.svg)
Building Trust in AI Development: Compassion
Compassion in AI means designing for the person who doesn't fit the average case. Here's why.
.png){kind=avatar}
Most AI systems are designed for the ideal user. The person who is calm, capable, and using the system under normal conditions.
That person is a minority.
The reality of human-centred AI design in healthcare, retail, and public spaces is that most interactions happen at the edges. The patient who is anxious and in pain. The elderly visitor unfamiliar with touchscreen interfaces. The person navigating a shopping centre with a child in tow and a phone that has run out of battery. The non-native English speaker using a government kiosk for the first time.
Compassion in AI development is the recognition that these are not edge cases. They are the main case.
What Compassion Means in AI Development
In human relationships, compassion means active listening, genuine presence, and care for others' experience, not just their outcomes. It is the difference between a leader who listens and one who merely waits to speak.
In AI development, compassion is the design commitment to understand who is actually using the system and what they experience when they use it.
That means user research that goes beyond the average. Testing with people who have low digital literacy, physical limitations, language barriers, or who are under stress. Making design decisions that serve the full range of people who will encounter the system, not just the ones who will have the easiest time.
Why the "Average User" Is a Design Fiction
The concept of an average user is useful for initial design frameworks. It becomes harmful when it stays the default assumption through delivery.
In a hospital wayfinding system, the average user is not a healthy, tech-savvy adult visiting a familiar wing of a building they know well. The average user is someone who received difficult news yesterday, slept badly, and is now trying to find a department they have never been to before.
Design for that person. The healthy, tech-savvy adult will find the system easy regardless. The reverse is not true.
Compassion as a Trust Mechanism
People trust systems that feel designed for them. The trust signal is usually subtle: a font size that is readable without squinting, a button that responds predictably, a tone that stays calm when the user makes an error, a system that offers help rather than a dead end when something goes wrong.
These are not small decisions. They are the moments that determine whether someone trusts the system enough to keep using it, or turns away.
In Justin Dean's research on trust in artificial intelligence, compassion consistently emerges as one of the principles most directly linked to user adoption. People adopt AI they trust. They trust AI that treats them as capable, considers their situation, and responds appropriately to their actual experience, not an assumed one.
What This Means in Practice for AI Developers
Ask these questions at every design stage.
Who is this system for, specifically? Not in the abstract. In the particular. Name the users. Name their situations.
What are they feeling when they use this? Stress, confusion, urgency, and unfamiliarity are not edge conditions. They are conditions the system needs to be designed for.
What happens when something goes wrong? Does the system respond by offering a clear path forward? Or does it produce an error and stop?
Compassion is not a soft principle. It is a design standard that determines whether real people, in real situations, can use and trust the systems you build.
