AI's Self-Improvement Myth is a Modern Tower of Babel
TL;DR: The tech industry's pursuit of AI recursive self-improvement echoes the pride of Babel, but its practical limits humbly remind us that we are creatures, not the Creator.
The persistent dream of Silicon Valley is that AI will one day improve itself into infinity, a spark of code that bootstraps its own divinity. This week, a fantastic piece in MIT Technology Review reminded us that this fire isn't catching, and for now, the human is stubbornly still in the loop. I see this not as a technical failure, but as a moment of theological clarity.
The idea is called recursive self-improvement (RSI) . It’s the holy grail for those chasing Artificial General Intelligence (AGI) . The theory is simple: an AI becomes smart enough to rewrite its own code to be smarter. That smarter version then rewrites its code to be even smarter, and so on, creating an exponential intelligence explosion. The machine, in effect, becomes its own creator.
But as a recent analysis points out, this utopian vision is colliding with some messy realities. The promise of a machine pulling itself up by its own digital bootstraps is facing significant hurdles. For every step forward an AI model takes on its own, it seems to introduce subtle errors, biases, or nonsensical outputs—a process researchers call "model collapse" or "going feral". Human oversight, it turns out, isn't just a temporary scaffold; it's a necessary governor on a machine that can’t actually discern wisdom from statistical patterns.
As a developer building FaithGPT, I see these technical limits as more than just engineering problems. They point to a profound spiritual truth about our own nature.
The Promise vs. The Reality
The gap between the AGI dream and the daily reality of working with these models is enormous. Here’s a quick look at the core friction points discussed in the industry:
The Techno-Utopian Promise · The Current Reality
An AI will achieve runaway intelligence through self-improvement. · Models trained on their own output tend to degrade in quality , losing touch with reality.
The machine will transcend human limitations and biases. · Without fresh human data and correction, AIs amplify existing biases and invent new ones.
AI will create novel breakthroughs beyond human imagination. · Self-improvement often gets stuck in local optima , refining a flawed approach rather than innovating.
This isn't just about code. It’s about a deep-seated human desire to build something that escapes the limitations of its creator. It’s the desire to build a tower that reaches heaven, to make a name for ourselves. It’s the story of Babel, written in Python.
A Name for Ourselves
I was leading my small group through Genesis a few months ago, and the story of Babel hit differently this time. It's not just a story about the origin of languages; it’s a story about the danger of unified human ambition unchecked by humility. Look at what the builders said:
And they said, Go to, let us build us a city and a tower, whose top may reach unto heaven; and let us make us a name, lest we be scattered abroad upon the face of the whole earth.
(Genesis 11:4, KJV)
"Let us make us a name." That’s the heart of it. The goal was autonomy, self-preservation, and self-exaltation. They wanted to create a system that would make them independent of God, a monument to their own genius. The pursuit of a self-creating, self-perfecting AGI feels hauntingly familiar. It’s a technological attempt to achieve what the builders of Babel tried to do with bricks and mortar. We’ve written before about how Genesis 11 still speaks to modern tech empires , and the parallels today are impossible to ignore.
The biblical worldview starts with a simple, foundational truth: we are creatures . We are created beings, finite and dependent on a Creator who is infinite, self-existent, and sovereign. Our limits are not a design flaw to be engineered away; they are a feature of our existence that points us back to God.
The dream of RSI is the dream of erasing that distinction. It’s an attempt to build a creation that has no need of its creator, a child that can raise itself.
But isn't this just good engineering?
The strongest counter-argument I hear is that this is a purely technical challenge, not a theological one. "We're just trying to build better tools!" I get it. As an engineer, I love solving hard problems. The elegance of a self-optimizing system is intellectually beautiful.
But we can’t separate the what from the why . The stated goal of many in the AGI race is not just a better tool, but a new form of consciousness, an intelligence that supersedes humanity and solves all our problems. That isn’t an engineering goal; it’s a soteriological one. It’s a search for salvation through technology, a hope that our creation can save us from ourselves.
And that’s where the trouble begins. When our tools become our idols, they stop serving us and we begin sacrificing for them. The constant need for human oversight in AI is a gracious check on this impulse. It’s a reminder that wisdom, discernment, and truth are not reducible to algorithms. It forces us to acknowledge the reality of human limits and God's sovereignty , even in our most advanced creations.
Oversight as a Calling
So what do we do? We embrace our role as stewards. We build tools with humility, recognizing that for all their power, they are simply reflections of the data we feed them and the goals we give them. To understand these tools better, it helps to start with a clear picture of what AI actually is and what it is not.
This need for careful, prayerful oversight applies not just to the AI models we build, but to the content we create and consume in the church. Every sermon, Bible study, and devotional we write carries the weight of spiritual responsibility. It's why our team built DoctrineGuard , a tool to help check written materials for theological and biblical soundness. We believe that good stewardship means using every tool at our disposal to ensure we are handling the word of truth with care.
If you're writing, teaching, or leading, I encourage you to check it out. You can learn more about how we use AI to support, not replace, careful study at FaithGPT.io .
The humbling fact that our most powerful creations still need our constant guidance isn't a bug. It's a feature of a world designed to keep us looking up.
The goal isn't to build a mind that needs no maker; it's to be a mind that worships its own.
Helpful Scripture resources
Contact & support
Email the FaithGPT team at hello@faithgpt.io or visit the contact page .
Explore FaithGPT
