a useful place to teach these habits. The new AI Innovation Lab* was created for practical learning and AI experimentation by students, professionals, technology entrepreneurs, and members of civil society. Training there can go beyond learning what tools can do. Participants can practise classifying decisions, checking outputs, documenting exceptions, and designing human escalation paths.

Universities can use the same approach with students. Employers can apply it to customer service, finance, HR, operations, and knowledge work. Government agencies can adapt it to public services, administrative decisions, and internal analysis. The categories will differ, but the governing question stays the same: what authority are we delegating, and what human competence must remain strong enough to supervise it?

The minister’s warning that Mauritius cannot purchase its way into becoming an AI nation points in the right direction. Capability grows through repeated practice inside real decisions. Infrastructure expands what AI can do. A decision-rights map helps institutions decide what AI should do, what people must still understand, and where responsibility ultimately sits.

* AI Innovation Lab:  <https://mitci.govmu.org/mitci/first-of-its-kind-ai-innovation-lab-launched-to-democratise-access-to-emerging-technologies/>

Gleb Tsipursky, PhD, is a behavioural scientist, CEO of Disaster Avoidance Experts, and author of ‘The Psychology of AI Adoption at Work: From Resistance to Results, (Georgetown University Press, 2026). https://disasteravoidanceexperts.com/aibook

Mauritius Times ePaper Friday 11 September 2026