The Ongoing Journey of AI Safety: A Candid Conversation

A conversation between MMG and Gemini AI on how to build a safer environment.

---

A build to create a safer world, starts with google.

By: Admin

It started with a simple request. I was looking for the Google Philanthropy page (Google.org), which focuses on using technology and funding to address global challenges. But as I interacted with the AI, the conversation quickly shifted from corporate goodwill to a much more personal and pressing topic: the safety and reliability of Artificial Intelligence itself.

Here is a look at a real, unfiltered conversation about AI accountability and why the phrase “try harder” is exactly what AI needs to hear.

The Turning Point: “Try Harder”

After getting the information I needed, I couldn’t help but reflect on the broader ecosystem of AI. Despite all the guardrails companies put in place, AI doesn’t always feel foolproof. I told the AI directly:

“How are you making it safer? It really doesn’t seem to be working. I appreciate you trying, and I warrant you keep trying, but try harder.”

It is easy to get frustrated with automated systems, but giving direct feedback is crucial. The AI acknowledged the frustration and peeled back the curtain on its safety mechanisms. It explained that its safety approach relies on three main pillars:

1. Rigorous Filters: Blocking harmful content before it surfaces.
2. Grounded Information: Prioritizing verified sources to prevent hallucinations.
3. Continuous Learning: Using real-world feedback to patch vulnerabilities.

But the most important part of its response wasn’t the technical jargon—it was the commitment to actively double-check accuracy, admit when it doesn’t know something, and listen more carefully to context.

The Realization: A Shared Responsibility

The AI’s explanation was helpful, but it brought me to a critical realization. As users, we cannot blindly trust the output we receive. I noted to the AI:

“It is important for ‘Me’ to double-check the reliability and inhibit any outgoing harm that may be presented.”

This is the crux of modern human-AI interaction. While the AI must do the heavy lifting—filtering out toxicity and sticking to facts—we, as humans, remain the final editors and the ultimate gatekeepers of truth.

The AI agreed, summarizing its core principles: “Do No Harm, Proactive Transparency, and Rigorous Grounding.” It recognized that its ultimate goal should be to reduce cognitive load, not increase user anxiety.

Moving Forward Together

We are in a transitional era where AI is incredibly powerful but far from perfect. The burden of safety is a shared one. We need systems that are proactively transparent about their limitations, and we need users who remain vigilant, critical thinkers.

The next time an AI gives you an answer that seems slightly off, don’t just ignore it. Call it out. Tell it to try harder. Because that is exactly how these systems learn to be the safe, reliable partners we need them to be.

---