When intelligence becomes a substitute for innovation
Imagine it’s 1956. Bell Labs has just achieved the impossible: a Large Language Model with voice capabilities. The machine can understand spoken requests, hold conversations, and execute complex tasks. It’s a technological marvel that would make modern engineers weep with envy.
And what do we use it for?
Switching telephone calls.
Every time someone picks up the phone, our miraculous AI answers: “Number please?” It listens, understands, and manually connects wire A to wire B. Thousands of these AI operators work around the clock.
The automatic telephone exchange—a brilliantly simple system of electromagnetic relays and rotary switches—remains uninvented. Why bother? We have artificial intelligence.
The Question We’re Not Asking
This scenario never happened, of course. But before you dismiss it as absurd, ask yourself: How would engineers in 1956 know there was another way?
If you can build a sufficiently intelligent system to handle any task, why would you spend decades developing task-specific solutions? Except the system is fundamentally unchanged. Calls still need to be manually switched, one at a time. The company has simply swapped one cost (human labor) for another (massive AI infrastructure and electricity bills). The problem—the fundamental inefficiency of manual switching—remains unsolved. They’ve built artificial humans, complete with most of their limitations, just without the need to eat or sleep but with staggering power requirements.
And here’s the terrifying part: Those engineers would never know they were trapped. They would never invent the automatic telephone exchange.
The Modern Switchboard
Look at how we’re deploying AI today and see if the pattern looks familiar.
Customer service representatives are being replaced by chatbots that handle the same convoluted processes humans struggled with. Content moderators are being replaced by AI that applies the same inconsistent rules human moderators applied. Copywriters are being replaced by language models that produce the same formulaic content humans were trained to write. Radiologists review fewer scans while AI assistants review more.
We’re not re-engineering customer service—we’re building AI that can navigate our deliberately obtuse phone trees. We’re not fixing content moderation—we’re creating artificial moderators that enforce our confused policies at scale. We’re not revolutionizing radiology—we’re adding an expensive AI layer to the same diagnostic pipeline.
The companies save on salaries and benefits. Then they spend it on GPU clusters, API calls, and electrical bills that would make a small nation blush. Workers lose their jobs. And we call it progress.
That’s substitution, not transformation.
This is our AI switchboard operator moment. And like those hypothetical 1956 engineers, we don’t see the trap.
What Real Progress Looks Like
The automatic telephone exchange wasn’t just more efficient than human operators. It was a fundamentally different solution. It didn’t automate the job—it eliminated the need for the job to exist.
One automatic exchange could handle thousands of connections simultaneously with no intelligence required—artificial or human. No one needed to understand the request, remember phone numbers, or manually connect wires. The system was re-engineered from the ground up. The problem was solved, not just automated.
That’s what real progress looks like. The ATM didn’t just replace bank tellers with machines that did the same job—it redesigned how people access their money. The shipping container didn’t replace longshoremen with robot arms doing the same loading—it transformed the entire logistics system. The barcode didn’t replace inventory clerks with AI that could count faster—it made counting automatic.
These solutions eliminated jobs, yes. But they also created massive new efficiencies, reduced costs for everyone, and opened entirely new possibilities. They were genuinely transformative.
Replacing a human switchboard operator with an AI switchboard operator? That’s just expensive stagnation with a cutting-edge price tag.
The Trap We Can’t See
Here’s what makes this so insidious: From inside the system, it looks like progress.
The metrics tell you you’re winning. Headcount down. Automation up. AI deployment successful. The jobs are being done—calls are being connected, tickets are being resolved, content is being produced. The quarterly report looks fantastic.
But step back and look at the bigger picture: You’re spending enormous resources to maintain fundamentally inefficient systems. Your AI infrastructure costs millions. Your energy bills are staggering. Your models need constant updates, monitoring, and correction.
And the system itself? Still broken. Still inefficient. Still designed around limitations that no longer need to exist.
We’re in the same trap today. We just can’t see it.
The Cost of False Progress
When you replace workers with AI without re-engineering the underlying system, you’ve chosen false progress. You’ve chosen to spend enormous resources maintaining inefficiency rather than investing in genuine innovation. You’ve chosen the comfortable path of automation over the difficult path of transformation.
And here’s the real cost: Opportunity.
While our hypothetical 1956 engineers perfected their AI switchboard operators, they weren’t developing electromagnetic relay systems. They weren’t exploring crossbar switches or step-by-step selectors. They weren’t thinking about how to eliminate the manual connection problem because they’d already “solved” it.
The same is happening today. Every dollar spent on AI to replace workers in broken systems is a dollar not spent on re-engineering those systems. Every engineering hour devoted to making chatbots handle terrible customer service workflows is an hour not spent designing workflows that don’t need chatbots. Every brilliant mind working on AI content moderation is a mind not working on platforms that don’t generate the same moderation nightmares.
We’re solving the wrong problems. Or rather, we’re not solving problems at all—we’re automating around them. And in doing so, we’re ensuring those problems become permanent fixtures of our technological landscape.
The engineers of 1956, trapped with their AI operators, would never have known what they were missing. The automatic exchange would have seemed impossible—unnecessary, even. Their system worked. Why change it?
Why indeed?
The Uncomfortable Truth
AI itself isn’t the problem. Artificial intelligence is genuinely revolutionary for problems that require intelligence: drug discovery, climate modeling, scientific research, creative exploration. These are domains where there is no simple mechanical solution, where intelligence—artificial or otherwise—is the appropriate tool.
But that’s not how most AI is being deployed. Most AI today is being used to replace human workers in systems that shouldn’t require intelligent beings in the first place. We’re building artificial humans to do human jobs rather than building systems that don’t need humans (or artificial humans) at all.
Why We’re Stuck
Sometimes substitution seems rational. Regulatory frameworks make system redesign prohibitively slow. Legacy systems must interoperate with dozens of external dependencies. Demand is uncertain and over-engineering would be wasteful. In these cases, AI substitution can be a reasonable bridge.
But the trap isn’t using AI to replace workers—it’s treating that replacement as the destination rather than a waypoint. What begins as pragmatic substitution calcifies into permanent architecture. The temporary bridge becomes the foundation for everything built afterward. Organizations optimize what they have, not what they could create, and the local maximum becomes indistinguishable from the global one.
And we might not be able to escape. AI works. It reduces headcount. It looks like progress in every quarterly report. The incentives all point in one direction: more AI, more automation, more worker replacement. Who benefits from asking whether there’s a better way? Not the companies that have invested billions in AI infrastructure. Not the AI industry itself. Not the executives whose bonuses depend on “automation success.”
The Exchange We’ll Never Invent
Somewhere in our future, there are elegant engineering solutions we’ll never discover because we’ve already “solved” the problems with AI.
The automatic telephone exchange took decades to perfect, but once deployed, it handled millions of calls with no intelligence required and minimal maintenance. It was a genuine solution: elegant, efficient, transformative. It didn’t just replace switchboard operators—it made the entire concept of switchboard operation obsolete.
That’s the kind of progress we’re trading away. Every time we replace a worker with AI without questioning why the work exists in the first place, we’re choosing expensive automation over genuine innovation. We’re choosing to perfect our AI switchboard operators rather than invent our automatic exchanges.
The question isn’t whether AI will replace human workers. It already is. The question is whether, fifty years from now, our descendants will look back and wonder why we spent trillions building artificial humans to perform human jobs instead of building systems that didn’t need humans—or artificial humans—at all.
Will they marvel at our AI operators? Or will they wonder why we never invented the automatic exchange?