Permalink
Switch branches/tags
Nothing to show
Find file Copy path
Fetching contributors…
Cannot retrieve contributors at this time
172 lines (123 sloc) 10 KB

Evolutionary Algorithms Might Reduce Costs of Fixing Climate Change

by Sven Nilsen, 2017

Disclaimer: This is an opinion piece for further discussion and experimentation with ideas.

The basic problem of fixing climate change is the following:

  1. Get rid of unneeded greenhouse gases
  2. Store the gases safely
  3. Do so on a super-scale at extremely low cost

Mankind is currently suffering from a cognitive bias that follows with how we spend energy:

  • The energy infrastructure is centralized
  • Only a fraction of the available energy is utilized

For example, when the sun heats the ocean, some of the surface water evaporates and form clouds. These clouds are then brought over landmasses by wind and fall down as rain. The rain is collected by rivers into dams of water and then falls through turbines that generate electricity.

In the end, only a fraction of the original source of energy (the sun) is utilized (electricity). When using a general ability to solve problems, such as electricity and computers, it creates a frame of mindset limited by how the electricity is generated and how computers are designed, which in turn leads to a cognitive bias (a bottleneck in the brain) for addressing the problem.

It is no wonder that when people look at climate change the problems that follow, that it seems like an insurmountable problem to overcome, which requires massive amounts of investement.

The problem seems unsolvable because the way we are thinking about the problem is wrong.

In this post I will address the 3 main bottlenecks to fix climate change.

  1. The energy source bottleneck
  2. The material bottleneck
  3. The algorithmic complexity bottleneck

I will first put some effort to explain the 2 first ones, because the 3rd is building on top of the two.

1) The energy source bottleneck

To build an engine that runs directly from the environment, all one needs is a temperature gradient or a source of photons (light).

The ocean environment has both things: Cold/hot water and available sunshine.

Cold water is cold because surface water blocks sun light from heating up the water underneith. This means that when the sun adds energy to the ocean surface, it creates a natural temperature gradient, which drives the forces of what we call "wheather" and in turn trigger some events that leads to "economic activity".

The world economy is just a system that humans have adopted to utilize a fraction of a particular coincidental, but natural, shape or organization of matter that respond to sun light in a specific way. With other words, the world economy is not something that drives itself, but rather is piggy-backing a much larger system, which one might call "the planet energy cycle".

A temperature gradient that is capable of driving the weather, is equally able to drive engines and chemical processes. It is a matter of repurposing some of this available energy into fixing the problem.

Most land based life is importing a lot of their energy from the ocean. If we want to get close to the source of this energy, we need to go to the ocean. This is how we get rid of the energy source bottleneck.

2) The material bottleneck

All known material is made of atoms, which when bound to each other through electron-pairs form molecules. Molecules comes in various shapes and magnetic fields, which determine their properties. The properties of molecules, such as how it responds to heat, pressure and electromagnetic radiation in various frequencies, determine the properties of the material in combination with it's designed shape. The shape can vary across microscopic and macroscopic scales in different patterns.

From a physical point of view, all one needs to produce material is a large amount of atoms, and a process that focuses the probability of creating the desired electron-pair bindings. This process requires energy, but as we have seen in 1) this problem is solvable.

The difficult part is finding the right sort of processes and duplicate them.

As a land-adopted species, humans have been occationally interacting with a complex, living kind of system, called "plants". Plants have the ability to reproduce and self-heal from many wounds to its structure. With other words, plants display the exact behavior we wish we had in a climate-change-fixing system operating in the ocean environment.

A plant spends much less energy to build one unit of mass compared to an animal, so the kind of systems that assemble material from atoms are more likely to come from plants. This process is designed by a long, blind chain of adaptions to the environment in order to self-reproduce, so it might be possible to increase the energy efficiency a lot.

When humans visualize plants, they think of some structure growing up, against the force of gravity. Still, every boat owner knows, learned through lot of frustration, that plants are affecting their floating transporting device in the opposite direction. In the ocean, plants grow down instead of up. What the plants need is a surface to grow on, sunlight, and nutritions from the sea water.

This is a source of material that can be utilized at relatively low level of energy. Instead of trying to defeat gravity by growing up toward the source of light, which requires extra energy, the plants are able to grow down from a surface held up by water pressure. The reason plants grow upward on land is because they compete against other plants for light, but this is not needed on the ocean where there is plenty of space.

This is also a demonstration of how energy processes that people are accustomed to, shapes their thinking. A simple thing like "plants grow up" must be changed to "plants grow down".

When plants grow down in the ocean, they extract CO2 from sea water. Since sea water is capable of absorbing CO2, one can indirectly extract CO2 from the air, without needing the surface area required to absorb CO2 efficiently. The ocean itself does half the work!

Basically, you need lots of lots of lots plants growing down from floating devices, which are produced by material produced by the plants, using energy sources available everywhere for assembly.

After enough plant material is collected, it can be compressed and sunk to the bottom of the ocean, or utilized for economic benefits.

3) The algorithmic complexity bottleneck

Humans are not able to write algorithms above a certain complexity, without the assistance of algorithms that teaches themselves how to solve problems.

In order to fix climate change, it would foolish to think that the complexity is low enough to be achievable by hand-coding in all the rules.

For example, to make a ocean-based system adopt optimally to various wheather, waves and streams, the number of parameters to test upfront would require a super-computer to check all cases.

Instead, one can program a system to simulate a small brain, called a "neural network". The neural network takes some inputs and produces some outputs, based on the weights inside the network. The weights are organized in a pattern that allows the network to approximate a wide variety of functions. By using some smart tricks of changing the weights, the network is capable of adopting to new environments.

This means that the same system can adopt to a widely different environments, as long as its hardware is generally capable.

One successful demonstration of a such network is the human brain, but it is not necessary to scale to such amount of complexity.

The complexity necessary to fix climate change is not about fixing many different problems from a single location in space and narrowed by time, but about fixing many problems across many locations of space and time. This means the same network could be used in a system with many units of relatively low complexity.

The bottleneck is the scale of the total complexity, not the complexity of each individual part.

For safety concerns, a such system will not pose any danger to humans or the environment, as long the hardware does not allow it to make actions that go undetected and which have not shown up under testing. Without evolutionary pressure that favors dangerous behavior, an evolved population of algorithms will not evolve dangerous behavior. Also, reduced complexity restricts the range of theoretically possible dangerous behavior.

A system designed to fix climate change does not need to evolve its hardware, but only the software that controls the hardware. If one unit displays undesirable behavior, it can swap its software from a unit that displays desirable behavior.

In nature, evolution happens by random mutations and cell mechanisms for turning genes on/off. This requires many generations before an adoptive behavior emerges across the population. When simulating this process in software, it is possible to speed up adoption to a much higher rate. The perception that "evolution is slow" does not apply to software.

Thus, it should be possible to design a system, given the right hardware capabilities, that is able to perform efficiently under a large variety of weather, waves and stream patterns, by utilizing local energy sources and then grow materials needed for tasks/construction, without posing any danger to humans, and with proper precaution, have little effect on the environment.

If one produces a single, centralized device to fix climate change, it would need to be extremely complex and therefore hard to optimize.

On the other hand, a distributed system consisting of many units that learns from each other, self-optimizes faster the more units there are, so choosing the right architecture in the beginning yields larger returns on investment in the end.

This means evolutionary algorithms might reduce the cost of fixing climate change, because of the reduced complexity within each unit, and the optimization benefits increasing with scale.

Another benefit is that one does not need to design the whole system upfront. Just give it general abilities for input/output to perform the tasks that are needed. This reduces cost in the planning and prototyping phase, which allows more planning and prototyping for the same amount of investment, enabling a more secure and capable system.