3D Fitness Landscapes

A fitness landscape is a helpful tool for thinking about biological and memetic evolution—which information will be successful in a given environment. Unfortunately, most fitness landscapes neglect the environmental dimension. This stops us from thinking clearly about how information flows. We should add "the environment" to our fitness landscapes!

Let's start by explaining fitness landscapes with a simple example. Let's think about what kind of gorillas reproduce best (have "high fitness"). We can imagine two axis: strength and speed. Strong and fast gorillas will have high fitness. These will be the "hills" on a fitness landscape. The gorilla population will "hill climb" these hills to find the best gorillas.

The dependent variable on the y axis is always "fitness" (ability to replicate). But we can have other independent variables. For example, a sequence space is a fitness landscape where the axes are amino acids in position 1, 2, ..., n. For just two amino acids, we can represent it like below:


In reality, sequence spaces have a massive amount of dimensions, not just two. SARS-CoV-2 has 10,000 amino acids, so its sequence space has 10,000 dimensions. The most famous variants of concern change have mutation E484K, meaning that amino acid E changed to amino acid K at location 484. There's a hill there. Many variants convergently evolved to "find" that hill.

Most visualizations of sequence spaces have three dimensions (like the one above) to make it look like the real-world three-dimensional hills we're used to. This means they show two dimensions of "characteristics" (or arrangements of information, like amino acid sequences) that determine the fitness.

I think this is a bad norm.

Instead, I think we should show all [arrangements of internal information] on one axis and show [arrangements of external matter] ("the environment") on the other axis.

Here's what I mean.

Think back to the Cambrian explosion, 550M years ago. Why did it happen? One of the big reasons is that oxygen had already filled up the sinks in the ocean and rocks, and so could fill up the atmosphere. An oxygenated atmosphere allowed new eukaryotic organisms to access tons of energy (through oxygen respiration) and live on land (because the ozone layer had just formed, protecting them from harsh rays).

This change in the environment allowed certain sequences of DNA to thrive. The sequence for "legs to walk on land" had low fitness when land bombarded by UV rays, but legs had high fitness when land was more hospitable to life.

The external matter environment changed the fitness for the internal arrangements of information (DNA sequences). We can create a 3D cube that looks like this:

A DNA arrangement that is low fitness in one environment can be high fitness in a different environment.

I prefer this three-dimensional view because it shows that fitness is determined by both DNA and the environment. We not only inherit DNA from our ancestors, but also the environment. Organism's DNAs are the answer to the question: "what can live here?" not just "what can live?"

In addition, these three dimensions map onto information, energy, and matter, the three core primitives of the universe. DNA (an arrangement of info) hill climbs in the environment (an arrangement of matter), trying to replicate through finding energy.

These three dimensions also help us see non-biological forms of evolution more clearly. For example, successful memes are determined by the environment they can live in. In the Agricultural Age, myths like religions became high fitness because we needed to trust each other in cities. (Moralizing god religions were low fitness in the Forager Age.) Today, immigration into white counties created fear of being left behind ("Great Replacement"), which created a ripe environment for newly high fitness memes like QAnon, StopTheSteal, and capitol insurrection. As a final example, the climate crisis has (literally) changed the environment and made "renewable energy" a higher fitness meme.

To understand evolutionary fitness landscapes, we should use information, environment, and energy as the three dimensions. This helps us reason about why certain genes and memes are successful, and how we can shape our environment towards more beneficial replicators.


  • It's always tough to do the correct dimensionality reduction. Principal component analysis is a great algorithm for this. I'm essentially claiming that the two principal components are internal information and external matter.
  • My favorite PCA https://everynoise.com/
  • The traditional way to show the environment is by showing a dynamic fitness landscape, i.e. by making an animation through time that changes the landscape. This is fine, but I still think misses the power of a still-image that has time "baked in." Most graphs have time as the x axis!
  • Relatedly: What are the units of the environmental axis? They are 4D space-time. For example, a plant can move through 3D space to find the sun and photosynthesize better. Or it can stay in the same space but have the sun "come to it" through time (e.g. not an ice age).
  • Niche construction is a way to modify the matter of the environment to create a high fitness hill. Beavers create dams to create a high fitness niche for themselves.
  • See developmental systems theory for "environment is inherited, not just genes."
  • We can think of the environment itself as a replicator, where matter replicates from t=0 to t=n with some degree of mutation, selection, and heredity. Imagine the world broken into 1m cubes. Treat them as the item of replication. Physical laws determine what matter "replicates" through time. A rock stays a rock, but eventually erodes.
  • I like to think of this from a Sciences of the Artificial perspective. Any agent/artifact takes external info, does computation on it, and tries to succeed. Any sign of it not succeeding shows that it is "weak" at computation. Apes couldn't create farms to access energy because they don't have brains and social learning to do so.
  • Relatedly, I like the framing that a given DNA sequence maps to a given environment. This is an adaptationist lens. We have a two-way function. Environment --> successful DNA. Successful DNA --> environment. Give me either and I'll show you the other.
  • This gives a feeling of incentive inevitability. Genes and memes are incentivized to replicate, have historical contingency and physical constraints, and then get to an inevitable answer given those inputs.
  • The environment also holds all the other biological and memetic species. This creates game theoretical dynamics and evolutionarily stable strategies


  • Given three primitives of info, matter, and energy, I'm not exactly sure where computation lives. Computation uses energy to rearrange info into other info. Computation is the process that allows a given replicator to find energy? idk. Help me!
  • We can apply fitness landscapes outside of genes and memes too. e.g. National economies are trying to find niches of production to make money (find high fitness hills). They are in game theoretic games with other economies to do so. I have less experience with thinking of fitness landscapes in these ways but curious to learn more.
  • Another question/thing I don't understand: For memes, what counts as information and what counts as the environment? For example, the informational medium has upgraded from voice, to text, to printed word, to the internet. I generally see these as different ways to represent information itself. i.e. They change the x axis of info, not z axis of environment. e.g. When DNA could be 10000x longer given eukaryotes and not being attached to cell wall, this lengthened the info axis by 10000x. But idk if this is true for memes. In many ways, it's better to think of the environmental axis as "homes" for memes, and so the internet coming along means there are new types of homes like hashtags. Information would stay as just the bits, not the medium that the bits can live in. idk.