AI Can't Kill Us. Our Own Stupidity Can
I don't think AI is going to kill us. I use it every day, and it still struggles to write a simple script without over-engineering it. The real risk isn't that it becomes smarter than us. It's what we do with it without thinking.
- AI
- Cybersecurity
- Opinion
Every so often someone comes out saying artificial intelligence is going to kill us all. Serious researchers resign, sign letters, talk about extinction probabilities. I understand the concern, but I don't share it, at least not in those terms. I don't think AI is capable of killing us. I think we're perfectly capable of doing it with it.
I use AI every day to code, and it still struggles to write a simple script without piling on layers of abstraction nobody asked for. I ask for a function and get back an architecture. I ask it to fix a bug and it touches five files that had nothing to do with it. That's the tool that's supposedly going to give rise to a superintelligence capable of improving itself without making mistakes.
I'm not saying this to downplay what AI can do. I use it because it helps me, a lot. I'm saying it because the doom narrative assumes a flawless execution I haven't seen anywhere yet. If it can't write simple code without over-engineering it, the idea that it's going to rewrite itself perfectly, without introducing a single error along the way, doesn't add up for me.
The real risk, to me, isn't in the machine. It's in the person using it without thinking. Someone who gives it access to something it shouldn't have. Someone who blindly trusts an answer without checking it. Someone who automates an important decision because it seemed faster than actually thinking it through.
For the first scenario, the machine going out of control, there's an entire discipline built for it: cybersecurity. Sandboxes, access limits, reviews, alignment, audits. It's a technical problem, and like any technical problem, it has technical defenses.
For human stupidity, there's no security that covers it. There's no firewall for the person who hands admin access to something they don't understand, and no patch for the one who'd rather not check what the AI handed back because they trust the machine more than their own judgment.
In the end, what worries me isn't AI becoming smarter than us. We're already pretty good, with no help from anyone, at making decisions that go wrong. All AI does is give us a faster tool for doing it.