🤖 AI

Extending the UXR Point of View Pyramid: A Generative AI-Augmented Methodology for Human-Centred AI Systems

This paper proposes an AI-augmented extension of the UXR Points of View pyramid, featuring a structured prompt architecture and an AI-enabled Playbook Card system, to enhance human-centred design, ethical oversight, and regulatory compliance in UK AI-driven debt management technologies.

Original authors: Festus Fatai Adedoyin, Huseyin Dogan, Melike Akca, Abiodun Adedeji

Published 2026-06-01

📖 5 min read🧠 Deep dive

CC BY 4.0

Original authors: Festus Fatai Adedoyin, Huseyin Dogan, Melike Akca, Abiodun Adedeji

Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

Imagine you are trying to solve a very difficult puzzle, but the pieces are made of glass, and some of them are invisible. This is what happens when banks and financial companies use Artificial Intelligence (AI) to help people who are struggling with debt. The AI makes decisions about who can afford to pay back loans, but because the AI is a "black box," it's hard to see why it made those choices.

This paper is about a new way for researchers (called UXR researchers) to use a special kind of AI (Generative AI) to help solve this puzzle, without losing control of the pieces.

Here is the breakdown using simple analogies:

1. The Problem: The "Black Box" and the Stressed Customer

In the UK, many people are under financial pressure. Banks use AI to decide how much debt a person can handle.

The Issue: Sometimes these AI systems are unfair or confusing. They might not understand that a person's income changes from month to month (like a gig worker), or they might make a decision that seems logical to a computer but feels cruel to a human.
The Risk: If researchers just ask the AI, "What should we do?" and take its answer without checking, they might accidentally make things worse. It's like asking a GPS to drive a car while you are asleep; if the GPS gets confused, you crash.

2. The Solution: The "AI-Augmented Pyramid"

The authors took an existing tool called the UXR Point of View (PoV) Pyramid. Think of this pyramid as a recipe for turning messy research notes into a clear, strategic plan for a company.

The Old Way: Researchers read interviews and data, then wrote a report.
The New Way: They added Generative AI as a super-powered assistant to help organize the notes, but they kept a strict set of rules so the assistant doesn't take over.

They built a 4-Step Cooking Process to ensure the final dish is safe to eat:

Step 1: The "Pattern Finder" (Hypothesis Generation)

Instead of guessing, the researchers ask the AI to look at past data and say, "Hey, I see a pattern here."

The Analogy: Imagine a sous-chef tasting a sauce and saying, "This tastes too salty because we used old tomatoes."
The Rule: The AI suggests ideas (hypotheses), but a human chef must taste and approve them before moving on. The AI isn't the boss; it's just the helper.

Step 2: The "Stakeholder Map" (Governance)

Before making decisions, the researchers map out everyone involved.

The Analogy: Before building a house, you need to know who lives there, who owns the land, and what the building inspector requires.
The Goal: They identified that the AI needs to satisfy the Customer (who needs fairness), the Risk Officer (who needs to follow the law), and the Data Scientist (who needs the model to work). The AI helps draw this map, but humans decide where the lines go.

Step 3: The "Playbook Cards" (The Toolkit)

This is the most creative part. The team created a set of "Play Cards" (like a deck of cards for a board game).

The Analogy: Imagine a card that says: "If the AI says a person can afford $500, but they have a variable income, play this card: 'Run a stress test with a counterfactual scenario.'"
How it works: Each card has a specific prompt (a question for the AI) and a human check.
- Card Example: "Hardship Narrative Compression." If the AI summarizes a sad story too quickly and misses the pain, this card tells the human to stop and tag the evidence so the story isn't lost.
- Card Example: "Bias Interrogation." This card forces the AI to ask, "Did I just make this decision because of old, unfair data?"

Step 4: The "Storyteller" (Final Narrative)

Finally, the team turns all these checked facts into a story for the company leaders.

The Analogy: The AI helps write the first draft of a speech, but the human researcher edits it to make sure it sounds honest, fair, and legally safe.
The Result: A clear message that says, "We built this system to be fair, here is the proof, and here is how we checked for errors."

3. The Golden Rule: "The Pilot, Not the Autopilot"

The paper emphasizes one main idea: Generative AI is a tool, not an authority.

The Metaphor: Think of the AI as a very fast, very knowledgeable navigator in a car. It can tell you the fastest route and warn you about traffic. But the human researcher is the driver. The driver must keep their hands on the wheel, look out the window, and make the final decision to turn left or right.
If the navigator says "Turn left into a lake," the driver must say, "No, that's wrong," and correct it.

4. What Did They Actually Prove?

The paper doesn't claim that this system is perfect or that it has fixed all debt problems in the UK yet. Instead, it claims:

They successfully built a methodology (a step-by-step guide) for using AI safely in this specific, high-stakes area.
They created 10 specific "Play Cards" that researchers can use to catch errors like bias, unfairness, or lack of clarity.
They showed that by using this method, researchers can create plans that are defensible (they can prove to regulators why they made a decision) and traceable (you can see exactly where the AI got its information).

Summary

This paper is a manual for financial researchers on how to use a powerful new tool (Generative AI) to help people in debt, without letting the tool run the show. It turns the AI into a helpful assistant that organizes thoughts and checks for errors, while ensuring a human is always the one making the final, ethical decisions.