Evolution is a general purpose idea about how knowledge can be created. Think of knowledge as useful, good information.
When something appears to be designed for a purpose, there is knowledge there. For example, hawks' eyes can see far – they seem designed for the purpose of long range vision; there is knowledge in a hawk's eye. And a wrist watch keeps time; it seems designed for the purpose of keeping accurate time; there's knowledge there.
There used to be a great mystery about how knowledge could be created. Some people thought the answer must be a designer. It's easy to understand how knowledge can be created by a designer because he already has knowledge and uses it for his creation. A watch does have a human designer, and that isn't mysterious. So people thought hawks were designed by God.
But where does a designer's knowledge come from? Even if you say human designers were designed by God, then where did God's knowledge come from? If designers are the only source of knowledge, God must have had a designer, and God's designer must have had a designer, and so on. Which doesn't work.
Evolution solves this mystery of where knowledge can come from without a designer. It's the only idea which has ever solved this mystery.
ReplicatorsA big idea of evolution is replicators. Some things make copies of themselves. It's in the context of this repeated copying that evolution happens.
Replicators aren't simple, but they aren't a huge mystery like knowledge creation. It's not that hard to imagine building a robot which is programmed to construct more robots using the same design as itself (assume that people come by periodically and give it raw materials).
Some types of crystals, placed in the right circumstances, create more crystals of the same type. If there was no crystal there, none would be created. But if the wind blows a crystal to a place with the right circumstances, or a human places one there, then it creates more crystals. (If you're curious about this, look up the idea of a "seed crystal" to get started.)
Variation and SelectionReplicators aren't enough for evolution. Variation and selection are also needed.
Variation is easy. Replicators aren't perfect. Errors happen. Some copies are a little bit different. This can be random or accidental. The result is some amount of change, some new things are created.
Selection is more interesting than variation. The general principle is that replicators which are better at replicating end up existing in larger numbers. Replicators which are inferior end up existing in smaller numbers, even zero.
The better something is at making copies of itself, the more copies of it will exist in the future. (On average. It could get unlucky.)
When you put variation and selection together, evolution happens. Lots of different replicators get made (due to copying errors). Because they are made randomly, not designed, most of them are inferior replicators. They make fewer copies of themselves. But a few variations happen to be improvements, and make more copies.
Inferior at what? An improvement by what standard? The standard of making more copies.
And not just immediate copies, but also copies of copies. And copies of copies of copies. In other words, great grandchildren count. In fact, it's better to look at great grandchildren than regular children. That's a good rule of thumb: the more great grandchildren a replicator creates, the better a replicator it is.
What if we aren't dealing with people or animals? Then instead of "children" think "copies". And instead of "great grandchildren", think "copies of copies of copies". I'll speak of great grandchildren for convenience, but the more general concept is copies of copies.
DeathWith genes and animals, the replicators die off regularly, and decompose, and the resources (like the atoms they are made out of) get reused. So a replicator which doesn't do very well ends up at zero copies, and better replicators use its resources. But this is just one possible scenario.
We could also imagine replicators which aren't living creatures, which don't die, and which don't have the ability to take resources (like building material) from other replicators. Then the inferior replicators wouldn't die off, there'd just be fewer of them compared to superior replicators. Over time, better replicators will create way way more copies.
Suppose a replicator is able to create 10 copies per year (they all take the full year to be created). But another replicator can do 20. After 20 years with no replication errors, do you think the better replicator will have twice as many total copies?
It's actually far more. The better replicator will have 413,554 times as many copies after 20 years. Over time, better replicators dominate, even if they're only a little better and nothing ever dies or runs out of resources to keep making copies. Direct competition between replicators is not required.
KnowledgeSo, replicators have copying errors and then over time there are more replicators that are better at replicating, and fewer that are inferior at replicating. Where's the knowledge? Where's the useful information, the appearance of design?
Well, over time these replicators get good at creating lots of great grandchildren. So, they appear designed for (approximately) the purpose of creating lots of great grandchildren. So there is knowledge there. There is useful information that's able to achieve a specific purpose.
It's not just any purpose. It's not knowledge about anything. It's knowledge specifically about (roughly) replicating great grandchildren. But that is knowledge.
So how are other types of knowledge created? Like a hawk's eyesight.
Knowledge About Other TopicsSo where does knowledge of a hawk's long distance eyesight come from? Or a tiger's sharp claws, an ant's scent trails, a cow's ability to create milk, a fly's ability to land softly.
The general principle is that creating knowledge about one topic often creates knowledge about other topics too. If I want to be a good physicist, I'll have to learn some math too. If I want to be a good doctor, I should know some chemistry. If I want to be a good lawyer, I should learn how to read. Pursuing one topic leads to knowledge about many topics.
Getting great grandchildren is a complicated problem. Let's consider animals. They don't just have to have babies. They also have to get food, and not become food. Animals do not have knowledge about just anything. But they do have knowledge about many things relevant to having great grandchildren besides fertility.
Hawk eyes and tiger claws are relevant to their survival, and survival is relevant to having children. That's why evolution was able to create these eyes, claws, and so on.
RandomnessSome critics portray evolution as randomness, and question the ability of randomness to create knowledge. And some of their opponent's laugh in their faces and call them ignorant. But, actually, there's an interesting issue here. It's a good topic to bring up.
The variation part of evolution is random. (Actually not exactly, but that's complicated, I'm not going to get into it.) And randomness doesn't create knowledge. Randomness doesn't design things (like eyes) for purposes (like seeing).
Although part of evolution is random, part isn't. The selection part of evolution isn't random. It designs replicators (like genes or ideas) for a purpose. What purpose? Roughly, to have a maximum number of great grandchildren.
How can something designed for one purpose (great grandchildren) be good at other purposes? This is an important question which some intolerant anti-religious evolution-proponents do not understand, let alone have an answer to. I discussed it above.
GenesAnimals are not replicators. And an animal's offspring are not copies of the parent animal.
What's actually copied are genes. Genes are tiny little sequences of information (made out of DNA). What they mainly do is something like control proteins to build baby animals. Genes are copied to child animals pretty much perfectly (except for infrequent tiny errors).
If you want to know more about genes, a good place to start is by reading The Selfish Gene.
MemesA "meme" is a word that means an idea which is a replicator. It's used for talking about the evolution of ideas.
If you want to know more about memes, you can look at my archives, or ask at the Fallible Ideas discussion group.
Why is idea space approximately organized into big semi-autonomous groupings?
On 2020-02-26 curi wrote in #fi (on the FI discord):
> like why is idea space approximately organized into big semi-autonomous groupings after you take out the crap?
> For the same reason that lifeforms can be organized into kingdoms, phylums, etc. It's because ideas evolve, like DNA. They build on , subtract from, and modify their predecessors.
curi replied to jordancurve:
> disagree. that suggests it's not inherent in the problem space. i think it partly is, both for ideas and organisms.
I (jordancurve) now agree with curi. I was wrong. I implied that my explanation was complete, but it was actually incomplete: evolution helps explain why semi-autonomous groups would *persist*, but it doesn't explain why those groups would *exist in the first place*.
Something else is needed for that. I think curi's own explanation works:
> i think that's to be expected because 1) it's sparse 2) problems aren't infinitely demanding. so often a solution can be adjusted a bit and still work b/c there is leeway in the problem
I'll try to explain it in my own words.
First, two definitions:
- An organism’s *environment* is the physical environment (including other organisms) in which it has to replicate in order to not go extinct.
- An idea's *environment* is the set of minds (including other ideas) in which it has to replicate in order to not go extinct.
Ok. I think curi's idea is that the environment itself (a.k.a. the problem space) causes the replicators in it to fall naturally into semi-autonomous groups. For example, some organisms live in water, some fly, some eat other organisms, some get energy from sunlight, etc. The environment is at least partly responsible for these naturally occurring groups.
Likewise, groups of ideas would also be influenced by their environments. For example, people have created different fields of knowledge over time. Each field consists of ideas that solve problems appropriate to the field. Ideas within a field are therefore related by the fact that they solve similar problems.
Part of my idea was simpler.
Imagine a very large, empty 3d space.
Now, in your imagination, *sparsely* add some tiny spheres in it. So there are some little dots here and there, but it's still mostly empty.
These spheres represent good ideas. The empty space represents bad ideas.
Now consider: lots of variants of a good idea are still good. Lots of changes don't ruin it. Therefore, near every sphere, add a bunch more spheres.
Now you've got clusters.
More on: Why is idea space approximately organized into big semi-autonomous groupings?
> Imagine a very large, empty 3d space.
> Now, in your imagination, *sparsely* add some tiny spheres in it. So there are some little dots here and there, but it's still mostly empty.
> These spheres represent good ideas. The empty space represents bad ideas.
I agree with this, as far as it goes. However, it doesn't answer the question I was trying to answer in #16361, namely: Why is the space of good ideas *initially* sparse?
The explanation that makes sense to me, which I took to be your explanation (i.e., it is "inherent in the problem space"), is that the problem space *itself* limits the areas in which solutions can most easily be found. Each initial good idea is ~randomly chosen from within those areas.
> Now consider: lots of variants of a good idea are still good. Lots of changes don't ruin it.
> Therefore, near every sphere, add a bunch more spheres.
> Now you've got clusters.
I agree with this part too, and I don't think I have anything to add to it. Restating what I think it means in my own words: the sparseness persists because evolution works via small incremental changes that generally don't bridge the relatively large gaps between the initial good ideas.
> Why is the space of good ideas *initially* sparse?
Oh, do you mean you want to know why most ideas are wrong? (It's confusing because the space of all possible ideas is timeless, so "initially" doesn't make sense.)
Here's an indication of why most ideas are wrong:
Let's suppose there are a million actions you can take. You can define goals which require any subset of those actions be taken and none of the rest. The goal is like a bitstring with the bits being on for the actions that should be taken and off for the others. So there are 2^1000000 such goals and for each one there is 1 way to get it right and 2^1000000-1 ways to get it wrong.
Many goals are less strict. E.g. you could have a goal of creating a bitstring (or taking actions) which match 99% or better with a target. This creates clusters of similar solutions.
There are many other ways to gain some perspective on why most ideas are wrong. Some are way simpler. I bet you could think of some that don't involve math or geometry.