Curriculum

This page is read only. You can view the source, but not change it. Ask your administrator if you think this is wrong.
====== Notes on Sapolsky 22. Emergence and Complexity ======

https://youtu.be/o_ZuWbX-CyE

[[Notes on Sapolsky Human Behavioral Biology|Back to top]]

Fractals. Scale-free. Chaos. Deterministic. Aperiodic. Tiny differences can have an amplified effect. Butterfly effect.

==Cellular Automata==
[[Cellular Automata]] from Wikipedia.

1:55
Very simple rules, out of which emerges a complex pattern.

Features required for **emergent complexity**.

  * Lots of building blocks.
  * Simple building blocks.
  * A starting condition 
  * Simple rules on how to form the next generation.
  * Each rule is based only on the state of neighboring building blocks.

Out come structured patterns.
In most cases, the patterns stop, go extinct.
A large number of starting states produce a small number of similar patterns.

Convergence. 

The notion that you can start with different forms and they will convergence over time.

You look at the mature for and you cannot know the starting state.

The starting state gives you no predictive power to know the mature state.

Cellular Automata encapsulate these features of evolution.

  * Most go extinct.
  * Only a relative few mature forms exist.
  * Convergence.
  * Butterfly Effects. Minor differences in the starting state can extend into very different consequences.

5:55

Starting state.
One space between. Boring. Static.
Two spaces between. Extinction.
Three spaces between. Extinction.
Four spaces between. Living. Dynamic.
Five spaces between. Living. Dynamic.

Asymmetric starting states produce more dynamic systems.

9:32
12:41
Patterns in nature.
Sea shells.

13:33
Robert in Kenya, equatorial glacial mountain. Friend in the Andes. Both photograph plants which look exactly the same, but have no taxonomic relationship.

Desert plants. Small number of plants. There are only a few forms that can survive.

Legs. Two, four, six, eight. Only a few ways to do legs.

Steve Wolfram. Large ego. Grotesquely wealthy.

Emergence. You can code for great complexity in the natural world with a small set of simple rules.

19:10

Another look at the ways reduction stops working after awhile.

Not having enough numbers of things. Not enough neurons. Solution: neural networks.

Not enough genes. To specify the bifurcated network of blood vessels, for example. We need fractal genes, with scale-free instructions. "Split when your length is five times your width."

In your body, there is no cell that is more than 5 cells away from a blood vessel. Yet, the circulatory system is less than 5% of body mass.

Start with a line. Remove the middle third, leaving two line segments. Remove the middle third of each of those, leaving four line segments. Repeat ad infinitum. The results is an infinitely large number of objects taking up no space.

Do the same thing with a triangle. Remove an upside down triangle from the middle third, leaving three triangles. Repeat continuously on all resulting triangles. The result is an object with an infinite amount of perimeter surface area within a finite space. This is impossible of course.

What would be a fractal mutation? Length 4.9 times width. Catastrophic.

A small number of diseases having to do with spatial relationships within the body. For example, one disease has things wrong in the midline of the body: the septum in the nostrils, the septum in the heart, the hypothalamus, Etc. These are not three different mutations. It's one fractal mutation that messed up the symmetry in the embryo.

Emergent complexity from biophysical properties.

Example from Paul Green. A disk where the inside material is softer than the outside rim. Apply heat. The disc will warp into a double saddle shape. Like the Potato chip.

In many plants, the branches come off the stem in a double saddle shape. This is the inevitable solution to the problem of packing things into a smaller shape.

Examples of **the wisdom of the crowd**.

At the State Fair there was a contest to guess the weight of an ox. 100 farmers guessed. No one got the right answer, but the average of all the guesses was correct within an ounce.

Who Wants to Be a Millionaire, when stumped on a question, ask the audience. They get it right 91% of the time.

The Pentagon asks questions like where will the next terrorist attack occur or where is the missing sub.

A group of ants carry a dead Beetle to the ant colony. No single and knows the correct direction to go but as a group they end up going in the right direction.

51:42


Emergence in an ant colony.
Bottom up organization.

==Swarm intelligence==

Traveling salesman problem.

Virtual ants.

Bee hive. How to find a new place for a hive.
Dance. Length of dance proportional to quality of food source.
Two generations. Random interactions.

==Repulsion and attraction rules==
Like magnets.


Urban planner. Simulation with simple rules of attraction and repulsion. 

Neurons in a petri dish. Looks exactly like an urban plan.

Tokyo subway system, similar to ant colony. Ant colony more efficient.

#pattern

==Origin of life==
1:08:38

Experiment at U of Chicago by Urey and Miller 1952. 

Primordial soup. Inorganic molecules of carbon, methane, water vapor, etc. Pass electricity through it repeatedly. Amino acids began to form.

Many scientists now working on origin of life. Later experiments show you don't need the electricity. Put molecules with attraction and repulsion rules, they form rational structures.

Another example. Child's toy, large number of small round magnets. They form patterns.

==Power law distribution==

[Curve sloping down into the right. 80/20 rule. long tail. popularity.]

  * Size of earthquakes.
  * Length of phone calls.
  * Distance traveled by a marked dollar bill.
  * Number of links that websites have two other websites.
  * The number of proteins showing certain degrees of complexity.
  * Number of emails sent by persons during one year.
  * Degrees of Separation among actors by filmography.
  * Length of neurons.

This is a fractal because it exists at many different scales.

1:16:26

Traveling salesman problem in designing a telephone Network. Distance traveled by phone calls is a power law distribution. Most calls are a short distance, carried by Urban cell towers, but occasionally you make a long-distance call over a long trunk Network.

Ditto for neurons in the cortex. Most connections are short but sometimes you can talk to another Point far away in the cortex. The length of the neural connection follows a power law distribution.

1:19:00 In patients with autism, the tail of the distribution is shorter. That means relatively fewer long distance connections.

Males have a steeper curve than females.

The corpus callosum is thicker in females than males to accommodate the larger number of long-distance connections from one hemisphere of the brain to the other.

==Bottom-up quality control==

User reviews of books movies music Etc. Recommendation engines based on people who liked this also liked these.

As opposed to top-down system of professional reviewers and editors and publishers.

Wikipedia. Bottom-up, self-organizing, self-correcting.

Rated comments.

Recommendation engines. Biased to conformity.

1:27:00
Developing cortex does a swarm intelligence solution.

Pioneer generation.
Radial glial cells.
Second random Wanderer generation: neurons. When they cross a glial cell, follow it.

1:29:20
How do we apply this stuff to humans?

looking at a single neuron you can't tell whether it came from a fruit fly or a human or what?

We have the same kind of neurotransmitters that a worm has.

We have a hundred million neurons for each one neuron and a fly brain.

From that large number of neurons comes emergent properties.

Deep Blue vs Garry Kasparov. "With quantity you invent quality."

What is the difference between a chimp brain and a human brain genetics. In the human some subset of the olfactory neurons have atrophied. Bone structure is different. chimps have more hair. Genes affecting cell division in the neurons. Humans go through more rounds of cell division.

1:30:00 summary

Emphasis on quantity.
The simpler the constituent Parts, the better.
More random interactions make more more adaptable Networks. Randomness is good.
Gradients of Attraction and repulsion.
Nearest neighbor interactions.
Generalists work better than specialists.


A future generation 

Optimal

Don't need blueprints. Don't need the people who make blueprints.