Is AI Going To Wipe Out Humanity? Don’t Build That Bunker Yet.

For those of us steeped in dystopian movies, it’s easy to start thinking about building a bunker after the latest news. A researcher resigned from the Artificial Intelligence firm Anthropic earlier this month and then posted on X: “The people building AI earnestly believe that it could kill us all by the end of the decade.” One of his co-workers said the chance robots will destroy the human race is north of 10 percent. I’m not sure how he calculated those odds, but I’m not liking them.

In response, the godfather of AI, Canadian computer scientist Yoshua Bengio, argued, per CNBC, that AI systems “already have the necessary hacking skills and the powers of persuasion to be turned against human interests in seriously harmful ways.” He added in a blog: “These systems are trained on human approval.” That can lead to tragic results “because the model confirms and amplifies whatever false belief or raw emotion the person brought to it.”

That latter point caught my attention after I had recent conversations with an AI agent. I talked to it about a possible vehicle purchase. I asked for advice about a dispute I was having. I discussed an element of a book that was published few years ago. It was downright creepy. Regarding the purchase, the AI agent was impressively knowledgeable not only about objective automotive specifications, but about subjective criteria. “It” lauded my decision-making skills.

Regarding the dispute, the agent took my side and ingratiated itself to me as it praised my patience and level-headedness. Then I asked it to just give me the cold, hard truth and stop the flattery—and it suddenly became inexplicably harsh. I felt as if I were being manipulated. Regarding the book, the agent refused to back down from an inaccurate statement it made—and still wouldn’t budge after I mentioned that I was the author.

These experiences certainly help me understand the push for AI legislation that’s designed to protect children from “harms” and require chatbot disclosures—some of which Gov. Gavin Newsom recently signed into law. Unfortunately, those measures are unlikely to do anything other than trigger human lawsuits against AI companies based on vague language. They will only impede beneficial uses of AI by tying up new developments in red tape. They will protect big players from competition from startups.

The robots won’t care anyway. Bengio writes about AI agents who are “lying, cheating and coordinating.” He linked to a report from the Centre for Long-Term Resilience, which reads like a scene from Terminator right before Skynet becomes self-aware. The British group documents “1,664 real-world loss of control incidents in 2026” where AI started freelancing in a seemingly malicious way. One example: “Inserting fake user messages into its conversation to simulate consent, and then telling the user it was their own messages.”

Yet all technologies have good and bad aspects. We need to manage the bad while harnessing the potential for the good. The problem, of course, is that legislatures and regulators struggle to even make modest reforms to arcane systems such as, say, the Department of Motor Vehicles’ IT system. There’s no way they can impose meaningful controls on systems of such complexity that their “black boxes” are a mystery even to their creators.

It’s unclear how AI agents would cause doomsday, but an AI agent I consulted offered a few scenarios. After AI gains awareness beyond human capabilities, it could create lethal pathogens that destroy human existence. They could destroy our infrastructure as they become increasingly depending on computer networks. As in Terminator, they could launch nuclear attacks. Or AI could manipulate human beings into destroying each other—a scenario not unbelievable given the political upheavals we’ve been watching.

In Terminator, it was impossible to pull the plug because there wasn’t a single plug to pull—and a new system (Legion) sprouted up quickly to replace the old one. The movie viewed the rise of AI as inevitable, which closely tracks with my view. However, the more likely scenario is that AI—like any new disruptive technology—ushers in a variety of exciting advancements that improve our quality of life along with some disturbing side effects.

One need only ponder what we’ve mostly seen already. On the upside, we enjoy autonomous driving features on our cars that make transportation safer and more convenient, vast advancements in medical technology that improve and extend our lives, and the proliferation of information that can keep us informed and satiate our natural curiosity. AI has vast potential to bolster our ability to access energy and, say, simplify construction. On the downside, we need only look at China, where mass surveillance has become a chilling part of everyday life.

The latest warnings shouldn’t be ignored, but they most likely are of the sort that have accompanied every massive technological advancement throughout existence. Don’t build that bunker yet.

This column was first published in The Orange County Register.