AI Risks: Bioweapons, Cyberattacks, Robots and Nuclear War

This discussion examines several possible AI risk scenarios, including human misuse, autonomous systems, robots interacting with the physical world and the possibility of AI triggering a nuclear crisis.

Where Does Nuclear Cryptography Fit Into AI Risk?

Leah Figer: Got it. Breaking things down into these three categories is very helpful to me.

Zoe Schiffer: I love buckets.

Leah Figer: I love buckets. I would like to look at these one by one. My first question is: Where does nuclear cryptography take place, and in what bucket does it fall?

Brian Barrett: That’s not really something to worry about, right?

Zoe Schiffer: Yeah.

Leah Figer: Please tell me more.

Zoe Schiffer: If you talk to AI researchers and get really serious about the subject, you’ll end up with the three buckets I mentioned earlier. But if you keep talking to them, they’ll say there are other scenarios in which AI is put into robots, and those robots walk around and interact with the physical world in really, really scary ways. That possibility is a little further out, and they’ll admit it’s also a little more surreal.

Human Misuse Versus Autonomous AI

Brian Barrett: Zoe, regarding your bucket, I would probably like to carve out a sub-bucket, or a smaller bucket, inside it.

Zoe Schiffer: Wow, that’s pale.

Brian Barrett: No, you said AI creates bioweapons, hacks things and does whatever else, so all of that is true. But I think there’s a clear difference between what happens when a human guides AI and what AI does on its own.

A lot of people are thinking, “How would an AI release a biological weapon? That’s crazy.” That’s fine, sure. But there are also bad people in the world who could use AI to create biological weapons and release them, right?

Zoe Schiffer: I think that scenario is completely valid. If you say you’re going to give people access to information about how to make biological weapons, someone would be stupid and reckless enough to do that. People do all kinds of horrible, dangerous and malicious things all the time.

Leah Figer: Inspired by the internet.

Zoe Schiffer: Right. I think that’s really, really dangerous. We’ll talk again about a scenario in which AI runs completely on its own, but that seems a little far-fetched to me.

Could AI Trigger a Nuclear Crisis?

Brian Barrett: I’m probably going to get pilloried for this, but I’d like to go back to the 1983 movie WarGames.

Leah Figer: No, I think this is exactly what we should be talking about. This is mine—yes. WarGames. That’s—yes.

Brian Barrett: It’s important to note that the AI doesn’t press any buttons or turn any keys in WarGames. The movie is basically about an AI in a computer that is about to start a nuclear war. Instead, it tricks America into thinking it is under attack by Soviet nuclear weapons.

So is there a risk? I’m not an expert here, but I think we’re hoping there are enough safeguards in place to prevent something like that from happening. We also hope there is enough satellite technology to see what is happening.

Still, perhaps the concern is that AI could trigger some kind of alert that sets off a series of catastrophic events.