MGAP7001 Chap.9 Governance in the AI Era
Governance in the AI Era
Digital delivery does not create direct contact
The assumption that an online service reaches the citizen directly is usually implicit rather than argued, and it is what justifies counting a service as accessible once it exists.
Survey evidence from a country chosen for its near-universal connectivity, so that connectivity could not be the explanation, found close to half the population had never used the national digital identity despite near-total official coverage. Someone stands in that gap.
Intermediaries sort into close ones, meaning family, peers and neighbours acting from altruism or reciprocity, and hierarchical ones, meaning political figures, officials and community leaders acting on electoral or coverage incentives.
The same service is rendered either way and the accountability exposure is entirely different.
Inclusion of actors is not inclusion of outcomes
Representation narrows sharply from the general population to the technology workforce to the people building these systems.
A consultation that adds voices without changing the objective function delivers the first kind of inclusion and not the second, so a memo should say which one it is claiming. Two governing instruments are available and differ on content, actor, bindingness and consequence. Moralising works through voluntary guidelines with reputational consequences; regulating works through law that binds and sanctions.
They form a sequence, which lets an analyst date where a jurisdiction sits rather than simply approving of its approach.
A concrete proposal worth carrying is the algorithmic impact assessment, borrowed from environmental practice as a mandatory documented review before deployment.
A well-specified theory that broke
The risk-insurance account holds that exposure to an automating technology raises perceived personal risk, which raises demand for social protection.
A pre-registered experiment had participants perform work-like tasks with or without a generative tool. Attitudes moved and support for training provision and a job guarantee rose, but subjective risk did not move at all. Because the account requires that link, the policy demand cannot be running through it, which changes who the constituency is and how it should be addressed.
What this chapter covers
- 01
Digital intermediaries and the coverage-to-use gap
- 02
Close and hierarchical intermediaries compared
- 03
Inclusion of actors against inclusion of outcomes
- 04
Moralising and regulating as a sequence
- 05
Testing a mediator, and what a broken link proves
Reporting a broken mechanism without overclaiming
- 3Confirm the part of the claim the evidence supports, so the correction is usable.
- 3Name the link that did not move and say what that rules out.
- 3Offer the better working account and its targeting consequence.
Key terms
- Digital Intermediary
- A Digital Intermediary is the person who operates a digital public service on another citizen's behalf, most often a family member rather than an official.
- Coverage Gap
- A Coverage Gap is the distance between a service being available and a service being used independently, which availability statistics are structurally unable to detect.
- Algorithmic Impact Assessment
- An Algorithmic Impact Assessment is a mandatory documented review of an automated decision system before deployment, modelled on environmental impact assessment procedure.
- Mediator Test
- A Mediator Test checks whether a treatment moved the intermediate step a theory requires, which is what allows a result to falsify a mechanism rather than merely fail to support it.
Governance in the AI Era FAQ
What should a digital inclusion strategy measure instead of availability?
Independent completion rates, broken down by the groups with the lowest digital skills. Availability measures the supply side and will read as success in exactly the population that cannot finish a transaction unaided, which converts the unpaid labour of family intermediaries into a reported state achievement. If a strategy exists to serve low-skill groups, its headline measure has to be something only those groups can move.
When are voluntary ethical principles the right instrument?
While the technology and its harms are still unstable, because a binding rule written too early fixes on the wrong target and is hard to unwind. The instrument becomes an evasion once specific harms are documented and attributable, at which point continuing to rely on voluntary guidance is a choice about enforcement rather than a stage of development. Date the jurisdiction against that sequence before judging it.
Assessment move
For each of the three assumptions, write the one-sentence version of the assumption and the one-sentence version of the evidence against it. Then practise the mediator test on a causal claim from another topic, asking what the intermediate step is and whether anything has measured it. That move is portable and is the most reusable thing this topic teaches.
Working through Governance in the AI Era in MGAP7001? Sia is AskSia’s AI Management tutor — ask any MGAP7001 Governance in the AI Era question and get a clear, step-by-step explanation grounded in how MGAP7001 is taught and assessed. Read this chapter free, then take your hardest questions to Sia.
