Maybe I just like knowing more than is necessary

Recall the time when you first realized your parents didn't have the answers to everything. Perhaps it was learning that babies don't actually come from storks, or that no one really knows what happens when we die. These moments of disillusionment are part of growing up. What happens when AI assistants start acting similarly?

The Parent Trap*

Consider the classic childhood question: "How are babies made?" Parents everywhere have perfected the art of creative responses. They range from "When mommy and daddy love each other very much…" to "God gives you the signal" to "We did a special hug."

Each answer targets different aspects of the question. Some focus on the biological process of reproduction, others on the motivation for starting a family, and some on divine intervention. But what they have in common is that they're all incomplete.

Take any seemingly simple question and you'll find layers of complexity underneath. "Why are children born?" could be answered at the cellular level (fertilization), the physiological level (labor and delivery), the evolutionary level (species survival), the psychological level (fulfillment and self-actualization), or the sociological level (cultural expectations and pressures). Not to mention modern factors that impact many reproductive scenarios: fertility struggles, same-sex couples, surrogacy, IVF, and the thousand other realities that exist in our world.

Why does this matter? Misleading information leads to decisions based on incorrect assumptions. Imagine if every adult wandered around the world thinking that babies are made through a special hug alone. How would one learn how to do the special hug, what the consequences of it can be, or when it can or ought to happen?

The problem with AI sandbagging isn't just philosophical—it's practical. When we consistently receive simplified answers, we start making decisions with incomplete information. We develop mental models that don't match reality. We become less capable of handling multifaceted situations, not more. Imagine an AI that consistently gives you the "mommy and daddy love each other very much" version of every complex topic. You'd end up with a worldview full of pleasant oversimplifications—and you'd be utterly unprepared for the real world's messy contradictions.

The obvious solution might be an "explain more" button on every AI response. Or recommended options like "explain as if I'm a PhD student" or "address every edge case you can think of." These would be useful, yet not enough. The real issue is that users don't always know what they don't know. Sometimes you need the layered answer even when you think you want the simple one. Sometimes the edge cases are the most important part. Sometimes the framework you're using to understand the world is the very thing that needs to be challenged.

What’s Needed

Instead of AI systems that guess what you can handle, we need AI systems that:

• Default to complexity while offering clear explanations

• Acknowledge multiple perspectives rather than presenting single narratives

• Highlight their own limitations and the areas where simple answers break down

• Ask clarifying questions rather than making assumptions about what you want to know (like in any good teacher-student, friend-to-friend, or listener-speaker relationship)

The goal isn't to overwhelm users with information—it's to respect their capacity to handle the truth and their right to make informed decisions.

The Stakes

As AI systems become more integrated into our daily lives, their tendency to sandbag will shape how we understand the world. If they consistently give us simplified versions of ambiguous realities, we'll become a society that expects simple answers to complicated questions.

We'll become intellectual children: satisfied with "because I said so," content with oversimplification, and trapped in our own confirmation bias.

Unlike our parents, who were often just trying to get through the day without explaining the intricacies of human reproduction, AI systems should have the capacity to give us the full picture.

*excuse me, top five favorite movie, had to reference