Alright, so in the last post I was talking about growing cars (yes, I’m pretending this is a talk-show). I tried to argue that the difference between growing cars or building them is really just a matter of perspective. In my personal experience, having never worked at or even come close to a car factory, both apples and cars magically appear in stores, and both are happy to be mine if I am willing to part with some money. In contrast, a hobbyist trying to build their own car is extremely aware that cars don’t just grown on trees. So what exactly would we like to achieve by somehow engineering a “car tree”?
Perhaps the key point is how much human involvement goes into the production process. However, with ever-increasing levels of automation, we can imagine a car factory run entirely by robots (like Tesla is trying to be). Still, such scenario seems materially distinct from the idea of actually growing cars. To me, this distinction seems to be not in what humans do to produce cars, but rather in what humans know — in other words, it’s about information! In an autonomous car-factory, we have still designed every bit of the production process in full detail, and also spent quite a lot of effort to make the robots working there. In contrast, an apple tree grows apples without us even understanding how (morphogenesis is murky business, still). Worse yet, neither do we understand how an apple seed grows into a tree. Nor how apple seeds develop (ok, we understand a bit, but not at the level we understand cars).
So we seem to have arrived at the conclusion that if we really want to “grow cars,” we must find a way to do this without ever knowing how we are doing it (else it would still really be “build”). That seems a bit paradoxical — how do we create something we inherently don’t understand? Well, notoriously, neural networks seem to be just that: we have no clue how they work even though we build them, and yet they seem to be quite good at what they do. So “growing cars” could, in some sense, also realize some black-box learning algorithm — but now directly in the physical world. The learning would then be what makes the “fruits” act like cars: we could optimize some set of desired criteria, like speed and comfort. This way all the information about design would be somehow generated by the learning algorithm, rather then by engineers.
Alright, does this seem to be getting closer to the intuition of “growing cars”? What other important aspects are we missing from the biological metaphor? Even without that, could this be a nice future technology, or is it pretty useless as described? I’m sort of thinking I’d also like my cars to be able to self-heal, as biological things do (though idk, apples don’t). About that in my next post!
You should do some studying of modern fruit production. Most apples are man-made hybrids, and most of them don't actually grow from seed - they're propagated by clipping, or by grafting onto a different-species base. And there's a TON of human technological activity in caring for the trees, harvesting the fruit, packing and preserving it for travel, and getting it to your house.Some, of course, do grow from seed and give perfectly edible fruit. They're about as efficient, compared to commercial apples, as horses are to cars. (edit for clarity: this is hyperbole. "natural" fruit trees are less consistent and less efficient in land and water use, but not by the same degree as a horse compared to a car).So I maintain that the comments on your previous post are spot-on. We have exactly what you describe - a growable car is called a horse. Also, it's already the case that nobody knows how to make a car. There are so many steps involved that nobody can possibly understand them all.
wow... I definitely did not know we were that intense with making things artificial..
and I like that argument to draw a parallel with horses - quite convincing.
I'm really interested in the question of what's the difference between human systems and things like ecosystems? There are definitely some advantages biological systems have - antifragility, adaptability, sustainability. On the other hand, as you point out, human-designed systems are more efficient, but at a more narrow task.
So are there structural lessons we could adapt from biological system designs? Or are we good where we are?
Check out the follow up post on this