The programming talent myth [LWN.net]

The programming talent myth

Posted May 4, 2015 14:54 UTC (Mon) by dskoll (subscriber, #1630)
In reply to: The programming talent myth by neilbrown
Parent article: The programming talent myth

Neil, I think you misunderstand the mechanism of natural selection. If there are no selection pressures for a certain trait, then it won't be selected for and cannot be expected to become common in the population. Your example of "side-effect" of the fact that its parent was able to outrun a completely different leopard misses the mark completely: Being able to outrun a predator has clear survival advantages, so that trait is likely to be selected for. Note that the trait is "being able to outrun a predator", not "being able to outrun Jennifer the Leopard, specifically."

When it comes to computers, I think you have it backwards. Humans evolved the ability to reason, plan and communicate (traits which indeed have significant survival advantages) so we invented the computer because that invention fit nicely with the brains that we already have as a result of natural selection.

The programming talent myth

Posted May 4, 2015 17:36 UTC (Mon) by andrel (guest, #5166) [Link] (1 responses)

In a finite size population, especially a small population, traits are expected to become common in the population solely due to genetic drift, with no selection pressure necessary.

The programming talent myth

Posted May 4, 2015 18:44 UTC (Mon) by dskoll (subscriber, #1630) [Link]

Yes, I know about genetic drift... I have a hard time believing that human brains are the way they are because of genetic drift. That seems a bit of a stretch.

The programming talent myth

Posted May 5, 2015 22:07 UTC (Tue) by nix (subscriber, #2304) [Link] (10 responses)

Quite. Note that e.g. we did not evolve the ability to do mathematics. We have exactly the same innate abilities for numerical manipulation as a vast swath of the animal kingdom: an exact counting mechanism which can instantly, reliably count up to about four or five, and another mechanism which can estimate magnitudes on an exponential scale. Both of these abilities are highly advantageous, and widespread -- even *honeybees* have them -- but there is nothing more to this ability. It can't count to ten, and it can't do more than roughly estimate the size of any large group of objects with respect to some other such group.

The ability to do mathematics comes not from this innate numerical ability but is a side-effect of a much more specifically human trait, language -- and language almost certainly did *not* evolve to enable us to do mathematics! Our ability to do mathematics is not an adaptation but an exaptation: a co-option of an existing ability into a different domain. (This, too, seems to be something humans do especially well, though it could simply be that we can't tell when other animals are doing similar feats of generalization.)

The programming talent myth

Posted May 6, 2015 3:20 UTC (Wed) by dashesy (guest, #74652) [Link]

Interesting! I think someone should do more research on dolphins then to see how good they are in math. So it is not a coincident that learning language and learning math seem to help each other.

The programming talent myth

Posted May 6, 2015 9:29 UTC (Wed) by paulj (subscriber, #341) [Link]

Funnily enough, humans actually suck at the exact-counting stuff compared to at least some of our great ape cousins. Chimpanzees utterly destroy humans at speed and accuracy in exact counting.

The programming talent myth

Posted May 7, 2015 9:22 UTC (Thu) by neilbrown (subscriber, #359) [Link] (7 responses)

> Quite. Note that e.g. we did not evolve the ability to do mathematics.

That's like saying that we didn't evolve the ability to speak English or German or any other specific language. This is obviously true if you use the word "ability" to mean "knowledge" or "skill" - a thing that can be learnt or practised and developed.

But if you use "ability" to mean something encoded in your genes, realised in your brain (and other parts of your body) - then "the ability to do mathematics" is exactly the same ability as "the ability to speak English" or "the ability to design electronic circuits". Depending on how the ability is trained and educated it will manifest in different ways, but that doesn't make it a different (innate) ability.

Evolution hasn't gifted us with specific skills - it has gifted us with a general ability to develop any skill.
We are born with a brain that knows (nearly) nothing. We can't even walk!!! Maybe we can count to 4 or 5, but we have no way to communicate that counting.
But we are born with a brain that can learn nearly anything. That is the ability that underlies all the others. That is the real gift that evolution has granted us.

Yes, the human brain does seem to have a particular affinity to language. I suspect that it because the languages we use are tuned (possibly by non-biological evolution) to the particular abilities of our brains.

As I reflect on my own thinking process (admittedly an error-prone activity) is see the two key strengths are:
1/ abstraction - the ability to recognize and identify patterns, and then associate all "things" that fit the pattern together. Very importantly, this is recursive so we build patterns of patterns.
2/ memory - both short-term (which gives a sense of time and sequencing) and long-term with parallel fuzzy-matching for very effective lookup.

Words, like numbers, are abstractions. Sentences build on these to create new abstractions. Their ideal length fits with the length of our short-term memory. The precision (or lack there-of) of words aligns with the fuzzy-matching of our memory access mechanism.

These same basic tools of abstraction and memory support "natural language", "mathematical language", computer programming, mechanical design, economics, politics, religion, war, etc etc.

I accept that the ability to solve partial differential equations or to write web-apps in Python did not lead to this ability helping my ancestors to survive, but I contend that you cannot point to any one resulting skill that did any more than any other. It was the "generality of ability", not any specific ability, that let our common ancestors dominate.

In times past there were more special-purpose VLSI designs than there are today. The general purpose CPU beat them all.... with the possible exception of image rendering, but "graphic chips" may yet turn into the "even more general purpose" processor which conquers today's general purpose processors.

*We* are the general purpose processors of the animal world. We may not have the best numerical co-processor, but we don't need one!

The programming talent myth

Posted May 9, 2015 1:02 UTC (Sat) by Wol (subscriber, #4433) [Link]

> Yes, the human brain does seem to have a particular affinity to language. I suspect that it because the languages we use are tuned (possibly by non-biological evolution) to the particular abilities of our brains.

The particular language is almost certainly non-biological evolution. Studies of bi-lingual children (ie two parents with different native languages) show that a child's preferred language correlates with abilities associated with that language.

The example I am thinking of is that French has strict rules as to where in a sentence, a leading or trailing consonant may "jump words". English couldn't care less. So when asked to listen for a given sound at the end of a word, a child who prefers English will spot it regardless of whether the consonant has jumped. A child who prefers French will be slower at spotting it if the consonant has jumped the "wrong" way. Note that there is no correlation whatsoever with whether the mother or father has that native language.

Cheers,
Wol

The programming talent myth

Posted May 9, 2015 23:13 UTC (Sat) by nix (subscriber, #2304) [Link] (5 responses)

Evolution hasn't gifted us with specific skills - it has gifted us with a general ability to develop any skill. We are born with a brain that knows (nearly) nothing. We can't even walk!!!

Ah, the blank slate hypothesis again. I don't believe it.

We are born very early in development, earlier than almost any other non-marsupial mammal: this is an unavoidable consequence of our bipedal stance, because we have to be born before our heads get too large to fit through our mothers' pelvises. Around two years pass after birth before large-scale neuron growth and streaming in the brain ends. Discount that, and, well, does a two-year-old know nearly nothing? Hardly.

And we almost certainly don't have a general ability to develop any skill, not even any cognitive skill, though it is hard to be sure of this because there are no entities known with superhuman cognitive abilities to give us an idea of what we might be missing: but there are certainly a lot of things that present-day computers can do that we can't. You can add memory to a computer, but can you learn to expand your short-term memory to 500 chunks? (No, you can only increase the average size of a chunk over time: we call that process "expertise"). Can you do learn to do mental arithmetic with the flawless, error-free efficiency, let alone speed, of a computer? No. (Human 'computers' had a far higher error rate than the ones we're used to). The list goes on.

Yes, the human brain does seem to have a particular affinity to language. I suspect that it because the languages we use are tuned (possibly by non-biological evolution) to the particular abilities of our brains.

If that were all there was to it, pigs, tortoises, and pigeons should have language. They don't (and we can be sure of that, even without knowing what the putative language might be: the informational complexity of their communications is too low).

It is known that humans have neural machinery specialized for the parsing and production of language (more specifically, for, roughly, lexemes and grammar). If we had brains optimized for learning any skill, and were using that ability to learn language, it would seem odd that the same brain regions serve to handle linguistic ability in all of us: but they do.

While it is certainly true that human language is optimized to be learned by human children, this is a selection effect: human children are the rootstock of language, almost all languages are predominantly learned by them, and almost all linguistic change comes from them. And they don't learn language in anything like the same way adults do: they can't, they don't have a language to start from. The process of learning a first language is frankly astonishing, and if it were the result of conscious deliberation -- of an 'ability to develop any skill', and in humans development of skills is the result of repetition, pattern-matching, and conscious effort -- humans would all be high-ranking geniuses.

Your just-past-newborn infants, just lying there with a brain that knows nearly nothing, not even capable of moving by dragging themselves around with their arms, are parsing phonemes out of the surrounding sound stream (also distinguishing human voices from background noise), identifying phonemes that are considered indistinguishable in the surrounding speakers' languages and equating them in their own auditory cortices (this is largely complete by the age of six months) including clustering them so that multiple languages with different such rules are correctly differentiated, then identifying and extracting words from the sound stream despite the fact that they don't know what any of the phonemes mean and the sound stream is, well, a stream, with little or no separation between the words they don't know anyway. Only then can they get down to the hard parts of grammar and vocabulary. The amazing thing to me is that they've not only done all this but also apparently derived most of the grammar of the language by perhaps the age of one and a half, the 'two-word' stage: those two words are far more often than chance the two most relevant words, in the correct order, that would be chosen from a larger grammatically correct sentence which would express what they want to say! To me that says they're thinking that sentence (or at least have its parse tree in mind, perhaps with a lot of the words missing) but just can't speak all of it yet. And I haven't even got into the grammatical parts of language acquisition, which have so many incredible feats all being accomplished simultaneously that I could go on about it all day.

Conscious thought can't begin to come close to doing this. Among other things, infants clearly aren't thinking consciously about this: if they were, they'd have no time to do anything else! It takes extensive training in field linguistics for a human adult to do this sort of thing: it's extremely hard and they are much less good at it than human infants, taking many times longer to do a much worse job (though it is true that they know what they have done, while your average six-year-old knows nothing of the native grammar she uses so effortlessly). All this stuff just screams unconscious parallel processing, and when it's happening in essentially all newborn infants we can be reasonably certain that that means that this is genetically determined. We have hardware for learning human languages. (Well, actually, we probably don't, not any more, but we did have it when we were much younger. There is a critical time window for these things, and we are both long past it.)

As I understand it, the things we are good at that other animals aren't are language, speech production (obviously tied to language), fine manipulation (i.e. toolmaking: many animals can do this to some extent, including some of our close relatives), long-distance endurance running, and things derived from those skills. That seems to me to be it for special human wossnames. It's just that language and speech production are hugely useful abilities: in the last few thousand years we've been combining them with fine manipulation (writing) to communicate beyond the bounds of a human lifetime, which is frankly amazing. If we've got that, I don't see that we need a universal learning machine in our heads that can develop any skill, if such a thing can even exist. Linguistic processing gives us everything we need.

If I'm right in this, there may well be lots of skills we intrinsically can't develop. The place to look for them is in those places that don't require language, or that require language plus some innate thing we don't have (gills, perhaps, or a lateral line sense, or the ability to detect and manipulate others' muscles from a distance like an electric eel).

The programming talent myth

Posted May 10, 2015 9:45 UTC (Sun) by paulj (subscriber, #341) [Link]

Can you do learn to do mental arithmetic with the flawless, error-free efficiency, let alone speed, of a computer? No.

Hell, we can't even learn to count as fast and accurately as a chimpanzee can! Same architecture as us!

The programming talent myth

Posted May 11, 2015 6:03 UTC (Mon) by neilbrown (subscriber, #359) [Link] (3 responses)

Wow. Thanks for your thoughtful reply!

> Ah, the blank slate hypothesis again.

"blank" is more extreme than I would claim - I probably should not have hidden those "nearly"s in parentheses. We still have vestigial instinct such as fight-or-flight. But I maintain that we have much more potential ability than practical ability.

> We are born very early in development, earlier than almost any other non-marsupial mammal

Good point. I hadn't thought about that. I'll withdraw the "can't even walk" quip.

> Discount that, and, well, does a two-year-old know nearly nothing? Hardly.

Is the knowledge of a 2 year old nature or nurture? Doubtlessly both. We probably lean in different directions on the question of proportion.

> but there are certainly a lot of things that present-day computers can do that we can't

Two responses to this:
1/ A thing can be very large and still have limits. I want to emphasize how large the range of possible learned abilities is. You respond by saying there are limits. I agree there are limits, I still think the range of learnable abilities is larger than we know.

2/ But we *did* do those things. We (humanity, not you and me personally) built the tools that perform the calculations. Our brains are so versatile, they allow us to do things that our brains cannot do themselves.

> Your just-past-newborn infants ... are parsing phonemes out of the surrounding sound stream

Amazing, isn't it?!?! Processing incoming sensory information, finding patterns, and patterns in patterns, and pattern in patterns in patterns. This is fundamental to how we think and learn and it starts very early.

> To me that says they're thinking that sentence (or at least have its parse tree in mind, perhaps with a lot of the words missing) but just can't speak all of it yet.

To me, that says that several of the important concepts are forming and differentiating from other concepts. Most of it - ideas, label, structure - is still fuzzy, but a few bits are clear enough to be worth reproducing and to be recognised when they are produced. Not sure if that is the same as what you are thinking.

> Conscious thought can't begin to come close to doing this.

Oh, absolutely. Conscious thought is only one little part of thinking. Conscious thought requires (thinking aloud here...) language. It requires those labels to apply to concepts so we can store them more efficiently and manipulate them more precisely.

Other thinking still happens, but it is harder to watch or to talk about. We call it intuition or hunches or "gut feelings" or taste etc etc. Such thinking is very important but it is also very unreliable. It is very hard to introspect our hunch to check the justification.

I think many of us tend to suppress our intuition for that reason. Finding a good balance where we don't suppress our hunches, but don't trusted them too much either, is not easy.

> Linguistic processing gives us everything we need.

I wonder what exactly you mean by "linguistic processing". Earlier you seemed to differentiate mathematics from language. If your definition of "linguistic processing" maintains that differentiation, that I think there is much more to how I, at least, think and the practical value that I get from my brain than just "linguistic processing".
Even if you do include mathematics, there is the whole realm of "subconscious thought" which doesn't seem in any way linguistic yet still brings value to the individual and the species.

> plus some innate thing we don't have (gills, perhaps, or a lateral line sense,

A question for philosophers: can a blind person learn to recognize "blue"?
Obviously they never will, but would their brain be able if the sensory input was there?

There is no doubt that the range and acuity of our sensory input is critical to what we can learn about our environment - you cannot work with that you cannot perceive. That doesn't mean that the brain does not have the ability to learn some skill, just that it has not the opportunity.

I recall at University talking to a Pure Mathematics professor who claimed that felt he had a (limited) ability to "think" in 4 dimensions, because that was related to his area of interest and research and with ongoing practice he found the manipulations easier - less linguistic, more intuitive. As humans we have incredible imagination. Sometimes it can provide input to our brain beyond physical senses, and give us the opportunity gain mental abilities that are "out of this world". Sometimes there are more things "dreamt of in your philosophy" than there are "in heaven and earth, Horatio"

The programming talent myth

Posted May 11, 2015 15:05 UTC (Mon) by nix (subscriber, #2304) [Link] (2 responses)

> Wow. Thanks for your thoughtful reply!

I wrote it and then thought 'oh crap, this is both offtopic and pointless rambling and also teaching Neil to suck eggs'. Nice to see I was being excessively paranoid!

> We still have vestigial instinct such as fight-or-flight.

I'm firmly of the believe that 'instinct' is a nearly meaningless category. All our neurons are placed into their final resting places by a feedback loop between genes and cellular microenvironment: many of them also include a feedback loop between cell and *external* environment (including those involved in things you'd think didn't, such as vision). Mammals have almost protean developmental pathways. Some of the behaviours encoded by those neurons are not very variable after they are established (e.g. heartbeat regulation). Others are more variable. It's a continuous gradation, not a division into enumerable "instincts" and "everything else". Elephants have to learn how to use their trunks with delicacy from other elephants: does that mean it's not an instinct? Even if it does, that surely doesn't mean we could learn to do the same thing!

> Is the knowledge of a 2 year old nature or nurture? Doubtlessly both.

The division is meaningless. The neural architecture of an unborn child six months in the womb is both nature and nurture (the neurons don't develop without environmental input, some internally generated and some from things like sound leaking through the womb wall). The architecture of a week-from-conception female child is both nature and nurture (which X is inactivated in which cell? What DNA methylation patterns are established? Hugely significant for all of future development, and more or less random with some environmental influence.)

Asking 'nature or nurture' is like saying 'this computer has really good hardware! Therefore its software must be pathetic, because its hardware does everything.' In reality, having more hardware often means the software too must be more complex: in biology, being a more complex organism often means that your relationship with your environment, and the regulatory networks it controls, just get more complex. We are a long way from getting to the bottom of this. Hell, in the last few years entire new *classes* of regulatory network have been uncovered...

Clearly some things (e.g. your knowledge of your mother's name) are strongly environmentally influenced, but even they will be the result of environment acting on neurons, changing their gene expression, and feeding back. When all of memory is like this, and so are most things that aren't memory, 'nature or nurture' really means very little. The most you can ask is 'is this behaviour variable across the population, and does it vary widely in closely related individuals' -- but that doesn't mean it's 'nurture not nature', it means it's nature with complex environmental responses.

> 2/ But we *did* do those things. We (humanity, not you and me personally) built the tools that perform the calculations. Our brains are so versatile, they allow us to do things that our brains cannot do themselves.

In that case it's impossible to ask whether there are things we cannot learn: if we encounter such a thing, we'll never know it's even there since we cannot learn of its existence. Therefore you have taken the question of the range of human mental capacity outside the realm of science.

It is still clear that we can't learn *anything*: there are lots of unsolved questions in mathematics that we don't know the answer to because we have no way to work it out, or because working it out is intractable. Much of the last century of science (computer science in particular) has been the story of the discovery of new limits: I strongly suspect that such limits exist in the human mind as well, simply because there is no reason for evolution to have selected for a universal thinking machine when something simpler and likely less energy-hungry would do (energy expenditure and heat dissipation are *major* constraints on human evolution, and the brain is a huge energy suck). Why would a generalist hunter/gatherer/scavenger on the African savannah evolve a brain capable of literally anything?

(It is definitely clear that in the domain of language we do a lot via metaphors and the like, often frozen into customary forms by centuries of use to such a degree that we don't even recognise them as metaphors any more. To me, this says that we *do* have cognitive limits, but that language at least sometimes helps us end-run around them. Even there, we can't handle everything: there are a lot of not especially complex grammatically valid linguistic structures that we just can't parse because of apparent arbitrary limits. A universal thinking machine would not have a limit of at most three on the number of centre-embeddings it could handle!)

> Processing incoming sensory information, finding patterns, and patterns in patterns, and pattern in patterns in patterns. This is fundamental to how we think and learn and it starts very early.

Except we *can't do that* for anything but language, and we can't do that conciously. We find patterns, yes, but in a conscious, linear fashion: we can't listen to a massive blur of sound and do all this extraction work without thinking about it except when a very young child in the critical language-learning window. This is hardwired behaviour -- or, in the form I describe above, neurons predetermined to respond to environmental input (usually auditory) by doing this massive set of pattern-matches.

> I wonder what exactly you mean by "linguistic processing". Earlier you seemed to differentiate mathematics from language. If your definition of "linguistic processing" maintains that differentiation, that I think there is much more to how I, at least, think and the practical value that I get from my brain than just "linguistic processing".
> Even if you do include mathematics, there is the whole realm of "subconscious thought" which doesn't seem in any way linguistic yet still brings value to the individual and the species.

Mathematics has language as a prerequisite, if you will: its foundation is the concept of a number line which is not only geometric but has numbers spaced regularly along it. Those numbers are linguistic entities, and in languages spoken by cultures with advanced mathematics are recursive combinatorial entities obeying a grammar (you can build a word for any number you like). Whether mathematics has *other* prerequisites is hard to say: no other animal has this prerequisite, and as far as we know no other animal has anything like the field of mathematics. (As opposed to, say, the sort of unconscious stuff you do when catching a ball, which is mathematics *implemented by neurons* rather than mathematics you think about. Almost everything alive has that sort of mathematics encoded in it somewhere, even bacteria, but that doesn't mean they can do mathematics in the sense we mean here.)

> A question for philosophers: can a blind person learn to recognize "blue"?
> Obviously they never will, but would their brain be able if the sensory input was there?

This has been tested in people blind from birth due to cataracts: yes they can, but they don't seem to have good outcomes due to just not being able to handle the flood of visual information as easily as we can: they can see, but it's difficult to make sense of the scene as a whole and they get overwhelmed very easily. It seems likely that a lot of the visual cortex either hasn't developed properly in such people, has been excessively trimmed due to inactivity or has been repurposed for auditory purposes (very common in blind people, as well as deaf people the other way: early parts of the auditory and visual signal processing pathways use very similar algorithms and very similar patterns of neural organization, so are easy to repurpose).

Again here we see the faultiness of the term 'instinct'. Visual processing is a vastly complex pile of neural machinery carrying out at least hundreds and quite possibly thousands of distinct tasks, with all sorts of environmental feedbacks in its development (from autogenerated test patterns on the retina in the womb, through to matching of corresponding points on the retinae of both eyes for stereo-vision development in a critical window in infancy, and doubtless more that is not yet known). It is reused by all sorts of things you might not expect to use it (e.g. remembering a visual scene and visual imagination both feed into the visual cortex 'backwards', and you can even guess at what's being imagined if you spy on the right parts of the cortex; if your visual cortex is lost, you can no longer remember what colours are or what things looked like). Things that were believed to be largely hardwired, like synaesthesia, have recently been discovered to be environmentally influenced: some people with colour/letter synaesthesia have colour/letter correspondences matching those in a particular Fisher-Price plastic letter set. Even conscious thought ripples down into the early visual cortex down feedback loops: this has been tested with the Necker cube, where you can lean to flip your perception of the cube by thinking about it, and when you do about 10% of the neurons in the earliest regions of the visual cortex flip with it.

So, is vision an instinct, given that a lot of it is hardwired? Is it thousands of instincts? Is it learned, given that you can easily break quite a lot of it through environmental deprivation, that things like synaesthesia are influenced by the colours of plastic letter sets, and that conscious thought can directly affect the activity of neurons in even the most fundamental parts of the visual cortex? It is both. It is neither. It is more complex than that. (This is biology. *Everything* is "more complex than that". There's a reason a lot of old-time hackers have gone into biology now -- it satisfies their desire for crazy levels of complexity! :) )

The programming talent myth

Posted May 16, 2015 4:11 UTC (Sat) by neilbrown (subscriber, #359) [Link] (1 responses)

There are a few points in your very interesting treatise that I'm tempted to drill down into, but I should probably go off and do my own reading instead.

The thing that particularly stands out for me (and which I think I can make relevant) is that you present a highly reductionistic view. You focus on the details and deny (some aspects of) the big picture (there is no beauty in the Mona Lisa, only brush strokes).

I certainly agree there is no bright line between "instinctive" and "learnt" behaviours, but nor do I find the distinction completely useless. There is a continuum, and by naming the endpoints it makes it easier to talk about them.

I seem to think of brain activity at a more abstract level than you present it. This is neither right nor wrong, just different. One of the (I claim) particular abilities of the human brain is to work at different levels of abstraction. I do a little volunteer teaching and "youth group" type work. Primary school children (up to about 12 years) are *very* concrete thinkers. I try to present abstract ideas and I'm lucky when one gets through. Concrete stories are easily remembered.
Just a couple of years later, they are making the leaps to abstract implications themselves. But going beyond two or three levels seems to take focused training. Dan Myers and others talk a lot about the "abstraction ladder" which is particularly helpful for maths. http://blog.mrmeyer.com/2012/loa-the-ladder-of-abstractio...

I think that a key part of competence with software (that's where we started, isn't it?) is the ability to move up and down the abstraction levels comfortable. To know that the brain is about chemicals and about neurons and about perceptions and about ideas and about paradigms. Or to be able to think about cachelines and stack foot print, and about interface design and maintainability, and everything between and outside these points.

I believe that I was trained in this from an early age by my father who loved to play with words and mis-interpret things whenever possible. In doing so he taught me the difference between "words" and "meaning" and importantly that there *is* a difference. Each level of abstraction is valuable, but it is important to be able to tell them apart and not believe that any one is "right".

Edsger Dijkstra touched on this 27 years ago:

>By evoking the need for deep conceptual hierarchies, the automatic computer confronts us with a radically new intellectual challenge that has no precedent in our history.

https://www.cs.utexas.edu/~EWD/transcriptions/EWD10xx/EWD...

Still very relevant.

The programming talent myth

Posted May 20, 2015 10:55 UTC (Wed) by nix (subscriber, #2304) [Link]

The thing that particularly stands out for me (and which I think I can make relevant) is that you present a highly reductionistic view. You focus on the details and deny (some aspects of) the big picture (there is no beauty in the Mona Lisa, only brush strokes).

There's plenty of beauty in the Mona Lisa, but we're such a long way from understanding the neural correlates of 'beauty' that it's hard to talk about in an objective sense.

One of the (I claim) particular abilities of the human brain is to work at different levels of abstraction.

Definitely, and this also appears to be a specifically human faculty: "integrative intelligence", if you will. (Some corvids and parrots might have something like it too. No other great ape seems to. I don't know about cetaceans.)

I think that a key part of competence with software (that's where we started, isn't it?) is the ability to move up and down the abstraction levels comfortable.

Of course, a lot of programmers can't do it. Does that mean they're not human, or just not conscious entities? (Thinking of some I have known in the past, I'd believe the 'not conscious entities' part in an instant :P ). Probably neither, which likely means that in most humans this ability is something used in childhood and then discarded when large-scale learning stops.

By evoking the need for deep conceptual hierarchies, the automatic computer confronts us with a radically new intellectual challenge that has no precedent in our history.

Mathematics was doing that long before, but of course software development is a mass phenomenon in a way that mathematics never was. There are many millions of software developers now, most fairly new to the field: there have never been anywhere near that many mathematicians.