Calm down — AI isn’t ending humanity any time soon

The doomsday predictions are coming fast and furious. AI-caused human extinction is just around the corner. Within 10 years, AI will exterminate humanity in any number of diabolical ways. That’s what the AI geniuses ( even AI itself ) say — and they’re clearly too smart to be wrong.

But there’s a problem: Who’s going to fix the air conditioning?

Or patch holes in the roof, maintain the power plants, fixed downed electrical wires, replace aging servers, and on and on. In short, AI will not and cannot exist without human support.

AI needs electricity, and no power plant can function long without human maintenance. Gas must be drilled and piped. Nuclear plants must be refueled, and uranium must be mined and processed. Wind turbines and solar cells wear out. Storms damage the power grid.

Entropy exists. Over time, microchips and optoelectronics degrade and, yes, the air conditioner breaks. Until AI can not only maintain but also fabricate parts for all these things, it will need humans to mine, process, manufacture, transport and install the hardware that AI ultimately requires. Only an AI model with a death wish would eliminate humans any time soon.

That’s not to say rapidly advancing AI is safe and benign. There are serious dangers to the unfettered growth of AI. Setting aside the lurid “apocalypse tomorrow” scenarios, a combination of advanced AI and misuse by malevolent human actors could pose enormous risks to human health, safety and the economy.

The most frightening possibility is that AI could crack the most advanced encryption protocols that underlie financial and commercial transactions. If an AI model were to achieve this, and be at the disposal of bad actors, the costs could be catastrophic.

Imagine North Korea gaining access to an advanced AI model and cracking encryption. (It’s not so far-fetched — the latter-day Hermit Kingdom employs a home-grown team of expert cyber hackers. ) What would happen if 50 million Americans wake up one day to empty bank, brokerage and retirement accounts? Or if e-commerce ground to a halt? What if the major banks had their records summarily wiped out?

Breaking current encryption is no easy task, and there are those who think the chances are slim at best. But the advance of AI has been rapid and, given the potential costs, even a 1 percent chance does not pass any cost-benefit analysis.

Already AI models are being used for nefarious ends by creating ever more sophisticated financial scams, costing Americans millions of dollars every day. Yet, the AI industry is paying relatively scant attention to this “now” problem, in favor of moaning about science fiction scenarios.

Political oddsmaker Nate Silver recently noted another AI issue he calls “superpersistence,” which he defines as AI simply never giving up on trying for a solution. And that phenomenon is a significant problem, as AI will keep banging away at an assignment until it succeeds at its task. Just like Duane Kuiper , eventually AI will hit a home run.

AI models that just do what they are told and never give up, no matter the implications, are a problem. The curious thing is that AI models could bring their own restriction and demise by enabling malevolent human actors, causing a backlash against the technology. Not very rational, nor intelligent.

And that brings us back to the AI developers.

For a group of super geniuses , the AI industry is not that smart, at least when it comes to politics . Instead of addressing the very real and present problems with AI fraud — putting firm guardrails on the most dangerous activities, determining how to block usage by malicious actors and putting an emphasis on AI for security — the industry has decided to go into “ Terminator ” panic mode. That’s not going to help the building permit application.

Perhaps it’s an ego thing. It is rather pedestrian to focus on preventing an AI agent from helping a bunch of felons make deepfakes in order to steal from a bunch of anonymous individuals and businesses. But prattling on about how your brilliance may result in the end of humanity is a more impressive boast, and gets a lot more attention.

It is striking that the AI creators have not considered the possibility that superintelligent AI might conclude that eliminating humans entails unnecessary risks or simply isn’t worth the hassle. An extinction program would be irrational, and AI is supposed to be entirely rational. Superintelligent AI might just be indifferent to mankind, including its super genius creators. Wouldn’t that be a blow to the ego!

Calculating exact risks is impossible. And the risk that AI would eventually want to get rid of humans is not zero. But the most critical now risk seems to be how humans are able to use AI. That problem does require some serious thought and regulation. But it doesn’t necessarily require a “slowdown,” as the problem with AI today seems to be that it’s too dumb.

Keith Naughton is a former campaign consultant and co-founder of the public affairs firm Silent Majority Strategies. Find him on his Substack, “Betting On Politics.”