We now know that human biochemistry is the result of an evolutionary process which when faced with a problem consistently comes up with the worst possible solution that is still workable. It's 3 billion years worth of crufty spaghetti code, literally.
As someone educated in theoretical biology and learning theory, I object to this kind of characterization of the evolutionary process. There is no evidence for a superior magical learning algorithm that would have done better-- see the no free lunch theorem. This characterization also lacks respect for the difficulty of many-dimensional optimization problems on non-fixed fitness landscapes, which is what evolution has to deal with. (Mathematically, the problem tackled by natural evolution is effectively infinite-dimensional. By comparison, the problems tackled by human engineers are toy problems.)
It would be more accurate to say that evolution, when faced with a problem, finds a working solution. There is no way to know how close or how far that working solution is from a theoretical optimum, since the theoretical optimum is non-computable. (If it were straightforwardly computable, evolution would find a gradient to ascend and discover it rapidly.)
"Life doesn't work perfectly. It just works."
That being said, it is likely that our mortality represents a hard compromise between the survival value of longevity, adaptations to prevent cancer (many of which have aging side effects, like telomeres), and the species-scale (or selfish-gene-scale) survival value of getting oldsters out of the way to make room for the next generation.
The fact that humans can live substantially longer than is merely necessary to reproduce is due to the fact that we're a K-selected species rather than an R-selected species. (Cicadas, for example, are an R-selected species. Dolphins, Elephants, and Humans are K-selected.)
There is no evidence for a superior magical learning algorithm that would have done better-- see the no free lunch theorem.
There's plenty of evidence that evolution is extremely inefficient. Due to its very nature it can't look forward, it can only hill-climb. Since evolution can't re-architect things, it gets stuck in local maxima all the time.
We can be reasonably sure of evolution's inefficiencies after looking at some some examples. Problems like the laryngeal nerve and backwards vertebrate retina are commonly brought up in these sorts of debates, but they forget that evolution doesn't even look at the vast majority of solution-space. It limits itself to squishy things in the discipline of biology. For example, neurons run at 10-200Hz and conduct signals at 0.000001c. Other substrates like silicon and diamond are much faster.
A human visual cortex has billions of neurons doing complicated procedures (edge-detection, FFTs, motion-detection, etc). But only a tiny fraction of humans can multiply two 4-digit numbers without the aid of pen and paper. Had the human brain been designed by a human, every programmer in the world would scream, "Why didn't you give us an API you idiot?!"
Of course, we'd also complain about the brain's lack of 4G, since evolution never invented a radio.
I have no authority on the subject, but, the fact that we cannot prove nor properly evaluate solutions that emerge from evolution does not change the fact that the first solution that fits is the one that will be kept. Intuitive reasoning that not work on such a subject, but still, I have hard time to accept that it is the best. The only proper qualifier, as you state, is that it is the existing one.
The same way I am not comfortable with using dynamic verbs when describing evolution. Evolution does not find a solution. It's just that a solution that fits emerges. After producing many dysfunctional living individuals. It's still the only way life can evolve by itself, and I wouldn't dare to say if it is a laborious or great way to sustain ifself.
The first solution that fits isn't necessarily kept, unless it can be refined to a high enough level of development to compete with other emerging solutions. The eye, for example, evolved separately many times.
BTW, in human engineering the first solution is also often kept. Look at nuclear power. We probably should be using the thorium cycle, but we kept light water reactors because they were the first big reactor commercialized (for military reasons). Another example would be x86 architecture. I am not arguing that evolution is magical in any way, just that human engineering doesn't show signs of being much better in many cases.
I see a lot of engineers who are ignorant of biology claiming that "evolution is slow" and that we should be able to snap our fingers and do better. It's a popular point of view among the singularity crowd.
Engineers are trained to be arrogant about their abilities. This is probably a good thing, since it causes them to fling themselves at problems fearlessly. But it's also not necessarily realistic.
That is the kind of thing we mean when we say evolution doesn't find the best solutions all the time. The horrible job evolution did with the human eye is another example.
Now:
> There is no evidence for a superior magical learning algorithm that would have done better
I think the point is that it's difficult to imagine anything doing worse.
As someone educated in theoretical biology and learning theory, I object to this kind of characterization of the evolutionary process. There is no evidence for a superior magical learning algorithm that would have done better-- see the no free lunch theorem. This characterization also lacks respect for the difficulty of many-dimensional optimization problems on non-fixed fitness landscapes, which is what evolution has to deal with. (Mathematically, the problem tackled by natural evolution is effectively infinite-dimensional. By comparison, the problems tackled by human engineers are toy problems.)
It would be more accurate to say that evolution, when faced with a problem, finds a working solution. There is no way to know how close or how far that working solution is from a theoretical optimum, since the theoretical optimum is non-computable. (If it were straightforwardly computable, evolution would find a gradient to ascend and discover it rapidly.)
"Life doesn't work perfectly. It just works."
That being said, it is likely that our mortality represents a hard compromise between the survival value of longevity, adaptations to prevent cancer (many of which have aging side effects, like telomeres), and the species-scale (or selfish-gene-scale) survival value of getting oldsters out of the way to make room for the next generation.
The fact that humans can live substantially longer than is merely necessary to reproduce is due to the fact that we're a K-selected species rather than an R-selected species. (Cicadas, for example, are an R-selected species. Dolphins, Elephants, and Humans are K-selected.)