Human+AI Interaction : Question -> Frame -> Unpack -> Own -> Iterate

Human+AI Interaction : Question -> Frame -> Unpack -> Own -> Iterate

By Christine Braun

A framework for educators to use with students to understand when and how to use AI. 

Human+AI Interaction is the discipline of engaging with artificial intelligence as critical collaborators. Collaborators plural as AI is trained through learning from inputs which are added by humans. Making it a global civilization of human knowledge: a community fed voice. An AI does not think in the human sense; it generates responses based on patterns: Generating the most statistically likely useful response based on patterns learned from a collective community voice created by data inputs and media. This means that every output carries the fingerprints of every human choice that shaped its training. Including all of the flaws, biases, challenges and knowledge held within any community.

• What AI tools actually are:
  • AI tools are trained, large language models.
  • Through that training it learned statistical patterns — which words, ideas, and structures tend to follow each other in meaningful ways. This also means it learned from our shortcomings and biases.
  • Image generators were trained on what the world looks like from millions of human-made images — which means they didn't just learn to make pictures, they learned every assumption about beauty, identity, and power that was embedded in those pictures.

The three things worth understanding about Text-based AI:

1. It learned from human choices. The text it trained on was written by people — with all their perspectives, blind spots, cultural assumptions, and gaps.
2. It generates — it doesn't verify by default. AI tools produce fluent, confident-sounding text.
3. That fluency is not the same as accuracy. It can be wrong, incomplete, or subtly skewed while the answer appears confident. The human must always be the judge.

The three things worth understanding about Generative AI:

1. It learned what things "look like" from human-made images. It learned every cultural assumption baked into those images. The biases aren't added, they're inherited.
2. It has never seen the world. It has only seen representations of the world made by humans. It doesn't know what a tree is, it knows what images labeled "tree" look like. 
3. It generates, it doesn't retrieve. It isn't finding an existing image. It's constructing a new one: something that looks completely real and credible and is entirely fabricated.

The Human+AI Interaction is a five step process for students, educators and others looking for a shared language approach to onboarding AI. This interaction steps out each aspect for ease of understanding and creates a shared vocabulary, based on professional usage. This shared vocabulary will allow for growth and understanding in how to generate the best outcomes while using AI.

Below is the breakdown of the stepped interaction process along with easy to remember questions when using this process.

Human+AI Interaction : Question -> Frame -> Unpack -> Own -> Iterate

•  Question → “What am I really trying to understand?”
• Frame → “How am I setting this up?”
• Unpack → “What’s here and what is missing?”
• Own → “What do I think and why?”
• Iterate → “What will I change next?”

Human+AI Interaction : Question -> Frame -> Unpack -> Own -> Iterate

1) Question: 1st Step

Let’s look at the collective community voice of AI as a large tapestry woven from millions of fine threads. For questions asked on a surface level, with little context or definition, the AI looks along one thread, offering a small, linear answer. If the question is constructed with True Authorship* the AI will look at the entire tapestry of knowledge to create a deeper, more robust answer. This is the response we are looking for as this drives learning for students. 

Defining the problem is the most consequential decision in any AI interaction. Every AI responds to the frame it's given: text or media. An unconsidered prompt produces a vague or misdirected output — not because the AI failed but because the question didn't do its job. In machine learning terms this is sometimes called problem formulation bias — the shape of the question determines the shape of everything the system can find or generate. Students who understand this intuitively will outperform students who don't regardless of which type of AI is being used.

*True Authorship is the quality of the Question’s construction that determines the quality of the output more than any other single decision. The conditions must be built to give the best outcome. The student is in complete control of this step as the develop the discipline of distinguishing between surface questions and root questions: 

• Surface/ Closed questions: The ones that arrive first: obvious, already half-answered and singular one thread without depth. 
• Root/ Open questions: Multilayered and strategic ensuring the response is from multiple knowledge threads with communal insights

By writing more intentional prompts students are exercising pure human agency — deciding what matters and what they're genuinely curious about. This experience of agency before engagement with an AI is pedagogically important. It establishes, from the very first step, where the human leads and the machine does.

The educator models the process by showing students examples of phrasing evolving questions. The most powerful thing an educator does during the QUESTION phase is demonstrate that even experienced thinkers do not arrive at the right question immediately. The Question phase is complete when the student can say in one sentence beginning with “As a <user type> I would like to be able to <Whatever it is> so that I can <get what result?" Fill in the blanks with specific enough information that a stranger could understand if in conversation. 

Human+AI Interaction : Question -> Frame -> Unpack -> Own -> Iterate

2) Frame: 2nd step in process

A long standing saying in software development states “garbage in, garbage out”. It isn't that poor inputs produce poor outputs in a simple linear way. It's that every input decision — what context is provided, what constraints are set, what vocabulary is used, what perspective is centered, what is left unsaid — shapes the entire possibility space of what the system can return. This is true of large language models, image generators, and every other type of AI students will encounter. Every input into any AI adds a thread to the tapestry of the communal voice. 

• Prompt Engineering: The professional practice of constructing inputs for AI.
• The Question step shapes the outcome of the prompt. Defining a well structured question is one part, nailing down the parameters the AI should work within is another. A good prompt encompasses the question and defines parameters for the AI to operate within. 
• Context Windows: Every AI tool operates within a limited frame of what it can hold and consider at once. 
• Comparative Construction —Creating two or three radically different Frames for the same Question allows for observation of output divergence. If the question pertains to whether free lunch should be offered at a particular school. Frame the question to look at multiple states, similar schools within those states along with ethics of pros and cons to develop an overarching understanding prior to moving to Unpack or Own phases.
• Bias Dimension: Student bias most visibly enters the process here and is invisible to the student. Perspectives, assumptions, and blind spots brought into the construction of the Frame prompt will shape the output confirming what is already believed and known. 
• Common Failures:

The student frames too narrowly — providing so much constraint that the tool has no generative room to work: the output is thin. The answer is linear considering only one thread of the tapestry.

The student frames for the answer they want — constructing input designed to confirm a conclusion they've already reached. 

The student frames without boundaries — providing context and intent but never considering what they're leaving out and why. The output reflects gaps the student doesn't notice.

The student treats the first frame as final — submitting once and accepting whatever comes back when asking a different construction might have served the inquiry better

FRAME is complete when the student can articulate what they included, what they excluded, and why both of those choices serve their QUESTION?

Example:

 “How do schools reduce loneliness" outputs a generic list.

"I am designing a peer connection program for a middle school of 400 students in a suburban community where students report feeling invisible during unstructured time — what approaches have shown measurable impact in similar contexts?" will get something genuinely useful.

FRAME is the most technically rich step in the Human+ AI Interaction. A teacher who truly understands what's happening at Frame — the Bias Dimension, the Construction Authorship, the Comparative Frame — for prompt engineering within the context window will produce students who use AI in a fundamentally different way than students who were simply taught to write better prompts.

Human+AI Interaction : Question -> Frame -> Unpack -> Own -> Iterate

3) Unpack: 3rd Step in the Process

Every AI output is a compression of information, meaning it takes an immense amount of data distilling it into a bit size summary to answer the prompt. What arrives as an answer, a recommendation, an image, or a generated text is the end product of a vast chain of human decisions — what data was collected, who collected it, what was labeled and how, whose perspective was centered in the training material, what was excluded as noise or irrelevant. 

Model opacity: The outputs of AI systems do not come with explanations of how they were produced or what they cannot represent. An AI will produce a confident, fluent response that gives no indication of the limits of its knowledge or the gaps in its training.

Hallucination: The tendency of large language models to generate plausible-sounding information that is factually incorrect. The model is not aware it is producing false information, it is working with confines to best answer the prompt given. Add to this that the AI is created to give the best answer it can, this does mean it will create an answer not based on data or truth.

Representation Bias: The systematic underrepresentation of certain demographics, geographies, cultures, and experiences in training data, which produces outputs that reflect those absences without announcing them. As in Weapons of Math Destruction, O’Neil warns these models are opaque, redeploying inequities at scale without fanfare or correction."

The UNPACK phase of the process will approach AI outputs with a permanent and productive skepticism using three distinct critical capacities:

Source interrogation: The habit of asking where an output came from and what shaped it. Remember the AI creates the best answer to the prompt, it does not mean there is research/data supporting the answer.

Absence recognition: Human perception is oriented toward presence, we notice what exists far more readily than what doesn't..

Perspective auditing: The practice of systematically asking whose experience is centered in an output and whose is not. This centers around equity, diversity and inclusion, all responses need to be sourced and evaluated.

The UNPACK phase has a creative dimension that is easy to miss because the step feels primarily critical. Imagining what's missing requires genuine creative projection — students must inhabit perspectives that are not their own, imagine experiences that the output didn't represent, and construct in their minds a version of the output that would look different if different people had been in the training data.

Counter-framing: Imagine and articulate what the output would look like if it had been generated from a completely different perspective or dataset. A student who can say "if this output had been trained primarily on data from this community, it would probably look like this instead" is thinking creatively and critically simultaneously.

Gap mapping: Draw or list the spaces the output doesn't address, decide which of those gaps matters most for the specific purpose. This turns absence from a vague concern into a concrete and actionable map of what needs to be found elsewhere, reconstructed differently, or acknowledged as a limitation.

The UNPACK is where the human-machine relationship becomes most clearly asymmetrical. The AI produced something. It did so without awareness of what it missed, what was factless, and without concern for whose story it didn't tell. The human is capable of noticing absence, auditing perspective, and caring about who is missing. 

The UNPACK phase is complete: Can the student identify specifically what the output reveals and  have examined the output honestly?

Human+AI Interaction : Question -> Frame -> Unpack -> Own -> Iterate

4) OWN: 4th Step in the process

This is a decision making commitment to use the information coming from the AI. From this point forward the information gathered by the student and being used by the student has been vetted and they now take ownership of that work.

Human-in-the-loop: The principle that consequential decisions made with AI assistance require a human being to review, interpret, and take responsibility for the outcome. A system cannot be held responsible for an outcome. 

Automation Bias: The human tendency to defer to automated systems even when those systems are wrong, incomplete, or operating outside their competence.

Reasoning transparency: The ability to articulate not just what the output was but why, including what evidence, which was weighed most heavily, what acknowledged uncertainty, and what was prioritized.

Students stand with their work, name it explicitly, and are willing to say: I decided this, because of this, knowing I might be wrong about this.

What it looks like in practice: A student has been using AI to research whether their school should shift to a four-day week. They questioned carefully, framed deliberately, unpacked thoroughly. The output presents evidence on both sides. Some studies suggest academic performance is unaffected. Others suggest vulnerable students fall behind. Community impact data is mixed. The output doesn't resolve the question, it clarifies it. The student must decide what they actually think, not what the data suggests, not what the most recent study found, not what the most authoritative source recommended but what is genuinely believed to be the right direction and why.            

The ethical core: To own a decision is to accept responsibility for its consequences.

Human+AI Interaction : Question -> Frame -> Unpack -> Own -> Iterate

5) Iterate: 5th step in the process

The Iteration process is the uniquely human ability to grow, adapt, and change. Within the AI tapestry, iteration becomes an opportunity for students to decide if the response uncovered during this Human+ AI interaction is fact or fiction. 

• terate offers students the chance to retrace their research thread from questions, frame, unpack and own. This sharpens thinking and grows skills ensuring students adapt prompts for richer responses from AI. 

CAPSTONE

Human+AI Interaction fosters Critical Creative Thinking by blending analytical rigor with the creative process. This equips students to question assumptions and spot biases, dissect AI outputs for gaps, and weigh evidence for unexpected connections and alternatives. Students learn to partner with AI as a community of collaborators, not a singular tool, building an understanding of both the boundaries within large language models and the vocabulary to navigate them efficiently. 

Human+AI interaction is a catalyst for deeper thinking through refined creative agency, curiosity-driven growth, and strengthening cognitive flexibility. This cultivates a life of joy in learning and discovery, where curiosity leads through evolving AI partnerships.