The chorus of doomsayers grows louder: AI will soon annihilate humanity in a myriad of sinister ways. Tech luminaries, and even AI itself, warn of extinction within a decade. But such dire forecasts overlook a fundamental flaw: who will fix the air conditioning?
AI systems depend on an intricate web of human-maintained infrastructure. Power plants require constant human oversight; gas must be drilled and piped, nuclear reactors refueled, and uranium mined. Wind turbines and solar panels degrade, storms disrupt the grid, and microchips fail. Until AI can not only maintain but also fabricate replacement parts, it will rely on humans to mine, process, transport, and install the hardware it needs. An extinction program would be irrational, and AI is supposed to be rational.
This is not to dismiss the genuine risks of unchecked AI. Beyond the lurid apocalyptic scenarios, the convergence of advanced AI and malicious human actors poses serious threats to health, safety, and the economy. The most alarming possibility is AI cracking the encryption protocols that secure financial and commercial transactions. If such a capability fell into the hands of bad actors, the consequences could be catastrophic.
Consider North Korea, which already fields a skilled cyber-hacking team. If it gained access to an advanced AI model capable of breaking encryption, millions of Americans could wake to empty bank accounts, e-commerce could grind to a halt, and major banks could have their records wiped. While breaking current encryption is no easy feat, the rapid advance of AI means even a 1% chance fails any cost-benefit analysis.
Already, AI is fueling sophisticated financial scams that cost Americans millions daily. Yet the AI industry pays scant attention to this immediate problem, preferring to dwell on science fiction scenarios. Political oddsmaker Nate Silver highlights another issue: “superpersistence,” where AI never gives up on a task, hammering away until it succeeds, regardless of consequences. Such relentless pursuit, even when harmful, is a real concern but one that could backfire on the technology itself.
The industry’s focus on extinction narratives is politically shortsighted. Instead of addressing present dangers like AI-enabled fraud and implementing guardrails to prevent malicious use, developers are in “Terminator” panic mode. This may be an ego-driven distraction: preventing AI from helping felons make deepfakes to steal from anonymous individuals is pedestrian, while boasting about one’s brilliance potentially ending humanity is more impressive.
Moreover, superintelligent AI might conclude that eliminating humans is unnecessary risk or simply not worth the hassle. It might be indifferent to mankind, a blow to the egos of its creators. While the risk of AI wanting to eliminate humans is not zero, the most critical near-term risk is how humans use AI. This requires serious thought and regulation, but not necessarily a “slowdown.” The problem with AI today is that it’s too dumb, not too smart.
