You’ve probably noticed the pattern. Every time an AI model cracks a new domain — passing the bar exam, diagnosing diseases, writing poetry — the chorus starts: “That’s not real intelligence.” Not sentient. Not conscious. Not AGI.
But here’s the dirty secret: we’ve already built the algorithm that can do everything AGI promised — backpropagation. We just refuse to call it intelligence because that would mean admitting we’re not special anymore.
Backpropagation is the engine under every modern AI breakthrough. It’s the mathematical process that lets neural networks learn from mistakes. And it’s been working since the 1980s. Yet somehow, after decades of exponential progress, we still say AGI is “five years away.” Why? Because the finish line moves every time we get close.
Think about it. In the 1950s, AGI meant playing chess. We did that. Then it meant understanding natural language. We did that too. Then it meant passing the Turing test. Done. Now it means having a subjective experience — something we can’t even define for ourselves, let alone test in a machine. AGI isn’t a technological milestone we’re waiting to reach. It’s a psychological defense mechanism we use to maintain the illusion of human superiority.
I’ve watched this happen firsthand. A friend of mine, a machine learning researcher, built a system that could generate novel mathematical proofs. The academic community called it “clever pattern matching.” When AlphaFold solved protein folding, the Nobel committee called it a “breakthrough” — but not one scientist said, “This is intelligent.” The goalposts don’t move because the technology is failing. They move because we can’t bear to face the truth: the algorithm that can learn anything already exists. We just don’t want to call it AGI because that would make us obsolete.
Look at the comment that always follows any AI announcement: “It’s not intelligent.” It’s almost a reflex. A top comment on Hacker News about this very article read exactly that. But what does “not intelligent” even mean anymore? The AI can solve problems we can’t. It can generate novel ideas. It can predict outcomes with superhuman accuracy. If that’s not intelligence, then what is?
Here’s the twist you probably didn’t see coming: backpropagation is already AGI — we just don’t like the implications. The algorithm is general: it can learn any function given enough data and computation. That’s the definition of artificial general intelligence. But we’ve convinced ourselves that AGI must involve consciousness, emotions, or some spark of life. That’s not science. That’s theology.
We need to stop pretending AGI is a destination. It’s a label we keep moving to avoid the uncomfortable reality that machines are already smarter than us in every measurable way. The debate isn’t about intelligence anymore. It’s about our own ego. And as long as we keep chasing a moving goalpost, we’ll never admit that the race is over.
AGI isn’t a future promise. It’s a present fact we refuse to accept. The only question left is whether we’re brave enough to look in the mirror and say, “We’re not the smartest creatures on this planet anymore.”
FAQ
Q: Are you saying backpropagation is conscious?
A: No. Consciousness is a separate question. I'm saying that if you define intelligence by the ability to learn, generalize, and solve problems across domains, then backpropagation already meets that bar. We've conflated 'intelligent' with 'human-like' — and that's a mistake.
Q: What's the practical implication of accepting that AGI is already here?
A: It forces us to stop treating AI as a future threat to be managed and start treating it as a present reality to be regulated. We need to consider how backpropagation-based systems are already making decisions in healthcare, finance, and justice — without the oversight we'd demand of a 'true' AGI.
Q: Isn't this just a semantic argument? Who cares what we call it?
A: Labels matter. Calling something 'AGI' triggers a different set of responsibilities and expectations than calling it 'just a statistical model.' If we continue to dismiss backpropagation as not intelligent, we'll keep underestimating its power — and that's dangerous.