Readit News logoReadit News
avar · a year ago
This article doesn't even try to address what I feel is the deeper and more interesting question (but probably one that can't be answered): Why is it that horses, cows, giraffes and birds have all had to come up with a purely passive solution of "locking" themselves in place, either via their joints (for the four-legged), or via the tendon mechanism described here for birds?

I.e. why wasn't in simpler in evolutionary terms to come up with some mechanism where 1% of the brain was dedicated to the relatively simple task of "station keeping", while the rest of the brain could benefit from sleep?

Agentus · a year ago
Evolution is like DFS depth first search, it isnt looking for an optimal way optimally its just going down a branch until it finds a satisfactory solution.

Sleep according to sleep scientist matthew walker, isnt what it seems. If i remember correctly theres even more brain activity and it serves certain goals like behavior refinement, among other things not just refreshing yourself. There might not be more benefit by more sleep.

BobaFloutist · a year ago
My simple conception is that evolution loves local maxima.
fennecbutt · 10 months ago
Exactly, it takes more time and energy to undo a mutation that unblocks a better solution than it is to mutate a solution that works with existing mutations.

It's why intelligent machines will absolutely destroy us when it comes to being "superior".

0xdeadbeefbabe · a year ago
It never left the "does need to sleep" branch. It's not much a sci fi writer.
danans · a year ago
> I.e. why wasn't in simpler in evolutionary terms to come up with some mechanism where 1% of the brain was dedicated to the relatively simple task of "station keeping",

Supposedly this is how dolphins sleep, shutting off part of the brain and using the other half to swim.

https://us.whales.org/whales-dolphins/how-do-dolphins-sleep/....

HarHarVeryFunny · a year ago
I think all aquatic mammals do this - they don't have much choice since they need to surface to breathe, so need to be semi-awake all the time.
roywashere · a year ago
Just as we can still breathe and digest food while sleeping!
mr_mitm · a year ago
We can even adjust our sleeping position while sleeping, possibly to avoid bedsores? But perhaps avoiding falling requires more processing power.
johnisgood · a year ago
Concurrency at its finest I say!
jahnu · a year ago
Does locking require any energy to maintain? I suspect very little if any.
meindnoch · a year ago
Also, why didn't any animal evolve a way to avoid sleep completely?
yann63 · a year ago
There's a species of bird (Chinstrap penguin) which kinda does that: it sleeps by intervals of 4 (four) seconds only. Many times through day & night. Can these 4 seconds naps be considered sleep? I don't know, but it goes along your question. Source: https://www.nationalgeographic.fr/animaux/insolite-decouvert...
LoveMortuus · a year ago
I thought that dolphins could avoid having to sleep while travelling long distances by having part of the brain asleep and then switching when tired.

I could just be misremembering things, so I’m not certain.

anvil-on-my-toe · a year ago
Better to whole-ass one thing than half-ass two things. Sleep allows a focused effort on cellular repair, garbage collection, memory consolidation, and learning.

The 24-hour dark-light cycle on Earth is the most energetically significant thing that happens here, and species capitalize on the parts of the day that play to their strengths.

nonameiguess · a year ago
Cnidarians and flatworms don't have a CNS and sponges don't have nerve tissue at all. I would think that means they don't do anything on a daily cycle that would match what we commonly call sleep.

I'd also wonder about creatures inhabiting extremely sparse environments like deep caves and undersea trenches, that spend most of their lives dormant and only seem to do anything at all when food happens to come around, which might only happen every few weeks or even every few years. They're probably not studied extensively enough to answer, but do they have anything like a sleep/wake cycle?

dystroy · a year ago
We don't know exactly why nature can't do otherwise, but any complex brain has to sleep.

Some animals, mainly sharks which can't stop or they would be asphyxied, deal with that by having kind of 2 brains which are never both sleeping at the same time.

interludead · a year ago
Sleep has crucial role in survival and well-being
michael1999 · a year ago
Because movement isn't 1% of the brain. Movement is why we have brains. Balance and head-righting do sensor fusion between vision, the vestibular system, and proprioception. That's a whole-body problem, and I don't see how that can be a low-energy activity. Brains are expensive.

We do have an unconscious falling reflex, but that works by startling us awake! Actually doing movement planning requires a running brain.

VoodooJuJu · a year ago
Because evolution isn't a nerd optimizing a design. Design implies intent. Evolution is more random and serendipitous than a nerd with any sort of intent. It's not engineering. There's no specs & recs or post mortems or blueprints. Evolution is too beautiful, random, and mysterious for the enginerd to appreciate and understand.

Deleted Comment

123pie123 · a year ago
may guess it's not as simple as we think it is

or it has evolved somewhere and we either don't know about it or the trait didn't survive

weinzierl · a year ago
"The mechanism’s passive character is surprising"

Are they saying that birds are in a stable equilibrium while standing? I always assumed that some low level activity in the Cerebellum would actively stabilize them even during sleep.

If this is not the case, a dead parrot should be perfectly able to stand, right?

ahoka · a year ago
Only if it's a Norwegian Blue, as far as I know.
cheschire · a year ago
Though in that particular case it may in fact be the nails in its feet causing the phenomenon.
louthy · a year ago
Lovely plumage

Deleted Comment

krisoft · a year ago
The article is clearly about why standing birds don't fall while they sleep.

An interesting and somewhat related fact is that flying birds also sometimes sleep: https://www.audubon.org/news/scientists-finally-have-evidenc...

The short story (at least in the case of the frigatebirds they studied) is that they do hemispherical sleeping, where only one half of their brain is asleep. And also it seems they only sleep while climbing in a thermal updraft. The theory being that "gliding up" is the safest portion of their flight.

Relatadely there is a famous videoclip of a paraglider colliding with an eagle mid-flight. https://youtu.be/g8_hyqlQXpk?si=alAvnt4Xva4RGYw3 (The clip is quite scary, but both the pilot and the eagle seems to be fine at the end.)

It is speculated that the incident happened because the bird was "dozing off" mid-flight in the updraft too. Of course this is difficult to verify.

willglynn · a year ago
The document to which this article refers was published in the Journal of the Royal Society Interface, and the article links there. It is also available as open access, which was not linked:

https://hal.science/hal-04287433v1

https://hal.science/hal-04287433/file/Version%20HALL.pdf

m463 · a year ago
When I read "tensegrity" I thought of these strange tables/scupltures you can buy:

https://www.amazon.com/s?k=tensegrity

dystroy · a year ago
If you look at the article, you'll see the picture of such sculpture (a rather impressive one, by the way).
m463 · a year ago
I didn't see that, are you talking about the bird's leg?
FiatLuxDave · a year ago
If that is what "tensegrity" makes you think of, I'm guessing you haven't seen this:

https://www.youtube.com/watch?v=hkzeE6BVNIk [Super Ball Bot, NASA]

prmoustache · a year ago
A good example are bicycle wheels made with ropes but if you think of it even of steel spokes as a single spoke by itself cannot maintain the structure and it all comes together when the spokes are tensionned. https://berdspokes.com/collections/berd-wheels

Obviously a wheel is the easiest design you can build based on tension given its symmetric nature.

Crazyontap · a year ago
When I was younger, I was fascinated by evolution, especially the intricacies of how things just work. This fascination also explains why many people believe in the intelligent design theory.

However, witnessing the rapid evolution of AI with just a few hundred GPUs, enough data, and power, I no longer wonder what a billion years of feedback loops and randomness can achieve.

TrainedMonkey · a year ago
AFAIK key insight into evolution is not randomness but rather sheer amount of compute. Specifically, evolution is a massively parallel flood algorithm that will try every single mutation. Barely any of them will have positive impact on organism fitness, but some will.
TeMPOraL · a year ago
That, and of course the other key insight is the "flood algorithm" part. I.e. evolution isn't about randomness (the "throw some parts into a bag and spin them until a 747 flies out" criticism), it's about bias and feedback: the environment itself isn't uniform, creating a bias in what would otherwise be entirely uniform selection, one which compounds with every generation. Randomness is just adding variance here, jitter preventing the process from getting stuck with one outcome.
SamPatt · a year ago
Also worth pointing out that billions of years already sounds like a long time to humans, but once you grasp how quickly everything is moving at the cellular or molecular level then it becomes a really reallllllly long time.
raffraffraff · a year ago
The way I understood evolution wasn't that "some mutations will have a positive impact", it's more like, when a species hits hard times, "some mutations allow it to survive long enough to reproduce".

Sure, you have dominant genes like eye colour. But evolutionary changes to a whole species are more about weeding out genes that cannot survive, right? Because if a species has no specific sexual selectors for breeding and all mutations survive and reproduce, then how does a specific gene thrive?

Edit: but chatting to my wife, she mentioned that species is a difficult concept, generally taken to be a generic group that actual mate in the wild. There are several different species that could technically produce offspring but through sexual selection, do not.

JKCalhoun · a year ago
I remember early "computer recreations" of life that seemed to suggest that unbounded randomness (mutation) was, as you mention, more often bad than good. Sexual reproduction, where genes are swapped (perhaps at random?) got you to the head of the class much, much quicker.
eru · a year ago
The key insight for both evolution and contemporary AI is that hill-climbing (either completely random a la evolution, or guided locally like in back-propagation) can work really well, if you have enough dimensions to play with.

Hill climbing obviously gets stuck easily in 2 or 3 dimensions. So our intuition ain't reliable.

carlmr · a year ago
>Hill climbing obviously gets stuck easily in 2 or 3 dimensions. So our intuition ain't reliable.

Also our intuition about overfitting from lower-dimensional representations seems to be less of an issue at high dimensions.

kortilla · a year ago
AI isn’t being trained on random though. It’s the corpus of a large portion of all of humanity’s written communication. I don’t think it’s a good analogy to evolution.

A single training session will iterate more than the number of generations of all birds.

tsimionescu · a year ago
> A single training session will iterate more than the number of generations of all birds.

But that's not the right analogy. Evolution happens at the individual level, and even to some extent at the individual gamete level. So it's actually every single fecundated bird egg that ever existed, and even every single spermatozoon and egg cell every time two birds mated. Not to mention every division of every bacterial cell in every bird gut, since microflora are a key part of the organism too.

And even this is an undercount, since the DNA and gene expression of an individual actually changes during its lifetime, and those changes can be passed down to offspring through various mechanisms. So there is a constant process of evolution that even all cells inside a living organism go through, that we're still trying to fully understand.

fennecbutt · 10 months ago
I would instead think of the llm training corpus as being equivalent to the physical laws that govern our reality.

LLM training is training an organism to "survive" in an environment consisting of languages/lexicons.

Not getting eaten by a tiger is equivalent to being able to produce a semantically and logically correct sentence.

zaptrem · a year ago
AI gets backpropagation whereas evolution is more like particle swarm optimization (pick a bunch of random values, then pick more random values near the ones that do best). Backpropagation is way better/faster since in expectation the gradient points toward a better set of weights, but it relies on a differentiable loss function.
openrisk · a year ago
Strictly speaking the two domains have very little in common besides "evolving" in a general sense (as opposed to something being static and unchanging). But if we generalize a bit our target system we can make the analogy more fruitful.

LLM Algorithms don't "evolve" when trained, they just fit data in a pre-existing and hardwired "DNA". More GPU, data, energy consumption etc. simply means different weights (parameters) for the same fixed algorithm. Training involves no feedback loop on the algorithm design itself. The biological analogy is like what happens when you starve or overfeed somebody: They become skinny or obese (but they will not pass on that attribute to their offspring).

The algorithm's DNA is explicitly designed and put in place by human intelligence. When thinking of the observed "evolution" of algorithms we need to include the sum total of the people involved in algorithmic design and deployment, their cognitive toolkit, incentives etc. Now that part is definitely "evolving" (various mathematical, technical or economic breakthroughs), not biologically of course, but culturally.

So-called AI Winters (and other AI Seasons) are indeed evidence of this collective cultural movement. You could say that the invention / adoption of the multilayered neural net pattern has led to a sort of Cambrian explosion similar to going from single cell to multicellular organisms.

wruza · a year ago
what a billion years of feedback loops and randomness can achieve

Half a billion, afaik. And also how technology matters. Oxygenless ~3.5B years were boring as hell.

For those unaware: Earth not always had O2 in its athmosphere - O2 is a result of one kind of ubiquitous goo that was emitting it as a byproduct. And while it oxidized what it could on the surface, it killed almost every goo, cause it is a poison. Then (and that is still an unproven theory) oxygen-breathing evolved and allowed for fast movement and carnivory, which started arms race of tracking and evasion, coordination, vision, swarming, etc. 0.5B years later here we are.

Btw, there were even higher peaks of oxygen concentration that allowed for animal-sized insects to exist. A time insectophobes wouldn’t want to live in.

Scientists know that dragonflies with wing spans as wide as a hawk's and cockroaches big enough to take on house cats lived during the Paleozoic era (245-570 million years ago). At the same time, mammoth millipedes longer than a human leg skittered across prehistoric soil.

mietek · a year ago
Big millipedes are still around! Although not quite as big as before.

https://youtu.be/oY32HPQrhYg?si=z0QO821u3-E5sJ6l&t=480 (about millipedes from 8:00)

utkarsh858 · a year ago
Rapid evolution of AI needs a director, a human training and guiding it to get tangible results.
fennecbutt · 10 months ago
Isn't our evolution directed by both the (fixed) laws of physics and by the (random) efforts of all other life forms on this planet?
bamboozled · a year ago
Which is built on all this "billions of years of randomness" to begin with.
cryptoz · a year ago
For now.
akomtu · a year ago
Where's randomness? AI is an intelligently designed algorithm trained to mimic words of highly intelligent species, and all that runs on GPUs that didn't evolve from a pile of mud, but were intelligently created according to a plan.
spiderfarmer · a year ago
The commenter is wrong. There is not a single analogy between the development of AI and evolution.
palata · a year ago
> However, witnessing the rapid evolution of AI with just a few hundred GPUs, enough data, and power, I no longer wonder what a billion years of feedback loops and randomness can achieve.

The thing with evolution is that it is robust. Even after the human species collapses (with everything it manages to bring down with it), evolution will still work.

Modern humans have been very quick at many things, but at the cost of breaking the conditions of their survival. One may say "that's wrong, with enough GPU a solution will appear", but on the other hand maybe we have already passed the tipping point where we actually cannot solve it anymore.

nkrisc · a year ago
The fascination with “intelligent design” is cherry-picking. The is no shortage of “unintelligent design” in the natural world.

Take humans, for example. You can block your trachea and die through the simple act of eating. An intelligent (and omniscient) designer could have avoided that by better designing our overall our overall structure.

Or take the fact or ear bones are modified jaw bones. Or if you believe in intelligent design, ask why our intelligent designer thought it wise to link our jaw to our ears so that it’s hard to hear things when you’re chewing.

smusamashah · a year ago
When you notice these flaws you are seeing it in very very short term. What we are today is what eventually worked for a million years. The design you see today is the way it is because it had to be robust enough (including those problems) to survive to this day.

The examples you quote do look like a problem today, but I think they must have worked to some benefit to bring us here.

lpat · a year ago
That's pretty weak argument against intelligent design.

1. The designer does not have to be "omniscient" only intelligent. Some people believe in it's omniscience but I don't think it's a requirement.

2. How do you know that our trachea is not an optimal design when you take into account all the tradeoffs?

3. You're surrounded by items created by intelligent design (us) and all of them have flaws, reused parts and tend to break. Obviously you wouldn't argue based on this that they weren't designed by intelligent beings.

ssener2001 · a year ago
Where has our mind shown the capacity to be an engineer of the universe? With this limited intellect, we cannot encompass the beauty of the whole. Even if it were beneath a forearm-length nose, if only attention were paid to it, beauty could still be found!"
hackeraccount · a year ago
My favorite version of this argument is from Catch-22.

It starts with a discussion about the miseries people suffer and if they have any utility at all. Is pain a useful warning of problems or could that information be gotten across in a better way?

It ends with the two characters agreeing that neither believes in God, with one not believing in a good and kind God and the other not believing in an evil malevolent God.

https://risingentropy.com/catch-22/

delichon · a year ago
I consider myself an intelligent designer but have to make compromises in design all the time. The only kind of designers that don't are ones with cartoonish super powers like deities.
erfgh · a year ago
Well I can crash my bicycle going downhill and neglecting to use the brakes, this doesn't mean it wasn't intelligently designed.

Design is all about tradeoffs.

Deleted Comment

jhanschoo · a year ago
As other commenters, it tickles me that this was your takeaway as optimizers mildly altering weights at every training step to a more correct representation is more analogous to intelligent design, where a God guides the evolution of species to develop the features they need.
interludead · a year ago
It makes the idea of randomness and feedback loops in nature more tangible

Deleted Comment

protonbob · a year ago
All of the data that has gone into AI has been intelligently created. Also, there are plenty of intelligent people cleaning the test data and guiding its training.

That is basically the entire premise of intelligent design. Not that there is no evolution, but that it is a guided process.

fellowniusmonk · a year ago
Right, genetic and morphological lineage data among hominidae and other families look exactly like you'd expect from unguided mutation. So any intelligent hand involved would have to be a real trickster to put so much effort into making the data look the way it does, or just shouldn't have bothered because it seems we end up at the same place regardless.
fennecbutt · 10 months ago
I mean even real biology like our tailbone, reflexes and organs we no longer actually need are all good indicators.

Deleted Comment

ssener2001 · a year ago
Notice that AI is not by evolution but the result of many scientists or works, devices etc.

If suppose evolution were to occur , then many wrong and absurd things would emerge or they would not have come into existence. But there is no disorder or mistake in the past or present and everything seems to be created perfectly.

"We see that in its existence, its attributes, and its lifetime, while hesitant among innumerable possibilities, that is, among truly numerous ways and aspects, each thing follows a well-ordered way in regard to its being in innumerable respects. Its attributes also are given it in a particular way. All the attributes and states which it changes throughout its life are specified in the same fashion. This means it is impelled on a wise way amid innumerable ways through the will of one who specifies, the choice of one who chooses, and the creation of a wise Creator. He clothes it with well-ordered attributes and states. Then it is taken out of isolation and made part of a compound body, and the possibilities increase, for they may be found in that body in thousands of ways. Whereas among those fruitless possibilities, it is given a particular, fruitful state, whereby important results and benefits are obtained from that body, and it is made to carry out important functions. Then the body is made a component of another body. Again the possibilities increase, for it could exist in thousands of ways. Thus, it is given one state among those thousands of ways. And through that state it is made to perform important functions; and so on. It progressively demonstrates more certainly the necessary existence of an All-Wise Planner. It makes known that it is being impelled by the command of an All-Knowing Commander. Body within body, each has a function, a well-ordered duty, in all the compounds that one within the other themselves become components of larger compounds, and has relationships particular to each, in the same way that a soldier has a function and well-ordered duty in his squad, his company, his battalion, his regiment, his division, and his army, and a relationship particular to each of these sections, one within the other. A cell from the pupil of your eye has a duty in your eye and a relationship with it, and has wise functions and duties in your head as a whole and a relationship with it. If it confuses these the tiniest jot, the health and organization of the body will be spoilt. It has particular functions with regard to each of the veins, the sensory and motor nerves, and even the body as a whole, and wise relations with them. That specified state has been given it within thousands of possibilities through the wisdom of an All-Wise Maker.

In just the same way, each of the creatures in the universe testifies to the Necessarily Existent One through the particular being, the wise form, the beneficial attributes given it among numerous possibilities. So too when they enter compounds, those creatures proclaim their Maker with a different tongue in each compound. Step by step till the greatest compound, through their relations, functions, and duties, they testify to the necessary existence, choice, and will of their All-Wise Maker. Because the one who situates a thing in all the compounds while preserving its wise relations, must be the Creator of all the compounds. That is to say, it is as though one single thing testifies to Him with thousands of tongues. Thus, from the point of view of contingency, the testimony to the existence of the Necessarily Existent One is as numerous, not as the number of beings in the universe, but as the attributes of beings and the compounds they form..."

So turn your vision again: do you see any flaw? Then turn your vision a second time; your vision will come back to you in a state dazzled and truly defeated(Quran)

state and point out, however much the human gaze tries to find faults, it can find none anywhere, and returns worn out to its dwelling, the eye, and says to the fault-finding mind who sent it: “I am worn out for nothing; there are no faults.” This shows that the order and regularity are most perfect

namanyayg · a year ago
Disagree.

Counterpoint: allergies, birth deformities, cancer

Aachen · a year ago
Tangential but for my information, does anyone else find this article's thin font takes noticeably more effort to read? I'm not sure if it's just me (I might need to get a better display that can output more nits in this bright room) since the site owners apparently think it's good
crazygringo · a year ago
Not just you. The font is absolutely too thin to be used for body text.

It might be more readable on low-res screens that hint the strokes to a full pixel width, but on Retina screens it's simply too hard to read.

Thin weights like that are meant for display uses (titles, posters, etc.), not body text.

dystroy · a year ago
People who do robotics, is efficient stable balancing at rest something which is studied and applied ?