Back in the late 80s the team I was in was evolving 2D aerofoil designs. The process discovered the divergent trailing edge, but McDonnell Douglas had already patented it.
It's a very powerful tool. By the mid 90s thanks to Moore's Law and a bit of ingenuity we were applying it to whole wings including optimisation of weight and fuel volume.
Recently, I had a requirement to simulate the processes going on in a cup of soup.
I thought of making a new language "soup_lang", to simulate these interactions, to run in parallel.
It turns out, I already had the necessary hardware in my kitchen, and used water, chicken, and a few other things to introduce the necessary entropy into the calculations.
The programs (I'll call them "recipes") were not hard to write, pretty tolerant of the supplied parameters.
Love the overlap between computation and biology. This is exactly the way I've always thought of natural selection, as a search algo run at a just *spectacular* scale.
Related fun fact: the way that DNA codes for amino acids in proteins is "fault tolerant", that is, it has features like:
1. there are 64 codons (3 letter sequences) mapping to just 20 proteomic aminos, so many single-letter mutations don't change the chosen amino acid
2. when a mutation *does* change the amino, it's usually a chemically similar amino, so that the resultant protein often "sort of" works, rather than being completely non-functional
3. the position most likely to have an error (the 3rd, basically) has the least critical information
Someone studied the possible DNA -> amino mappings and found that the one life uses is very nearly mathematically optimal.
And I would be unsurprised to eventually learn that the very minor suboptimality has some hidden reason why it's actually better.
I am increasing impressed with the tiny size of today's human genetic pool, having 'processed' 5000+ pictures of naked females as I search for a partner. That's what comes to the spam folder. The sole variables are color and size. Phenotypicals are identical.
Back in the late 80s the team I was in was evolving 2D aerofoil designs. The process discovered the divergent trailing edge, but McDonnell Douglas had already patented it.
It's a very powerful tool. By the mid 90s thanks to Moore's Law and a bit of ingenuity we were applying it to whole wings including optimisation of weight and fuel volume.
Recently, I had a requirement to simulate the processes going on in a cup of soup.
I thought of making a new language "soup_lang", to simulate these interactions, to run in parallel.
It turns out, I already had the necessary hardware in my kitchen, and used water, chicken, and a few other things to introduce the necessary entropy into the calculations.
The programs (I'll call them "recipes") were not hard to write, pretty tolerant of the supplied parameters.
No Qbits were inconvenienced.
Fascinating, thank you.
Love the overlap between computation and biology. This is exactly the way I've always thought of natural selection, as a search algo run at a just *spectacular* scale.
Related fun fact: the way that DNA codes for amino acids in proteins is "fault tolerant", that is, it has features like:
1. there are 64 codons (3 letter sequences) mapping to just 20 proteomic aminos, so many single-letter mutations don't change the chosen amino acid
2. when a mutation *does* change the amino, it's usually a chemically similar amino, so that the resultant protein often "sort of" works, rather than being completely non-functional
3. the position most likely to have an error (the 3rd, basically) has the least critical information
Someone studied the possible DNA -> amino mappings and found that the one life uses is very nearly mathematically optimal.
And I would be unsurprised to eventually learn that the very minor suboptimality has some hidden reason why it's actually better.
Some other examples of provable near-perfection in life systems:
1. retinal sensory systems are near-perfect at information transfer given noise + metabolic constraints (Shannon optimality)
2. E. coli's ability to sense chemical gradients is near the Berg-Purcell limit (given diffusion + noise)
3. vascular branching follows r^3 scaling that near-perfectly balances energy lost to pumping vs the cost of maintaining vessels
I suspect that kinesin + microtubules will be proven near-perfect someday. A great PhD for someone!
I am increasing impressed with the tiny size of today's human genetic pool, having 'processed' 5000+ pictures of naked females as I search for a partner. That's what comes to the spam folder. The sole variables are color and size. Phenotypicals are identical.