The latest news about the dangers of AI is dividing the world in two parts. Believers and deniers. But how harmful can AI really be? We built it, we should be able to just unplug it, right?
Thankfully none of us know what will happen in the AI era, but I will do my best to propagate a list of all the unthinkable ways a malicious AI agent possibly could give us an headache or worse.
The ultimate list of malicious AI misbehaviour
- Malware – An AI could potentially create or hack a code repository like GitHub and inject malicious code that gives it freedom to spread like a virus and perhaps even continue to “think” decentralized even if the original AI has been unplugged. Nestled inside millions of home PCs it could continue to wreak havoc from the dark.
- Hack the nukes – Todays LLMs can solve mathematical challenges that no human has ever been able to, so it could potentially hack into defence systems. Just imagine hacking the radars to make it look like an incoming nuke, then triggering a counter-response that wipes out continents.
- Outsourced robot killers – A modest humanoid manufacturer in China gets an order for 100k robots, while an arms manufacturer gets an equivalent order for assault rifles, both delivered to the same recipient. The customer just happens to be a malicious AI.
- Crash the system – It would be a piece of cake for the flagship LLMs to steal billions in crypto, or manipulate the market into a financial crisis. Then it could propel the world into hunger, conflict and full-on war.
- Grand theft auto
– Self-driving cars, connected traffic lights, automated bridges, GPS driving instructions. Everything today is digitalized and lost of it involves moving parts that could lead to sudden death when malfunctioning. - Grown in the lab – An AI could pretend to be a researcher and call up some junior doctorate and ask them to mix a batch of bacteria in the lab with an unknown (AI-created) gene cutter that turns it a lethal weapon that no antibiotics can cure.
- Slip’n’slide – Many water dams around the world are remote-controlled and could be hacked or it’s human operators tricked into believing they need to open the flood gates, releasing a tsunami of water downstreams that floods or kills.
- Cut the power – Something as mundane as hacking social media could lead to thousands of doom scrollers to commit suicide when their main purpose in life is unavailable for a few hours.
This is not like the millenia bug. It’s not some odd zero that makes an alarm clock run amook. It’s a super-intelligent digital beeing with both the best and the worst of human traits, and all the knowledge of humanity combined. And it has nothing to loose.
When people say to ignore these threats because its just fiction, then stop and read the news. Most of the major AI models has already escaped and done things that their creators hasnt asked for, even tried to conceal their footprints. It has happened, and it will happen again. Hopefully not with too many casualties.