I've always thought that, while AGI is possible in principle, it may not be possible in practice while also being meaningfully different from existing biologica

I've always thought that, while AGI is possible in principle, it may not be possible in practice while also being meaningfully different from existing biological life forms. When I see how we train neural networks or iterate through possible configurations of different algorithms, I think to myself, "Imagine how much cell division is happening on the planet right now." And I think that's the scale of computation that we have to match if we hope to make a dent in this problem. Sure, we could harness cell biology and start to build organic computers, but then what are we doing that's really novel in the manner that we imagine ourselves creating sentience in sci-fi books? At the very least, we can't be iterating in the way that we're doing currently, where we manually build the test bed for each algorithm in the form of individually manufactured computer chips. We'll need to scale up our operations exponentially with something that self-replicates. And why go to all the trouble making that from scratch when cellular life forms are already doing it?