Murderbots?
The real risks of AI, and how to make sure that we don't go extinct.
RedEye spoke on background with a Baton Rouge-based computer expert who has worked at the highest levels of the federal government on cybersecurity and AI risks. Because of the sensitive nature of his work, he could not speak on the record.
On the risks: “High consequence, low probability events are always possible. But you have to look at both factors: consequence and probability. AI models are certainly progressing quickly. And I do think they could be used to accelerate development of biological pathogens and to unleash destructive cybersecurity events. Those are real risks that should be identified, mitigated against, monitored, and protected against. Much like we do today for both, and for nuclear weapons. The risk was always there with high consequences, but the probability is now higher. We have to work to bring the probability lower.”
How to control the risk: “Stop building data centers and constrain available compute and memory capacity. That’s the number one through ten ways to throttle AI advancement and let policy and regulation catch up.”
On AI’s most likely consequences: “We’ll probably just decay into more ignorant and religiously zealous people, fight each other in civil disturbances, crash our democracy and economy, and become another third-world hellhole. You can totally use AI to do that because it’s already being used to do so in social media and influence campaigns. We used to call it PSYOPS.”