The Framework
HI — HO
HI-HO is an ethical framework for using generative AI. It separates what only humans should do from what the AI may do — and binds the entire process to a single standard: the non-marginalization of humans.
HI · Human Input
The human controls the information and details that go into the AI.
You decide what context, data, framing, audience, and intent reach the model. Nothing is generated from a vacuum. HI is the moment where authorship begins and where bias, scope, and ethical guardrails are set.
Human Input
The human is the source of intent. You define purpose, audience, context, evidence, standards, and boundaries — and you protect private or sensitive information. A prompt is not the beginning of thought; it is the visible expression of thinking that has already begun.
The Dash: AI Understanding & AI-Assisted Work
The dash is AI literacy in action. It covers tool selection, model awareness (AI predicts responses rather than knowing facts), information and source awareness, privacy, iteration, and limitation awareness. ChatGPT, Claude, Gemini, Perplexity, Copilot, and NotebookLM are not interchangeable — the best tool depends on the purpose, the information involved, and the outcome expected.
Human Oversight
Oversight is more than proofreading. It asks about verification, relevance, fairness, privacy, consequences, professional judgment, and accountability. AI can offer a recommendation; it cannot accept responsibility for the decision. Ask: would I be comfortable explaining how this was created, how I checked it, and why I decided to use it?
From the foundational guide
The HI-HO™ Practice Cycle
A repeatable cycle — simple enough for everyday use, rigorous enough for classrooms, professional practice, and organizational decision-making. It repeats; responsible AI use is iterative rather than linear.
1. Intend
State the human purpose. What are you trying to accomplish? Who may be affected? What would a responsible outcome look like?
2. Inform
Give the AI useful context, evidence, criteria, boundaries, examples, and audience expectations.
3. Interact
Ask, revise, challenge, clarify, and try again. Treat the first response as a starting point, not a finished product.
4. Interpret
Consider how the chosen system, model, sources, and limitations may have shaped the response.
5. Inspect
Check important facts, citations, calculations, privacy concerns, bias, tone, and professional or academic standards.
6. Integrate
Decide what to keep, change, reject, disclose, or use — and accept responsibility for the final product and its effects.
Common misunderstandings
“AI produced it, so AI is responsible.”
AI does not hold responsibility. The person or organization choosing to use the output remains accountable.
“A detailed prompt guarantees a correct answer.”
A strong prompt improves direction, but it does not remove the need for verification and judgment.
“All AI tools are basically the same.”
Systems differ in features, information access, privacy, source handling, and suitability for particular tasks.
“Human oversight means checking grammar.”
Oversight includes accuracy, fairness, relevance, privacy, consequences, and professional standards.
“Ethical AI means using less AI.”
Ethical use is measured by whether humans remain purposeful, informed, engaged, and responsible.
“Disclosing AI use is enough.”
Transparency matters, but disclosure does not replace sound input, informed use, or careful oversight.
The non-negotiable
Humans are aided, not replaced. Human dignity is the ultimate ethical standard. Any application of generative AI that marginalizes humans — by skipping HI, collapsing HO, or treating the dash as a decision-maker — is out of alignment with the framework.