"Where is this coming from?" Uncovering Trustworthiness Ideals in AI-powered Peripartum Information Seeking
Through focus groups with birthing people, clinicians, and health workers, this paper argues that in high-stakes peripartum care, AI systems must prioritize inspectable trustworthiness over asserted authority by incorporating transparency, recourse, and ecosystem-aware integration to address historical inequities and misinformation.
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 pregnant or just had a baby. You have a question: "Is it safe to drink coffee?" or "Why does my back hurt?" In the past, you might have called your doctor. But today, many people turn to the internet, social media, or AI chatbots for quick answers.
This paper is like a group of detectives (researchers) who asked three different groups of people—parents-to-be, doctors, and support workers (like doulas)—what they really need when they are looking for health answers online. They wanted to know: If we build a super-smart AI to answer these questions, how do we make sure people actually trust it?
Here is the story of what they found, explained simply.
The Big Problem: Trust isn't just about being "Right"
The researchers found that for parents, getting the "correct" medical fact isn't enough. It's like a GPS that tells you the fastest route but ignores that the road is closed due to a protest, or that you are driving a truck that can't fit under a bridge.
In the world of pregnancy and birth, history matters. Because of past unfair treatment (especially toward Black and Brown communities), many people don't automatically trust hospitals or doctors. They might feel rushed, unheard, or treated differently. So, when they ask an AI a question, they aren't just asking for data; they are asking for reassurance and safety.
The Four Rules for a Trustworthy AI
The researchers discovered four main "rules" (themes) that any AI tool needs to follow to be trusted in this sensitive area. Think of these as the ingredients for a good recipe:
1. The "It's Not Just You" Rule (Social Sensemaking)
The Metaphor: Imagine you are lost in a forest. A map (the AI) tells you the direction. But what you really need is a friend to say, "I was lost here too, and I found a path. You're okay."
The Finding: People don't just want facts; they want to know that others like them have had similar experiences. They want to hear stories from peers.
The Solution: The AI shouldn't just be a robot giving a textbook answer. It should acknowledge feelings, normalize worries, and maybe even show stories from other parents (if they are safe and verified) so the user feels less alone.
2. The "Many Ways to Check" Rule (Verification Pluralism)
The Metaphor: Imagine you are buying a used car. One person checks the engine, another checks the tires, and a third asks a mechanic. There isn't just one way to know if the car is good.
The Finding: Not everyone trusts the same things. Some people trust their doctor 100%. Others think doctors are too slow or outdated. Some trust community groups.
The Solution: The AI shouldn't force one way of checking. It should offer different tools: "Here is what the doctor says," "Here is what the research says," and "Here is a checklist you can use to ask your doctor better questions." It lets the user choose how they want to verify the info.
3. The "Show Your Work" Rule (Inspectable Governance)
The Metaphor: Imagine a magician pulls a rabbit out of a hat. If you ask, "Where did the rabbit come from?" and the magician says, "Trust me, it's magic," you won't believe them. But if they open the hat and show you the secret compartment, you trust the trick.
The Finding: People are suspicious of AI. They ask: "Who wrote this? When was it written? What if it's wrong? Who do I call if it hurts me?"
The Solution: The AI must be transparent. It needs to show its "ID card" (source), its "expiration date" (when it was last updated), and a clear path for what to do if it makes a mistake. It can't just say "I am trustworthy"; it has to prove it.
4. The "Don't Add to the Burden" Rule (Ecosystem Complementarity)
The Metaphor: Imagine you are already carrying a heavy backpack full of groceries, kids, and work papers. If someone hands you a new heavy box and says, "Here, carry this too," you will drop everything. But if they help you organize the groceries so the bag is lighter, you will love them.
The Finding: Doctors and parents are already overwhelmed. They don't want a new tool that creates more work or confuses the system.
The Solution: The AI shouldn't try to replace the doctor. It should act like a helpful assistant that organizes the questions so the user can talk to the doctor more efficiently. It should know when to say, "I don't know, please ask your doctor," and help the user get there safely.
The Main Takeaway
The paper concludes that trust cannot be claimed; it must be shown.
If we build an AI for pregnancy that is just "smart" but doesn't understand history, doesn't show its sources, and doesn't fit into the real lives of parents and doctors, it will fail. The researchers suggest that for AI to be truly helpful, it needs to be transparent, flexible, and human-centered, acknowledging that for many, the journey to health is also a journey through history and emotion.
In short: Don't just build a robot that knows the facts. Build a tool that understands the people asking the questions.
Drowning in papers in your field?
Get daily digests of the most novel papers matching your research keywords — with technical summaries, in your language.