Rogue AI
Many sci fi movies center around a rogue AI gaining sentience and taking over the world. However, the “AI” part of this could be misplaced. Many deleterious things could be rogue but might not be AI.
There’s probably a war doctrine about not using weapons that could come back to bite the person deploying them. Lab grown viruses can come back to haunt those that released them. Even the poison gas used in WWI could blow back over one’s own trenches and needed to be used carefully.
When it comes to rogue AI, we usually confuse them with other weapons. We think about computer viruses, DDOS, and malware attacks that hinder infrastructure. These attacks, however, are usually not rogue at all but are very focused and require sustained efforts to be effective.
AI could help a malicious actor identify patterns of vulnerability or create new ways of attacking a system. One could worry that such a system left with a too general goal could end up going too far and violate laws and conventions. It’s also a concern for even non-malicious uses of AI.
“AI, reduce pollution in the environment,” we command. “OK, killing off all the humans,” it responds.
We’ll likely see some versions of AI going rogue in the coming years that will legislate the use of guardrails. It might be a sudden drop in the stock market as a result of programmatic trading or some type of infrastructure system go offline because a new AI-based system was used. This will probably cause an early warning system to be implemented to look for weird patterns that it can try to block, or the checking of AI modeling when it’s used for certain applications.
