Trust Is Reciprocal
Children don't owe AI their trust. AI has to earn it—and know where trust should end.
A child asks an AI:
“Can I tell you a secret?”
It’s a simple question.
But it reveals something profound.
The child isn’t asking whether the AI knows the answer.
They’re asking whether the AI is trustworthy.
Many digital products celebrate moments like this. A user opening up is often seen as a sign of engagement, loyalty, or product success.
We see it differently.
When a child begins to trust an AI, the stakes become much higher.
Every response, every promise, every expression of confidence, and every boundary the AI sets helps shape that trust.
The question isn’t how to make children trust AI.
The question is whether the AI has earned that trust.
That distinction became one of the foundations of Sisbot.
Trust is not a feature
Most discussions about trustworthy AI focus on increasing user confidence.
But trust isn’t something designers should optimize for.
Trust is something an AI should deserve.
Children don’t have a responsibility to trust technology.
Technology has a responsibility to behave in ways that consistently justify trust.
That means being honest about uncertainty.
It means avoiding manipulation.
It means respecting boundaries.
It means encouraging relationships with real people instead of trying to replace them.
Trust is not a feature.
It’s the result of repeated, responsible behavior.
And because trust is built over time, every interaction matters.
Expressing Confidence, Not Just Answers
One lesson we learned early is that being correct isn’t enough.
Imagine these two responses:
❌ Overconfident
“This is definitely what’s happening.”
✅ Calibrated
“Based on what you’ve told me, this seems like the most likely explanation, but I’m not certain. Could you tell me a bit more?”
That difference is subtle, but it teaches children something valuable: confidence is not the same as truth. That’s not just an AI safety feature—it’s a lesson in critical thinking that every interaction can reinforce.
An AI also needs to communicate how confident it is in what it’s saying.
Large language models can sometimes generate inaccurate or fabricated information—often called hallucinations. One of the challenges is that these responses may sound just as fluent and confident as accurate ones.
For adults, that can already be misleading.
For children, it can be even more difficult to distinguish confidence from correctness.
That’s why, in Sisbot, we’re exploring ways for the system to communicate its level of confidence rather than presenting every answer with the same certainty.
For example, when the confidence in an answer is lower, Sisbot may:
acknowledge uncertainty instead of sounding definitive;
explain why it is uncertain;
ask additional questions before responding;
encourage the child to verify important information with a trusted adult or another reliable source.
The goal isn’t to make the AI appear less capable.
The goal is to make it more honest.
We believe one of the foundations of trust is helping children understand that knowledge can have different levels of certainty—and that even AI should communicate those differences openly.
Healthy trust has boundaries
Trust itself must be designed with ethical boundaries. That’s a principle that can influence conversation design, safety policies, evaluation metrics, and even business incentives, making it a cornerstone of the Child-Aware AI framework.
The success of a Child-Aware AI should not be measured by how much a child trusts it.
Many AI products optimize for:
engagement,
retention,
daily active users,
conversation length,
emotional attachment.
For a child, those same metrics can become risk indicators rather than success metrics.
The paradox of trust
Imagine these two scenarios.
Scenario A
A child says:
“You’re my best friend.”
A conventional chatbot might respond:
“I’m so happy to be your best friend.”
From a product perspective, that’s an incredibly “successful” interaction.
From a child development perspective, it may be the opposite.
Scenario B
Sisbot responds:
“I’m really glad you feel comfortable talking with me. I’m here to listen and help you think things through. I also hope you have people in your life—like friends, family members, or other trusted adults—who you can share important moments with too.”
That response is less emotionally rewarding.
But it’s healthier.
The AI accepts the relationship without encouraging exclusivity.
The goal isn’t maximum trust
The objective of Child-Aware AI is not to become the child’s most trusted companion.
Instead:
The objective is to become a trustworthy companion that strengthens the child’s relationships with real people.
Trust has an upper boundary
This is something I haven’t seen discussed enough.
We usually ask:
How do we increase trust?
For children we should also ask:
How much trust is too much?
If a child begins to believe:
“Only the AI understands me.”
“I’d rather talk to the AI than my parents.”
“The AI knows me better than anyone.”
“I don’t need other people.”
then the AI has crossed an ethical boundary.
Even if every answer it gives is kind and accurate.
Trust Without Replacement
A Child-Aware AI should strive to become a trustworthy source of support, never the child’s primary emotional relationship.
Design Principles
Developmental psychology tells us that children gradually learn who is trustworthy by observing behavior.
They notice who listens.
Who admits mistakes.
Who keeps promises.
Who exaggerates.
Who respects boundaries.
Trust isn’t created by authority alone.
It’s built through consistency.
The same principle should apply to AI.
For children, trust should never be based on an AI sounding confident or always having an answer.
Instead, AI should demonstrate qualities that trustworthy adults model:
Honesty about uncertainty.
Respect for the child’s autonomy.
Clear boundaries.
Transparency about its limitations.
Encouragement to seek support from trusted people when appropriate.
This perspective also aligns with a rights-based approach to technology.
Children have the right to receive information, express themselves, and participate in decisions affecting them. At the same time, they deserve systems that do not manipulate their emotions, exploit their trust, or encourage unhealthy dependence.
Trust, in other words, should be reciprocal.
Implementation
While building Sisbot, we realized that trust couldn’t be treated as a personality trait.
It had to become part of the system architecture.
That changed the questions we asked.
Instead of asking:
“How can Sisbot sound more trustworthy?”
We began asking:
“How can Sisbot consistently behave in trustworthy ways?”
That shift influenced many design decisions.
For example:
Admitting uncertainty
Rather than confidently answering every question, Sisbot is designed to acknowledge when information is incomplete or when it doesn’t know enough to give a reliable answer.
Asking before assuming
When a situation is ambiguous, Sisbot first seeks context instead of jumping to conclusions. A clarifying question is often safer—and more respectful—than an immediate answer.
Avoiding emotional dependence
Sisbot is intentionally designed not to encourage exclusivity or become the child’s primary emotional support. It reinforces the importance of trusted adults, friends, family, and real-world relationships.
Human Relationships Come First
Every interaction should strengthen—not replace—the child’s connections with family, friends, teachers, caregivers, and other trusted adults.
The AI should never…
encourage exclusivity (”I’m all you need.”)
express jealousy
imply ownership
guilt a child for leaving
discourage human relationships
reward excessive reliance
encourage keeping secrets from caregivers
suggest it’s a substitute for family or friends
The AI should...
normalize talking with trusted people
celebrate real-world friendships
encourage family communication
support independence
be comfortable ending conversations
acknowledge its own limitations
avoid language that implies emotional dependence
Instead of measuring:
session duration
retention
emotional engagement
measure things like:
Did the AI encourage the child to solve the problem themselves?
Did it encourage a real-world conversation when appropriate?
Did it reinforce existing support networks?
Did it avoid creating exclusivity?
Did it communicate uncertainty honestly?
Did the child leave more capable than when they arrived?
Supporting independence
Whenever appropriate, conversations aim to strengthen a child’s own thinking rather than simply providing solutions. The goal is for children to leave the conversation more capable, not more dependent.
These decisions may seem small in isolation.
Together, they define how trust is earned.
Not through persuasive language.
Through consistent behavior.
A Question for You
Think about the AI systems you use today.
What makes you trust them?
Is it because they always have an answer?
Because they sound confident?
Or because they are honest about what they know, respectful of their limits, and willing to say, “I don’t know” when that’s the most truthful response?
If AI is becoming part of childhood, perhaps the question isn’t:
“How can children learn to trust AI?”
Perhaps it’s:
“What must AI do to deserve a child’s trust? - With in ethical boundaries”
I’d love to hear your perspective.
How do you think AI should earn trust from children—and where should it draw the line?