Mark Whittington asked in these pages whether artificial intelligence might become the only way humanity can survive. It is a question worth taking seriously. If AI can help us detect pandemics earlier, discover medicines faster, manage complex energy systems, identify approaching hazards, or solve problems too large for human teams to process in time, refusing to develop it could carry risks of its own.
If AI becomes humanity’s lifeboat, we should build the lifeboat. We should also keep the oars.
There is an awkward fact hiding underneath the debate over whether humanity will become dependent on AI. A large part of humanity is still only loosely connected to the digital infrastructure on which advanced AI depends.
The International Telecommunication Union estimates that 6 billion people used the internet in 2025. Another 2.2 billion, roughly 1 person in 4, remained offline. The latest global energy-access assessment reports that 666 million people still lack basic electricity access, about 1 person in 12. Most live in remote, low-income, or fragile areas, particularly in sub-Saharan Africa.
Those numbers should never be romanticized. Lack of electricity and connectivity limits education, healthcare, economic opportunity, and safety. The goal should be to connect those populations, not preserve deprivation as some kind of technological virtue.
But the numbers establish something narrower and important: human existence and the digital stack are still different things.
Advanced societies are rapidly closing that distinction. Electric grids depend on software. Hospitals depend on networked records and logistics. Transportation, finance, communications, food distribution, and government increasingly rely on cloud services and automated decision systems. AI will make many of those systems better. It may eventually make some dramatically better.
The danger appears when “better with AI” quietly becomes “unable to function without AI.”
That is a resilience problem before it is a science-fiction problem.
Design for graceful degradation
Engineers already have a vocabulary for dealing with this. One principle is graceful degradation. When the most capable component disappears, the larger system should lose performance in stages rather than collapse completely.
A hospital deprived of its AI scheduling service may become slower. It should still treat patients. A logistics network that loses an optimization model may move fewer goods. It should still know where essential deliveries need to go. An electric utility may lose predictive intelligence while retaining protection, control, and restoration functions.
The objective is not to reproduce the AI’s full capability without the AI. The objective is to preserve the function that matters most.
Keep an independent fallback
A second principle is independent fallback. A backup that depends on the same model provider, cloud region, communications network, electrical supply, or identity service as the primary system may be a redundancy on an architecture diagram and a single failure in reality. Critical systems need some recovery paths whose failure modes are genuinely different.
That may mean simpler software. Sometimes it means local control. Sometimes it means procedures that look inefficient during normal operation precisely because they exist for the day normal operation disappears.
The same logic applies to human judgment. The Amish population of the United States and Canada is estimated at about 411,000. Amish communities should not be described as technology-free. They make deliberate distinctions among technologies rather than accepting every innovation simply because it exists. Their choices arise from religious and social commitments that most Americans do not share. The relevant lesson is the existence of choice itself: a community can ask whether a technology serves its institutions before allowing the technology to reshape them.
Modern society cannot adopt the Amish technological model, nor should it try. But it can recover that question.
What are we allowing the machine to become necessary for?
Capability without captivity
That leads to a third engineering principle: bounded authority. AI can recommend, optimize, search, diagnose, plan, and discover while the surrounding architecture determines which functions may depend on it and which remain independently recoverable.
This is capability without captivity.
The distinction becomes especially important if Whittington’s underlying concern proves correct and future AI systems become exceptionally valuable against existential threats. A technology powerful enough to protect civilization will naturally migrate into more parts of civilization. Every success will create pressure to connect it to the next system.
That can produce extraordinary capability. It can also produce correlated dependence.
The answer is not to reject the technology. It is to design recoverability while dependence is still optional.
There will be costs. Maintaining degraded modes consumes money. Independent systems create duplication. Human operators lose proficiency when automation performs a task better every day, so preserving meaningful fallbacks requires training and exercises. Some older procedures will eventually become so inferior that retaining them makes little sense.
Resilience always carries friction.
The relevant test is therefore not whether society can recreate an advanced AI system with people, paper, and telephones. It is whether losing that system also means losing the essential function it supported.
AI may become one of the most valuable technologies humanity has ever built. Let it help discover drugs, operate complex infrastructure, model dangerous events, accelerate science, and reveal risks that humans would otherwise miss.
We should want the strongest AI we can build safely.
We should also want institutions capable of surviving its absence.
A civilization that needs one technology to survive has given that technology extraordinary leverage over its future, regardless of whether the technology ever develops intentions of its own.
OPINION: TRUMP SAYS AI NEEDS NO NEW GUARDRAILS. THEN WHO GETS TO TURN IT BACK ON?
So build the systems that may save us. Use them. Improve them. Give them difficult problems. Just preserve graceful degradation, independent fallback, bounded authority, and the ability to recover when the intelligence goes dark.
Build the lifeboat. Keep the oars.
Burak Oktenli is an independent researcher based in Washington, D.C., focusing on assurance, verification, and bounded machine authority in autonomous and AI-enabled systems. He holds a bachelor’s degree in computer science and engineering from the University of South Florida, a Master of Business Administration from Lynn University, and is completing a Master of Professional Studies in applied intelligence at Georgetown University.