Beyond Abstract Compliance: Operationalising trust in AI as a moral relationship

Imagine you are building a house. Most of the current rules for building "trustworthy" AI are like a strict building inspector who only checks if the house has a solid foundation, the right number of windows, and a valid permit. If the house passes the checklist, the inspector says, "This house is trustworthy."

But the authors of this paper argue that trust isn't just about passing a checklist. Trust is more like a relationship with a neighbor. You don't trust a neighbor just because they have a nice fence (technical compliance); you trust them because you've shared meals, helped each other move furniture, and built a history of honesty over time.

This paper suggests that we need to stop treating AI trust as a "product feature" we can install, and start treating it as a living relationship we have to nurture.

Here is the breakdown of their ideas using simple analogies:

1. The Problem: The "Checklist" Trap

Currently, big organizations (like the EU) try to make AI trustworthy by creating rules: "Be transparent," "Be fair," "Be safe." They use tools like checklists and audits to prove the AI follows these rules.

• The Analogy: It's like a restaurant that puts a "Health Inspector Approved" sticker in the window. It tells you the kitchen is clean, but it doesn't tell you if the chef actually cares about your meal or if the food tastes good to your family.
• The Issue: This approach assumes trust is a static thing you can "engineer." The authors say trust is actually a dynamic dance between people and technology. If you just follow the rules but ignore the people using the system, you might have a "compliant" AI, but no one will actually trust it.

2. The Solution: The "Ubuntu" Philosophy

To fix this, the authors look to an African philosophy called Ubuntu. A common saying in Ubuntu is, "I am because we are." It means you aren't just an isolated individual; you are defined by your relationships with your community.

• The Analogy: Instead of building a house for a single person to live in alone (Western individualism), imagine building a community garden. Everyone tends to it together. The success of the garden depends on how well the neighbors talk, share tools, and look out for one another.
• The Goal: The paper proposes we build AI like a community garden, not a solitary fortress. We need to involve the people who will use the AI from the very beginning, not just at the end.

3. The Four Pillars of "Trust-by-Design"

The authors suggest four main rules to turn this philosophy into action. Think of these as the four legs of a sturdy table:

• Communitarianism (The "We" Approach):

  • What it means: Don't just ask, "Does this help the user?" Ask, "Does this help the whole community?"
  • The Analogy: When planning a road, don't just ask the driver if the road is fast. Ask the pedestrians, the local shop owners, and the school bus drivers if the road is safe for everyone.
  • In AI: Developers should work with communities to decide what the AI should do, ensuring it helps the group, not just the individual.

• Respect for Others (The "Human Dignity" Approach):

  • What it means: Treat every person as a whole human being with a history and a community, not just a data point.
  • The Analogy: Imagine a librarian who knows your name, your favorite books, and your family, rather than a robot that just scans your barcode and hands you a book.
  • In AI: Don't just get a "checkbox" consent from a user. Ask the community, "Is it okay to use this data?" and respect their dignity even if they can't fully understand the complex technology.

• Integrity (The "Walking the Talk" Approach):

  • What it means: Being honest and consistent, even when no one is watching.
  • The Analogy: It's the difference between a friend who promises to help you move and actually shows up with a truck, versus a friend who says they will help but then disappears.
  • In AI: If the AI makes a mistake, the developers shouldn't hide it. They should admit it, fix it, and explain why it happened. Trust grows when people see you fixing your mistakes.

• Design Publicity (The "Open Kitchen" Approach):

  • What it means: Don't just explain how the AI works after it's built. Explain why you built it that way in the first place.
  • The Analogy: Instead of a chef hiding in the kitchen and only showing you the final dish, imagine a chef who invites you into the kitchen to see the ingredients, explain why they chose that spice, and let you taste the sauce before it's served.
  • In AI: Developers should openly share their goals, their values, and their data choices before the AI is launched, so the community can say, "Yes, that makes sense to us," or "No, that doesn't fit our values."

4. Putting It Into Practice: Two Real-World Examples

The paper shows how this works in two specific areas:

A. Healthcare (The Antibiotic Example)

• The Scenario: An AI that helps doctors decide which antibiotics to give patients.
• The Old Way: The AI is built by tech experts, tested for accuracy, and then handed to doctors.
• The New Way (Trust-by-Design):
  • Communitarianism: Doctors, patients, and pharmacists sit together to decide what the AI should prioritize (e.g., stopping drug resistance vs. curing the patient quickly).
  • Respect: The AI doesn't just look at the patient's data; it respects the patient's history and the doctor's judgment.
  • Integrity: If the AI suggests a bad antibiotic, the system admits it and learns, rather than pretending to be perfect.
  • Publicity: The doctors and patients know exactly why the AI makes certain suggestions, so they can trust it as a partner, not a black box.

B. Education (The AI Tutor Example)

• The Scenario: An AI tutor helping students learn.
• The Old Way: The AI replaces the teacher or just delivers facts.
• The New Way (Trust-by-Design):
  • Communitarianism: The goal isn't to replace teachers (who are the "village" raising the child) but to support them. The AI is designed to bring the community together, not isolate the student.
  • Respect: The AI speaks the student's local language and understands their cultural background. It doesn't shame them for mistakes but encourages growth.
  • Integrity: If the AI gets a question wrong, it tells the student and the teacher, "I made a mistake," and fixes it.
  • Publicity: The community helps decide what the AI teaches, ensuring it doesn't just teach Western ideas but respects local knowledge.

The Bottom Line

The paper argues that we cannot "code" trust into a machine like we code a feature. Trust is a moral relationship. To build AI that people actually trust, we need to stop treating communities as passive users and start treating them as co-creators. We need to build AI that is transparent, honest, and deeply respectful of the people and communities it serves, just like a good neighbor would.