Part 10 of 12 in the Try It Yourself series — practical AI guidance for parents at blog.parentsguidetoai.ca


When a computer gives you an answer, it can feel objective. Neutral. Like a calculation. It isn't.

AI tools produce the most statistically common response based on their training data — not the most accurate one, not the most balanced one, and not the one that best reflects your family's perspective or values.1 If certain voices, cultures, or viewpoints were underrepresented in the data the AI learned from, those voices are underrepresented in what it produces.

This isn't a flaw that will be fixed in the next update. It's an inherent characteristic of how current AI systems work. Understanding it is one of the most important media literacy skills of the next decade — and it's something parents can model for their children starting today.


Where AI bias comes from

The parent's guide explains it this way: AI systems learn from large amounts of text and other data from the internet. That data reflects the world as it was written about — which means it reflects the biases, assumptions, and gaps of the people who wrote it.

If most of the text describing a particular profession, culture, or history was written from a specific perspective, the AI's understanding of that topic reflects that perspective. Not because the AI was programmed to be biased, but because its understanding of the world is built from biased source material.

Some types of bias appear more frequently than others:

  • Cultural and geographic bias: AI tends to give answers that reflect dominant English-language, Western perspectives. Local context — BC-specific information, Indigenous perspectives, immigrant experiences, regional differences — is often underrepresented or incorrect.2
  • Historical bias: When describing historical events, AI reflects the framing most common in its training data, which may not be the most accurate or complete framing.
  • Representation bias: Visual and descriptive AI tools often default to particular demographics when no specification is given — default assumptions about what a "doctor" or a "student" looks like, for example.
  • Confirmation patterns: AI tends to produce confident-sounding answers. It doesn't naturally say "there are multiple legitimate perspectives on this" unless prompted to do so.

What biased AI output looks like

Bias in AI output is usually subtle. It doesn't look like an obvious error. It looks like a perspective that has been normalized so thoroughly that it reads as neutral.

Some things to watch for:

  • The "universal" perspective: An answer that treats one cultural practice as the default and others as variations or exceptions.
  • Missing representation: A description that doesn't include perspectives your family would consider obvious or important.
  • Oversimplification of complex topics: Historical events, political issues, or social questions described as if there were one clear answer.
  • Slightly off for local context: Information about Canada, BC, or Vancouver that's just a bit wrong — policies, institutions, timelines — because most of the AI's training data was American.

The parent's guide includes a list of questions from Section 11 that apply directly here:

Whose perspective does this reflect? Could this be unfair to certain groups? Would a different source give a different answer?

These questions are useful for children. They're also useful for parents who are reading AI-generated content and want to think critically about it before accepting it.


A hands-on exercise

This exercise takes about ten minutes and gives you a concrete example to discuss with your child.

  1. Open any AI tool — ChatGPT, Copilot, or Claude.
  2. Ask it about a topic that's meaningful to your family — your cultural background, your community, a historical event relevant to your heritage, or a local issue in BC.
  3. Read the response carefully. Does it reflect your experience and understanding? Where does it feel right? Where does it feel off or incomplete?
  4. Then ask the AI: "What perspectives or voices might be missing from that answer?"

The second prompt often produces something more nuanced — and the contrast between the two responses is a useful lesson in itself.


Why this matters for raising children in a diverse world

Children who grow up using AI tools are going to form some of their understanding of the world through those tools. If they never learn to question the perspective embedded in AI answers, they may absorb a particular worldview without realizing it — and without realizing there are others.

This is the same skill we teach when we talk about media literacy and critical thinking. AI is just a new surface for an old challenge.

The parent's guide frames it this way: the goal isn't to make children suspicious of AI, but to help them understand that AI reflects the world it was trained on — and that world had gaps, biases, and dominant perspectives that didn't speak for everyone.

A child who has been taught to ask "whose perspective is this?" when reading a book, watching a film, or looking at a news article already has the skill they need for AI. The application is new. The habit of mind is the same.


The short version:

  • AI produces statistically common answers, not neutral ones — it reflects the biases in its training data.
  • Common types: cultural/geographic bias (often Western-centric), representation bias, local context gaps.
  • Three questions from the guide: Whose perspective? Could it be unfair? Would another source differ?
  • The hands-on exercise: ask AI about something meaningful to your family, then ask what perspectives are missing.
  • This is media literacy for the AI era — the same habit of mind, applied to a new surface.

We're nearly at the end of the series. The next post brings everything together: how to write a family AI agreement that your children will actually follow.

Try it today: Ask an AI tool about a topic that's personally meaningful to your family or community. Then ask it what perspectives might be missing. Read Section 11 of the parent's guide for the full context on AI and critical thinking.

Next up: Post 11 — Writing Your Family AI Agreement: A Step-by-Step Guide


Notes


  1. Bender, Emily M., Timnit Gebru, Angelina McMillan-Major, and Shmargaret Shmitchell. "On the Dangers of Stochastic Parrots: Can Language Models Be Too Big?" FAccT '21: Proceedings of the 2021 ACM Conference on Fairness, Accountability, and Transparency, March 2021. https://dl.acm.org/doi/10.1145/3442188.3445922 

  2. MediaSmarts. "Resources for Parents — AI and Algorithms." MediaSmarts, 2024. https://mediasmarts.ca/digital-media-literacy/general-information/ai-and-algorithms/resources-parents-ai-and-algorithms