Research-led analysis · 14 November 2026
Who decides? Who gets paid? Who gets human time?
A payment, a say in a decision and time with a person offer different kinds of protection. Here are four questions to ask when someone promises to protect people as AI advances.
Two selected arrangements, described at the editions below. This is an argument about policy design; current applicability and outcomes have not been established.
A person can need more than one kind of protection
Imagine a performer whose digital replica delivers new lines, and a resident who needs someone to help them eat, move or explain a care decision. These are illustrations of different policy questions. The performer might want control over a use and payment for it. The resident might need human help to be available, with enough staff time to make the relationship real.
A rule reserving work to people, a consent requirement, a compensation claim and a staffing obligation do different things. They can be combined, but one cannot be taken as evidence that another exists. Paying someone when their work is automated does not ensure they keep practising it. Reserving a service to people does not ensure someone can afford or obtain it.
Human by choice is exploring how people can keep meaningful control and share AI’s economic gains. Before backing a proposal, ask what it secures: human work, authority, an income claim or human support. Then ask who remains outside the promise. The four questions below can help examine a proposal at work or in a public service.
Payment can accompany substitution
SAG-AFTRA’s published 2022–2028 Interactive Media Agreement links defined Digital Replica uses to consent, compensation and usage reports, subject to exceptions. This account covers the ordinary Digital Replica regime, not separately regulated Independently Created Digital Replicas. It is a contractual example, not a universal law; later changes, interpretation and individual coverage remain unchecked.
Payment can accompany machine-generated performance. It may recognise claims over identity or prior labour. That does not establish new human work, higher total earnings or gains for the wider workforce.
There is a difficult distributional question. An identifiable established performer may have leverage that a newcomer or an unrecognisable contributor lacks. A compensation regime could benefit some workers while reducing opportunities for others. That is a hypothesis to investigate, not an outcome demonstrated by this agreement.
Sharing AI productivity gains across society raises further choices: who qualifies, what value is measured, whether payments are individual or collective, how claims are checked, and how people outside a particular contract participate. A convincing proposal has to answer those distributional questions.
Human care needs time, not only a boundary
If the aim is access to a person, a rule also needs to address whether that person has time to help. Australia’s aged-care requirements offer a comparison. The inspected October 2026 compilation requires covered residential-care providers to meet calculated care-time averages, including a nursing component. These are home-level calculations with programme-specific conditions, rather than identical minutes promised to every resident. Applicability on 14 November 2026 still needs checking.
The department separately describes public funding for eligible care homes, including a performance-linked supplement. This is an official scheme account; we have not checked the full statutory funding chain, actual payments or funding adequacy.
The lesson for an AI proposal is to ask both who should provide a service and what human capacity must be available to deliver it. This comparison is not presented as an AI ban.
Time is also an imperfect measure. A home-level average can conceal different experiences, and a documented minute need not be attentive or useful. Recruitment shortages can make an obligation difficult to meet. A credible proposal has to examine quality, individual need, staff supply, access for people waiting outside the service and the consequences of diverting money from other uses.
Four questions for any proposed protection
First, what would be lost? Identify the value: a relationship, accountable judgment, creative practice, control of identity, income or access to support. Explain whether it would remain valuable even if an AI system became more capable. A temporary performance gap and an enduring reason for a human role lead to different policies.
Second, who should have what power? A worker may need a say before a workflow changes. A service user may need a person who can change the outcome. A creator may need control over a use. Name the actor, trigger and actual authority; a signature without time or discretion may add little.
Third, who receives the benefit and who pays? Separate the person performing the work from the person receiving a payment or service. Ask about entrants, contractors, unpaid carers, people outside a bargaining agreement and those who cannot get into the service. Then identify money, staff time, information and enforcement capacity.
Fourth, what would count against the proposal? Test whether it worsens queues or prices, protects incumbents without public benefit, imposes ceremonial checking, or blocks useful assistance. Compare it with funded human escalation, worker negotiation, outcome standards or a different distribution of AI gains. Explain what evidence would change the choice.
A practical test for a UK proposal
Consider a hypothetical UK service exploring AI-assisted administration. No saving or improvement is assumed. An initial proposal might retain staff and redirect any verified time saving into conversations with service users. Another might reduce working hours while preserving pay. A third might use the saving to increase throughput. Each distributes value differently.
For each option, ask for a written answer: how would time saved be measured, who would decide how to use it, and who would gain or lose? If the promise is more conversation, which staff time would be kept for it, and how could service users check that it happened? Check who has authority and funding to deliver the choice. Include people waiting for the service, as well as current staff and users.
The overseas examples cannot settle this UK choice. Applying a mechanism in a new setting requires checking who benefits, institutional responsibilities, costs and how a commitment could be enforced. Begin with the intended benefit and available alternatives, rather than copying a rule because its headline sounds protective.
Put the questions to work
Take a proposed AI change into a workplace, union or service-policy discussion. Make a short record: what should be protected; who has authority to secure it; who benefits and pays; and what evidence would change the plan. State what remains unknown and where people disagree.
The next test is what happens in practice: who receives money, authority or time; who is excluded; what it costs; and whether another arrangement would serve people better. Those findings should be allowed to change our categories and priorities.
This small research programme has not established which policies work best or what workers and care users want. It does not speak on their behalf.
When a proposal promises human protection, ask: who decides, who benefits, and what must actually be provided?
Original sources and limits
The locators identify the basis for the comparison. A live page may change after inspection.
- AUACT2
Aged Care Act 2024, Compilation 2 ↗
Locator: Section 7 definition of direct-care staff; sections 175–176 for context.
Edition: Compilation dated 1 July 2026, inspected 14 November 2026.
Limits: Used for definitions and statutory context. Later amendments and current applicability are not fully checked.
- AURULE13
Aged Care Rules 2025, Compilation 13 ↗
Locator: Sections 176-5–176-20, particularly 176-15 and 176-20.
Edition: Compilation dated 19 October 2026, inspected 14 November 2026.
Limits: Named-edition comparison. The full commencement and amendment chain, and applicability on 14 November 2026, remain unchecked.
- AUFUND
Australian department: care minutes supplement ↗
Locator: Applies to; What is the supplement; How the supplement works; Reporting.
Edition: Official guidance dated 31 March 2026, inspected 14 November 2026.
Limits: Contextual scheme account. The full statutory funding chain, actual payments and funding adequacy were not audited.
- SAGIMA
SAG-AFTRA 2022–2028 Interactive Media Agreement ↗
Locator: Article I §47 and Exhibit D, especially definitions and §§1–6; PDF pages 85–91.
Edition: Publisher-linked copy inspected 14 November 2026; Exhibit D states prospective effect from 12 October 2025.
Limits: Defined Digital Replicas only in this comparison; the separate Independently Created Digital Replica regime is excluded. Later amendments, interpretation and case-specific coverage are not fully checked.