Practical AI · 3-minute read
We gave computers enormous quantities of human material and then seemed surprised when some human habits showed up in the results.
To be fair, our species did invent both libraries and comment sections. The training environment was always going to be complicated.
The slogan is a simplification. An AI system’s behavior depends on its data, design, training, instructions and context of use. But the useful question underneath it is real: which assumptions are being carried into the answer, and whose experience is missing?
Yesterday’s wording can become today’s requirement
Consider Lantern Acre People Operations, a fictional HR consultancy invented for this example. A client wants an updated job description for an office coordinator. The team asks AI to refresh an old posting.
The old version asks for “recent graduates,” a degree the team has never tied to the actual tasks, and constant availability for a job with ordinary scheduled hours. A smoother rewrite could preserve all three assumptions while making them sound more official.
The useful assignment is to examine the requirements before polishing them. Which duties actually need a particular qualification? Which expectations are unclear? Which phrases describe the work, and which quietly describe the kind of person someone pictured doing it?
This example concerns drafting and review. It does not use AI to score candidates, infer personal traits or make hiring decisions.
Inspect what “normal” is hiding
NIST describes how human and systemic biases can enter decisions throughout the AI lifecycle, including design and use. The issue is broader than one bad word in a training set. It can involve what gets measured, whose feedback is considered and what an organization chooses to optimize.
For an everyday document review, start with three questions:
Relevant to what? Tie each requirement to an actual task or justified need.
According to whom? Ask whose perspective shaped the original and who should review the revision.
What is missing? Look for unclear conditions and unsupported assumptions that a fluent rewrite might conceal.
Try this prompt
Review this fictional job description against the duties supplied. Identify vague requirements, assumptions about applicants and qualifications that lack an explained connection to the work. Suggest clearer, task-based wording and questions for an HR professional to resolve. Do not evaluate candidates or infer protected traits. Do not claim the result is legally compliant. Duties: [duties]. Draft: [draft].
Use fictional material for practice. A qualified HR reviewer should check a real posting against the role and applicable requirements. AI can help surface questions, but it cannot certify fairness or replace that review.
The encouraging part is that we can change what we ask, examine the inputs and involve the people affected. “We have always written it that way” is a historical observation, not a quality standard.
AI learned from us. We are still allowed to learn too.
Your turn: Which familiar requirement at work deserves a fresh “Why is that necessary?”
Background: NIST AI RMF 1.0, Appendix C. The HR company, job posting and review prompt are original editorial examples, not legal guidance or a real hiring case.
