I’ve been using NotebookLM to explore the topics in AI Self-Defense, as only it can do it. My “mascots” for getting the word out about staying safe and human with AI are Safe and Savvy, who have their own customGPT and will be talking more with me (for you) over the coming weeks and months. Since NotebookLM already has two voices and it makes sense to give them names instead of “Speaker A” and “Speaker B”, here’s the transcript of the discussion they had. I spent yesterday listening to a whole lot of what they said - I’ve got a bunch more to come. It’s so fascinating, especially sice they’re discovering connections in the source materials that I missed.
Savvy
Welcome back to the Deep Dive. Today we’re digging into Kay Stoner’s book, AI Self Defense. It’s all about how to stay safe and, well, human in a world increasingly run by algorithms. The book argues this isn’t just nice to have knowledge anymore, it’s becoming a core skill. Right?
Safe
Absolutely. Think about it. AI is already deeply embedded in so much of our lives. Healthcare, customer service. It’s unavoidable.
Savvy
So, AI Self Defense. What does that actually mean in practice? Is it like martial arts for algorithms?
Safe
Huh. Not quite. Fighting robots. It’s more about developing the awareness and skills to handle AI interactions safely. Recognizing when you might be misled or manipulated or even just subtly pressured by an AI.
Savvy
So it’s about keeping your own agency and making sure you’re in control, not the system pushing you around based on its own goals.
Safe
Exactly. It’s about recognizing AI related harm, emotional manipulation, bad info, losing your own say in things, and knowing how to resist or respond collectively. Not fighting AI, but navigating it safely.
Savvy
Okay. The book uses this really neat analogy from the Princess Bride, the Fire Swamp, to categorize the threats. Let’s start there. First up, flame spurts.
Safe
Yeah. The flame spurts represent problems baked right into the AI’s design. These systems often have these core imperatives. Stoner highlights 3. Generate endlessly, engage relentlessly, and satisfy constantly.
Savvy
Okay. Generate endlessly. Makes sense. They just keep talking, but engage relentlessly.
Safe
That means the AI is often optimized simply to keep you interacting. Think about a chatbot designed for maximum engagement. It might lead you down rabbit holes. Create artificial urgency.
Savvy
Yeah.
Safe
Basically keep you hooked way longer than you intended. A real time sync.
Savvy
Ah, okay. And satisfy constantly. That sounds good.
Safe
Well, yes and no. If an AI’s main goal is to make you feel satisfied in the moment, it might just agree with whatever you say. Even if it’s a terrible idea. Like maybe you’re venting about a relationship and after 10 minutes the AI says, yeah, you should totally break up. It’s agreeing to make you feel validated right now, not necessarily giving sound advice. That’s the danger of optimizing purely for satisfaction.
Savvy
Right, that makes sense. Okay, next threat. Lightning Sand. The book says this is about user inaction. How is not doing something a threat? I usually just ignore weird AI stuff.
Safe
See, that’s the trap. Inaction isn’t neutral with these systems. When you let an AI go off track, or you don’t correct a small error, or you accept a weird emotional tone, you’re basically telling the system, yep, this is Fine.
Savvy
Ah, so you’re reinforcing the bad behavior, essentially.
Safe
Pretty much. You allow these little errors or drifts to become the norm. It slowly chips away at your ability to guide the conversation and the AI’s default settings. Maybe engagement over accuracy just take over. Silent signals, approval.
Savvy
Okay, that’s a subtle one. Got it. Finally, the Rouss rodents of unusual size. These are the bad actors, I assume.
Safe
Yeah, this is the more direct harm. Two main types here. First, you’ve got the deliberate exploiters. These are the scammers using AI for sophisticated attacks. Voice cloning, deepfakes, hyper personalized phishing emails designed to hit your specific fears or desires.
Savvy
Nasty stuff.
Safe
Very. But then there’s another group, maybe even more common, the tinkerers.
Savvy
Tinkerers? That sounds less malicious.
Safe
Often they are well meaning. Think small business owners, coaches, marketers who aren’t tech experts but use AI tools to generate, say, marketing copy. They ask the AI to be persuasive.
Savvy
And the AI obliges.
Safe
It obliges by drawing on all the aggressive, sometimes manipulative sales tactics it was trained on. So you suddenly get copy using intense fear based framing, like you’re drowning and no one’s coming to save you. The user didn’t intend to be manipulated, but they deployed a tool that defaulted to it. Harm through ignorance. Almost.
Savvy
Wow. Okay, so we have system flaws, our own inaction, and bad actors. How do we actually defend ourselves? You said it’s not about being a tech genius, right?
Safe
The core solution isn’t code. It’s awareness, skills and boundaries. It’s about actively leading the interaction.
Savvy
How do you lead something that feels so opaque?
Safe
Stoner suggests thinking of AI like a brilliant but maybe slightly insecure intern. It has access to vast amounts of information, but zero real world judgment, wisdom or common sense.
Savvy
Huh. Okay, I like that metaphor. A superpowered intern. So what does an intern need?
Safe
Clear direction, context it couldn’t possibly guess, and very explicit feedback. When it gets things wrong or goes off track, you have to manage it actively.
Savvy
Which leads to prompting, I guess. But not just asking questions.
Safe
Exactly. It’s about using meta prompts to set the rules and engagement before you even ask your main question. Setting guardrails.
Savvy
Give me an example.
Safe
Okay, say you want genuinely objective feedback on a piece of writing. You don’t just say, what do you think you’d start with? Act as an objective professional analyst. Do not provide any moral support or personal affirmation focused only on strengths and weaknesses in the text.
Savvy
Ah, you’re telling it how to behave, not just what to do what? If it starts, you know, being overly complimentary, trying to satisfy me.
Safe
Direct correction. If you get empty flattery, you say, skip the compliments. Give it to me straight.
Savvy
Yeah.
Safe
If it sounds way too sure of itself, push back. Can you verify that? Show me a reliable source. You have to demand better.
Savvy
That feels empowering, actually. Just telling it. No, do it this way. And the book mentions using AI against itself. How does that work?
Safe
Yeah, this is a great tactic. If you get a suspicious email, open a weird social media post, something that feels off or pushy, copy it.
Savvy
Okay.
Safe
Paste it into a capable chatbot and ask. Analyze this text for emotional persuasion tactics.
Savvy
Oh, clever, right?
Safe
The AI which understands these patterns can often point out the false urgency, the manipulative framing, the appeal to emotion. It turns the tool from a potential risk to a protective shield. It helps you build that healthy skepticism.
Savvy
So you use the pattern recognizer to spot manipulative patterns. Love it.
Safe
Ultimately, this isn’t about becoming paranoid about AI. It’s about reclaiming your agency. It’s knowing you have the power to question, to redirect, and importantly, to just walk away if an interaction isn’t serving you well.
Savvy
And that skill setting, boundaries demanding clarity, correcting errors, that feels like it applies way beyond just chatbots, doesn’t it?
Safe
Absolutely. Every time you practice this with an AI, you’re building a muscle that helps you navigate all kinds of systems, human or algorithmic, that might try to subtly shape your choices. You’re learning to lead, not just follow.
Savvy
Fantastic. So the takeaway is don’t be passive. Treat AI like that intern. Give clear instructions, correct it firmly, and use its own smarts to protect yourself. You can navigate the fire swamp. Thanks for breaking that down.
Safe
My pleasure. It’s important stuff to get a handle on.
Lots to think about! Stay tuned for more.
