The HUMAN Test for AI-Enabled Work
A practical five-part decision aid for leaders assessing whether an AI-enabled workflow protects human purpose, participation, oversight, accountability and future capability.
Suggested citation: Khatani, S. A. (2026, October 9). The HUMAN Test for AI-Enabled Work. Personal website of Dr. Salman Ahmed Khatani.
AI adoption is a work-design decision
An organisation may purchase an AI system in weeks, but redesigning work responsibly takes longer. Leaders must decide which problem the system should solve, whose judgement it will influence, how errors can be challenged, who remains accountable and what capabilities people will need next.
The International Labour Organization's 2025 global index estimates that one in four workers are in occupations with some degree of generative-AI exposure. Because most occupations still contain tasks requiring human input, the ILO identifies transformation, rather than full replacement, as the more likely effect and emphasises social dialogue.
The OECD AI Principles connect trustworthy AI with human rights, fairness, transparency, robustness and accountability. The NIST AI Risk Management Framework provides a voluntary risk-management structure; NIST states that AI RMF 1.0 is being revised in 2026. The HUMAN Test translates these broad principles into five practical questions for one workplace workflow.
H: Human purpose
Core question: What worthwhile human outcome should this system improve?
Start with the work problem, not the product. “We need AI” is not a purpose. Reducing repetitive documentation, improving accessibility or creating more time for human interaction are clearer aims. Leaders should also name what must not be weakened, such as independent judgement, care, professional responsibility or the quality of a consequential decision.
Evidence to record: defined problem, intended users, baseline, desired outcome, non-AI alternatives considered and the human capability or value to preserve.
U: User voice and inclusion
Core question: Have the people who perform and experience the work influenced its redesign?
Employees should not first encounter a system during mandatory training. People closest to a workflow often know where exceptions occur, what information cannot be trusted and which parts depend on tacit knowledge. Participation should include people affected by the output, not only those operating the tool.
In Pakistani organisations this may reveal differences in language confidence, device access, disability access, workload, digital capability and authority to question a supervisor or automated recommendation.
Evidence to record: groups consulted, concerns raised, changes made, accessibility and language tests, excluded users and workable alternatives.
M: Meaningful human oversight
Core question: Can a qualified person understand, challenge and stop the system's influence?
“Human in the loop” is meaningful only when the person has relevant knowledge, enough time, access to supporting information and authority to intervene. The level of review should rise with the consequence of the decision: drafting an internal summary and recommending disciplinary action do not require the same controls.
Evidence to record: reviewer's role and competence, review point, verification information, escalation route, shutdown conditions and prohibited uses.
A: Accountability, evidence and appeal
Core question: Who owns the outcome, how will performance be evaluated and how can an affected person seek review?
Accountability cannot be delegated to a vendor or model. A named organisational role should own the decision to introduce the system, its continuing performance, incident response and withdrawal when harms exceed value.
Adoption, output volume and time saved are incomplete measures. Quality, correction rates, employee workload, accessibility, affected-user experience and the distribution of benefits may matter more. When an AI-influenced decision affects employment, assessment, opportunity or access to a service, the person affected should have a route to explanation and human review.
Evidence to record: accountable owner, baseline, benefit and harm indicators, evaluation date, incident log, appeal route, stop criteria and exit plan.
N: New capability and next futures
Core question: What must people learn, and what future could this workflow create if it scales?
Training should extend beyond operating a tool. Employees may need to frame problems, verify outputs, recognise uncertainty, protect sensitive information, document AI use and decide when not to automate.
Strategic foresight adds a second test: imagine the workflow after one or two years. One future may strengthen human capability; another may distribute advantages unevenly; a third may create platform dependence or erode professional judgement. These are not predictions. They help leaders notice consequences before scale makes them difficult to reverse.
Evidence to record: capability gaps, learning plan, protected practice time, three plausible future conditions, early indicators, continuity plan and review triggers.
The 15-point HUMAN checklist
Human purpose
- We can name the work problem and the human outcome to improve.
- We compared AI with simpler or non-AI alternatives.
- We identified a human capability, relationship or right that must not be weakened.
User voice and inclusion
- Employees performing the work helped identify needs, exceptions and risks.
- People affected by the output were represented in testing or review.
- Access, language, disability, workload and digital-capability differences have workable responses.
Meaningful oversight
- A qualified person can verify important outputs before consequences occur.
- The reviewer has enough information, time and authority to challenge the system.
- Escalation, prohibited-use and shutdown conditions are documented.
Accountability, evidence and appeal
- A named role owns the system's outcomes and continuing review.
- The pilot has a baseline, benefit indicators, harm indicators and stop criteria.
- People affected by consequential decisions can obtain explanation and human review.
New capability and next futures
- Employees will learn verification, judgement and safe-use skills, not only tool operation.
- We tested the workflow against at least three plausible future conditions.
- Review triggers and an exit or continuity plan are in place.
Decision rule
Do not average away a serious weakness. A Stop relating to sensitive data, consequential decisions, meaningful review, appeal or accountable ownership means the workflow is not ready to scale. “Needs work” items should become named actions with owners and dates. A bounded pilot is appropriate only when the leadership team can show evidence for every high-consequence safeguard.
This rule is intentionally qualitative. A neat numerical score can create false confidence when one unresolved issue could cause significant harm.
A 60-minute leadership exercise
- Define one workflow, 10 minutes. Choose a specific task, such as AI-assisted recruitment screening, customer-response drafting or faculty feedback.
- Map the people, 10 minutes. Identify who performs the work, who receives the output, who may be excluded and who carries the consequence of an error.
- Complete the HUMAN Test, 20 minutes. Require a document, observation or named source for every Ready judgement.
- Stress-test the future, 10 minutes. Consider capability strengthening, uneven advantage and capability erosion. Name one early indicator for each.
- Make a decision, 10 minutes. Select stop, redesign, bounded pilot or continue with a scheduled review. Record the owner and next decision date.
Illustrative example
Imagine a Pakistani services firm considering an AI tool to draft customer responses. The stated purpose is faster service, but employees identify that complex cases involve Urdu-English switching and exceptions not captured in standard procedures.
The HUMAN Test changes the pilot. Routine drafts may proceed with employee review, while complaints involving refunds or vulnerable customers require specialist approval. The firm measures correction rates and customer resolution, not only speed. Staff receive time to practise verification, and leaders monitor whether reliance on suggested wording reduces employees' ability to explain decisions independently.
This example is hypothetical. It does not describe a client engagement or measured result.
Leadership principle
Human-centred AI is not achieved by adding a person at the final approval stage. It is achieved when human purpose, participation, judgement, responsibility and future capability shape the workflow from the beginning.
Authoritative sources
- International Labour Organization, Generative AI and jobs: A 2025 update, published 20 May 2025.
- OECD AI Principles, adopted in 2019 and updated in May 2024.
- NIST AI Risk Management Framework, version 1.0 released 26 January 2023 and listed by NIST as under revision in 2026.
Sources checked 9 October 2026. The HUMAN acronym, questions, checklist, decision rule and exercise are original proposed analysis by Dr. Salman Ahmed Khatani.