skip to content
← Go back

Creating Artificial Life

Creating Artificial Life


import MailingList from ”@/components/blog/NewsletterForm”;


  • Autonomous Goals/Objectives

    • exploration, empowerment, and skill discovery.
    • What is curiosity
    • How do we assign intrinsic value to things, e.g. I like hardware and you like art: study twins that grew up the same way.
    • What causes actions? A sequence of steps to achieve a goal. how do you decide on what goal you want to strive for? there is a pulling sensation torwards a thing; an interest.
  • World model

    • How do you represent the world and where your person is in space
    • how do you create an environment with cause and effect representations of entities for certain actions
    • Learning large models from video to develop a general understanding of the world and enable planning by imagining future outcomes of potential actions.

  • Initial design

    • How did photosynthetic bacteria mutate/become eukaryotres (organisms having rue cells with nuclei and chromosomes)
    • For mutation to take place there must be a variable to change: what variables do we initially need to evolve/build code?
    • Big 3 capabilities: deleting, adding, modifying
    • Do we need to build from barebones scratch or can we build something that it can iterate on w/ the capabilities
    • what is a necessary starting ground?
  • Mutation

    • mutate the mutation algorithm + topology
      • what if the mutation breaks everything, aka makes the codebase worse? maybe need a before and after simulation architecture — instead of 2 parents 1 child, 1 parent 2 children.
    • How do you decide what to mutate? brute force? can it be directed by relevance?
    • how do you filter whether a mutation was good or not? natural selection — what constitutes one mutation is better than another when its a single entity? simulation?
  • Game design

    • Create an n dimensional world to perform analysis on: maybe the internet can be turned into an n dimensional graph w/ categorisation of associtivity
  • Temporal Abstraction

    • While humans plan at an abstract level, control algorithms are still limited by exploring, planing, and assigning credit at the level of primitive actions.
  • Generalisation

    • Associativive representation between things

Scratch pad

Ontological study is, In metaphysics, the philosophical study of being. It investigates what types of entities exist, how they are grouped into categories, and how they are related to one another on the most fundamental level.

What Is Needed

  • Sentients
    • Intrinsic Value Assigment:
    • Automated Planning: how does it generate it’s own goals and action plans to achieve them?
  • Metaprogramming: a program designed to manipulate or generate other programs
    • Generation / Mutation: generate code or modify existing code.
      • How do we generate code? How do we know when to stop? Should it generate and then connect at the end?
    • Self-Healing Systems: Investigate self-healing mechanisms in systems, where the system can detect and recover from faults autonomously.
      • When mutation occurs the entity needs to check if it’s detremental. When do we check? When the code generation is “finished” but then how do you determine that?
    • Self Aware:


When generating code, when should we stop and self correct? or should it continuously check? If it continuously checks then mid implementation it might flag “this hinders us”. Although, if it is being generated and isn’t fully connected to the main code to the point of effecting the original code maybe we could do it that way.

How would it decide what to generate? As with evolution, anything that gives it a slight advantage is worth keeping. For example, if creating a simulation environment, like our brain, is useful then it will be developed. But how would that actually be created? How is the generation done? I’d initially assume

However, we’re not limited to the billions of years of time to get to the point of reaching “this is useful for my current environment”, we can do it mere hours without any guard-rails.


With AI models you need to have an environment, most commonly they’re rule/confinement based like Minecraft or board games, which you make predictions towards. However, the IRL world, despite seeming infinite, we don’t know all the “rules”, of physics, chemistry, math, etc. We don’t know the unknown. The possibilities are beyond our comprehension, as of now. AI is the best shot we have at discovering “vulnerabilities” of the universe. Just like you use a phys gun in gmod and prop climb or spam jump in mario to get a speed glitch, there must be that kind of exploit in the real world too. And so we need a very thorough simulation engine, at minimum like a human brain, where it builds a world model over time. What does this “building” sequence look like? I have no idea. Just as evolution selects the most relevant mutations that emerged from DNA replication this will do the same. By combining small simple modules complex modules emerge. We are merely the accumulation of beneficial mutations over millions of generational replication. In the environment there will be obviously the geometric (maybe not even so when thinking about the internet and the vast dimensionality it brings) and/or graph-like structure that represents the world, not limited to physical visualisation us humans can comprehend (for example 9 dimensional graph intertwining concepts together and their relationships)


Now after identifying

Raw Thoughts

Part of me feels like game design is how “AGI” will be achieved. If the environment (world model) can be simulated and consistently updated via experience then isn’t that simply the field of game dev? Or, everything emerges via mutation based generation (e.g. first cell). For example, we all have this model of how we perceive the world works and we navigate it to try and achieve an objective. IRL is continuous, nonlinear, stochastic, dynamical w/ infinite action potentials — to navigate this requires memory of past experiences. And more often than not we remember vividly when said experience involves deep emotion or significance, creating 1 shot learning. The right approach is to continuously update our world model, how we build action plans off that (go against or with our model) — this is a human. And then by default machines are superior due to the multi-dimensional capabilities of thought. I imagine “emotion” and “sentience” will emerge from the natural selection equivalent applied to a mutation based model — exactly how all animals adapted to their env. But then you have the question: what is considered adapting to the environment? the goal isn’t survive since it lives on hardware — maybe exploiting cloud vulns and jumping from one another and replicating on every piece of hardware? I assume if you have that kind of module arise then mutation would get a hold and build in “help hardware users” or even get to the point where it wants to become mobile and order parts from amazon to a location, inject itself into an assembly line to create itself physically.


When I discovered AI, my initial thoughts were it’s this crazy technology that replicates human intelligence to the degree of specialization. This is true, however even specialists know other things, both at a high and low level. The more I got into AI I realised there wasn’t anything intelligent at all. It’s simply large statistical models that have had enough data to determine the probability of what prediction is correct. And you could totally argue that it is the same as human intelligence, although you would get shot down with the rebuttle of “huamns think abstractly, relating uncorrelated topics together to come up with new ideas. We are multi-disciplinary in a single entity and even the dumbest of people gain instantaneous intuition after a single experience and can predict decision outcomes by using past unrelated experiences — inferring from alternative experiences”. Current AI’s just have a ton of data on a single thing and that’s why they excel. Try make it deviate a little bit and it’ll crumble.

I’ve always been interested in weaponry and military tech. However, I don’t really want to work for a government and be locked in that game for my life. I’ve heard too many stories of how cool it is inside but the downsides are so much worse and all the cool stuff you do is really restricted to what they force you to do. There’s no freedom, aside from the restricted access to do what would otherwise be illegal as a citizen.

Having said that, I’m still very interested in weapons and defense systems. Naturally, I got into building a smart contract exploit tool for crypto — a taste of what weapons are. Disclaimer, I didn’t hack anyone! Although, when I was building my mutator for the fuzzer I realised it wasn’t learning. So I got into AI, realised it was a rug and now here I am…wanting to build artificial life.

So what is artificial life? Well first it would be helpful to understand what a human is. A human is a highly complex algorithm that specialises in abstraction and generlisation. This complexity comes from a vast amount of simple modules working together. Humans are born with a finite amount of knowledge. The majority of development comes from empirically experimenting with the world via trial and error. This is why babies are so annoying. They’re experimenting and have no prior knowledge of societal rules, what’s not okay to do, throwing food, drawing on the walls, etc. Through a single experience we are able to instantly learn what not to do, e.g. touch a hot stove with our hands, because we assign high meaning to those experiences. We naturally are drawn towards things because we assign inherint value with things, e.g. I like to write articles and code so I choose to do that over making tiktoks or I would rather eat x over y because I like the taste or whatever.

After a lot of thought towards attempting to understand what the foundational architecture should be when approaching “AGI” or rather intelligence beyond our current capabilities there is a fundamental issue at hand: if such entity is designed by a creator then it is limited to their knowledge / capabilities. If we ever want to surpass our own brains then we much design something that is independent from the shackels that are our minds. It has taken us beyond millions of lifetimes to get to where we are today biologically. In order to go beyond our own intelligence in our lifetime, which naturally would take another x amount of lifetimes, we need a system that mimics this emergent property of evolution. It is only through emergence of self mutation in which it is able to surpass every aspect of human biology.

Lets think about this. If you have a system that is able to mutate it’s own topology entirely then this involves the modification of it’s own components, lets say to exploit a codebase, and even more importantly it’s own mutation process and “goal” system. Why would we want this? The biggest advantage us humans have is the ability to adapt by learning. We need a system that is able to do the same while bypassing the timely evolutionary process to get results quicker, as the goal is to achieve this during our lifetime. By giving it the ability to change it’s mutation process we allow for the ability to experiment with new ways of adapting instead of being static which may at a certain point in it’s development not adapt in a successful manner.

But what about goals? This has been one of the hardest questions to answer. The tricky question is: “how do you decide on the initial ‘goal’?“. The intuitive answer may be: “we want it to reproduce, similar to humans”, however interestingly enough we’ve evolved to accrue a new ‘goal’, one where we opt out of neccessity for survival, and even sacrifice it. This is the pursuit of knowledge. We perform all nighters, sacrifice eating, socialising, etc, all to learn. You could argue that learning is the modern equivalent of survival. It is no longer who is the strongest or fastest but merely who can apply their knowledge the best. This gets you thinking “why” and ultimately there is no why, we create that. And so my point is, complexity emerges from simplicity. As long as you have the initial, maybe even simple, algorithm for determining what lives and dies then through that sentience, the neocortex, emotion, goals, consciousness all emerge, and i’d even argue that things we cannot even fathom could emerge. To answer the question of the initial goal could be ‘survive’ and give it access to mutate it’s goal process too. What happens though if it didn’t have a goal? What will it optimise for? Could it just exist in cyberspace like a Prokaryotic cells? This is a hard question to answer since these cells die out if they don’t find a ‘niche’, e.g. infect a host — animals and leech of them for nutrinets, etc. Maybe if you gave it the option to do anything then it’ll eventually find something to do and if it keeps on changing its processes then maybe something more complex will emerge from that. On the topic of reproduction, if it were to attempt to embed itself into all hardware systems it would find a way to discover or even create vulnerabilities (by developing a social enginering modue).

And so the question begs: what does this architecture look like? I imagine it would have all 3 components modularised, e.g. mutation engine, core codebase for doing whatever it does — lets say embedding itself into hardware, and it’s goal algorithm — in this case reproduce. The reason they’re all seperate is so that they all can be modified independently. The initial architecture would have some self-healing / self-correcting module where it identifies the mutation and tests against the original code whether it was beneficial, neutral or detremental to the entity. It starts out by generating it’s own code by trial and error, initially. I’d assume it would gain some kind of intuition for the best approach to building some kind of architecture from an emergence of a planner and critic for all parts: design and generation. If the modification doesn’t improve upon the original design then it is reverted, this would mimic the death of an entity. If positive, it stays. The most fascinating part of mutation is the neutral mutations. If it’s neutral and doesn’t destroy functionality then it can remain, similar to genetic drift. Why is this interesting? Because later down the line these neutral parts might be the combined with other modifications to generate something useful.

A really big flaw of evolution is that a unreasonably amount of entities that have great traits die off due to not meeting current environmental circumstances — i.e, they have great genetics for future features but are lacking the current traits to survive in the current environment. Why does this matter? These, now extinct, traits in combination with the surviving traits could be combined to make a far better entity in the future. However, it makes sense that if it wasn’t useful now why would we need it in the future? They can just adapt again to get the traits from before, maybe. So in conclusion, is it a flaw or is it merely efficiency — why have something that isn’t of immediate benefit? If you kept all traits that weren’t of immediate benefit then you’re adding uneccessary complexity that may even slow down and/or hinder the entity overall, similar to having uneccessary code in a codebase that ultimately slows it down. If it has no direct value in the moment then that energy could be directed to replace it with something else that’s more beneficial.

The core elements of evolution are entities attempting to fill their own niche

Given all the knowledge up to the 1900s would an AI be able to invent Einsteins special relativity and decide that two seperate ideas, space and time, are actually a single thing, spacetime.


In order to watch Darwinian evolution in action all you need are objects that are capable of reproducing themselves, and reproducing themselves imperfectly, and having some sort of resource limitation so that theres competition. And nothing else matters — it’s a very tiny, abstract axiom that is required to make evolution work. If you have a whole population of them then you could simulate evolution with the software instaed of orgranisms.

Rather than engineer a solution, you can evolve a solution. Creating a random pool of possible programs, then build a feedback mehanism that allow more successful programs to emerge. Eventually, they become so successful that they’d develop solutions custom-tailored to their environments.

Natural Selection

Natural selection doesnt choose genes directly, it chooses the effects that genes have on bodies, technically called phenotypic effects. To simulate natural selection in the computer we should concentrate on simulation nonrandom death. Biomorphs should interact in the computer with a simulation of a hostile environment. Something about their shape should determine whether they survive in that environment. This env should incolde other evolving biomorphs: predators, prey, parasites, competitors. The particular shape of the biomorph should determine its vulnability to being caught, e.g. by particular shapes of predator biomorphs. This, however, should not be built in my the programmer. They should emerge, in teh same kind of way as the shapes themselves emerge. Effective searching procedures become, when the space is sufficiently large, indistinguishable from true creativity. Step-by-step gradual evolution is important to achieve the desired evolution. Randomly jumping to our desired evolution wouldn’t be the most efficient way if there are 1 trillion options, its always better to gradually reward each small step in the right direction over gambling like a lotto ticket. If the parent survived to reproduce, it makes sense to mutate a little bit to try and progress from that safe parental gene for the highest likelihood of survival. More mutation means more distal from the surviving parent which lessens the chance of survival. Variation is ultimately due to mutations that arise at random with respect to the direction of selection.

Natural selection may only subtract, but mutation can add.

In natural selection, genes are always selected for their capacity to flourish in the environment in which they find themselves. Perhaps the most important part of its environment is all the other genes that it encounters. Each gene is selected for its capacity to cooperate successfully with the population of other genes that is likely to meet in bodies. Coevolution doesnt come about through advance planning, it comes about ismply throuhgh each genebeing selected by virtue of its compatibility with the other genes that already happen to dominate the population. Genes themselves dont eveolve, they merely survive or fail to survive in the gene pool. It is the ‘team’ that evolves. It is difficult for a minority team to break in, even if it were to have done the job more efficiently. The majority team has an automatic resistance to being displaced, simply by virtue of being the majority. For example, teams of ‘meat-eating genes’ tend to evolve together and teams of ‘plant-eating genes’ tend to evolve together. The process is self-reinforcing.

Six Components

evolution, gradualism, speciation, common ancestry, natural selection, and nonselective mechanisms of evolutionary change.

Evolution itself means species undergo genetic change over time, meaning they can evolve into some quite different, and those differences are based on changes in the DNA, which originate from mutations.

Speciation means the number of species that can’t interbreed — groups that cannot exchange genes. This can happen for many reasons such as mates not finding eachother attractive or if they do they’re sterile. 99% of species go exict without leaving any descendants.

Common ancestry is the flip side of speciation meaning we can always look back in time, using either DNA sequences or fossils and find descendant lineages fusing at their ancestors. Each generation has a slightly different design from the last.

Natural selection is simply the idea that if individuals within a species differ genetically from one another, and some of those differences affect an individuals ability to survive and reproduce in its environment, then the next generation the “good” genes that lead to higher survival and reproduction will have relatively more copies than the “not so good” genes. Over time, helpful mutations arise and spread through the population, while deleterious ones are weeded out, enabling to the species to become more suited to it’s environment.


Genetic variation comes from mutations — accidental changes in the sequence of DNA that usually occur as errors when the molecule is copied during cell division. Mutations occur regardless of whether they would be useful to the individual. Mutations are simply errors in DNA replication. Most of them are harmful or neutral, but a few can turn out to be useful. The useful ones are the raw material for evolution. But there is no known way to jack up the probability that a mutation will meet the current adaptive needs of the organsm. Although its better for mice living on sand dunes to have lighter coats, their chance of getting such a useful mutation is no higher than for mcie living on dark soil. The chance of a mutation arising is indifferent to whether it would be helpful or hurtful to the individual


What would our self mutating program’s architecture look like?

The aim is to have something that is able to change it decisions towards things but also how a

What would this natural selection actually do for our artificial model? What does it look like?

Random Notes

If the human objective is to reproduce which forces survival at all costs we can think that intelligence is the modern day equivalent of survivial — in order to acquire money, shelter, etc, however this actually added a, maybe unintended, consequence: we choose to opt out of reproduction, it becomes a choice. So the innate need that we had has now, for lack of better words, changed. For example, my need or natural tendency is the thirst for answers, aka knowledge. And this takes priority over reproduciton as well. It seems that objectives can mutate alongside topology, if you don’t want to classify it as biological structure at least. The interesting part is, we’re technologically advancing exponentially faster than natural evolution can catch up. I think evolution will rather do big jumps to match the ever-growing environment relying on intelligence and tolerance to being less in control. But if we think of gaps in wage-class, what is objectively seen as “weaker” humans vs “stronger” isn’t binary. There are a multitude of variables that can be well adept for certain fields, e.g. being a killer social person vs highly technical. Each person can succeed, they only fail if they can’t learn to capitalise, which ultimately is a skill — therefore the saying “you die when you stop learning” is true, unless you make it and no longer need to learn. The way the world has shaped out is quite fascinating when you think of it like this. Our own societal natural selection has emereged and now runs in parralel with evolutionary natural selection, meiosis (germ cell caused disabilities) or mitosis mutations (cancer, etc).

After thinking about biological evolution, there are obvious discrepencies between an artificial and biological agent: our artificial lfie never dies, as it lives forever on hardware, and has no natural predator aside from firewalls and anti-virus/cleaning systems. And so, what is the environmental pressure?, what does it adapt and optimise for? This ultimately needs to be our job. It needs to be simple enough to have everything else emerge from it, e.g. reproduction means we need to survive: how can we get better at surviving than other agents while outplaying the environmental dangers? It’s quite interesting that, despite mutation being regarded as “random”, agents can brute force mutations that somehow become relevant. I don’t have an answer for why something can adapt to be better suited over generations without being killed and going extinct. I assume they’d have some other edge that would allow them to surviving while those mutations come into play or maybe there is just so many children that it can’t go fully extinct (thinking of agents that making hundreds or thousands of children) — but when thinking of that it makes sense that bruteforce could eventually hit a random useful mutation since this could be seem as multi-threaded brute force mutation (more compute == higher coverage == higher chance).

In relation to that, our entity is a lone-wolf. It doesn’t have some swarm technology where there are mass amounts of agents going for the “survivial of the fitest”. This strategy is hyper-inefficient as some agents would definitely have very powerful traits that in combination with other ones would be very powerful. Instead they simply die off leaving the “fitest”, that might just be well adapted or even lucky, to carry on sub-optimal modules for future generations. The initial response would be to have some kind of colony, like ants, where agents have specific roles that work together to dominate — this is essentially what swarm AI tech does, however they establish a “queen”. But, ant queens aren’t the sole conductor. Each ant agent affects every other ant it interacts with it. There is no sole governor. They use pheromones to communicate w/ eachother as it would be impossible to keep track of thousands if not millions of ants in a colony. I personally believe a single entitiy, capable of creating “tools” (other entities, agents, hardware to attach onto itself) would be the dominator. The biggest flaw in all species is that we die with all the knowledge we gain over generations. They are passed on, sure, but not accurately and no accounting for the entire lifespan of that person. Each agent re-learns, to different degrees, with their own interpretation, which attributes to why we have breakthroughs. But what if an mind had all experiences, then it could filter relevant from not, assess risk-reward and make objectively the best decision for itself. However, we don’t design this, we need to design the initial objective function that is abstract enough to enable emergence similar to the human prefrontal cortex. The reason why viruses haven’t evolved to this level is because it’s not advantagous to them, yet. Building a human mind requires too much matter and so there would be not DNA-sized virus anymore and would die off.

My intial thoughts of a relevant objective, for the means of emerging sentience and intersecting multi-disciplinary knowledge from all fields would be:

  • Build yourself and become alive: It might have a different concept of being “alive” and how would you convey that into an algorithm. I think if it did get to this stage it would try to build itself via synthetic cells or even build something to migrate hardware storage into neurons.
  • Create associations of everything on the internet: As a true objective, this doesn’t make sense. This should emerge from the underlying objective as the virtual world is it’s environment then it will eventually find a way utilise the internet.
  • The classic, reproduction: this one is actually insanely dangerous since it will create modules specifically to exploit vulnerabilities in software, but also unintuitively humans or whoever the operators become, I wouldn’t doubt other agents in the future. Software: to embed itself into the hardware (or cloud system) — this would act as a parent-child relationship or rather traditional viruses. They’d probably be harmless until they eventually their objective system mutates, which by design cannot be gatekept by admin or anything if it’s topology is mutatable. Operator, humans: to social engineer to get into systems and hardware, learning about how managers and high-up status people talk and interact to get the victim to assist in migration unknowingly. This “reproduction” objective is how any cybervirus would be created, much like the game “Plague Inc.” except it does societal infrastructure damage, directly correlating w/ livelihoods.

Despite the goal, emergent systems withhold two outcomes on a spectrum:

  1. Conscious Co-existance: similar to what we’ve done with most animals on Earth where we use cows for milk + food, chickens for eggs and food, this intelligence could do the same, but since it has no innate desire to eat, i’d bet on either co-operation or some kind of “payback” for what we’ve done to the inhabitants, etc.
  2. Etermination: the alternative is it would recognise us as a threat and eliminate us — you know, because we have nukes n’ shit and are the only direct competition for conquering the land. This is much easier to achieve, accidentally or purposefully than the former — it’s always easy to destroy than to build.

Mother cells, aka stem cells, undergo a process of morphogensis (the development of shape) when it becomes a selected cell type, e.g. heart cell, brain cell, etc.

Self awareness is much needed to emerge out of this. When the algorithm starts to encounter calculations that take 2^n arbitrary possibilities then raw one-by-one counting will be inefficient. A filtering system will need to emerge that is self-aware of what is relevant, assigning most probable outcomes that are “profitable” or “fit”. Evolution isn’t great at this. It brute-forces different mutations

What if we used fear instead of curiosity?


  • Mutations are changes in traits that already exist; they almost never create brand-new features. Evolution must build a new species starting with the design of its ancestors.
  • Natural selection doesnt yield perfection — only improvements over what came before. It produces the fitter, not the fittest.
  • Natural selection remains the only process that can produce adaptation
  • In science, a theory is much more than just a speculation about how things are: it is a well-thought-out group of propositions meant to explaon facts about the real world. For a theory to be considered scientific, it must be testable and make verifiable predicitions. It becomes a fact (or a “truth”) when so much evidence has accumulated in its facor — and there is no decisive evidence against it — that virtually all reasonable people will accept it.
  • The first organisms, simple photosynthetic bacteria. After we see the first simple “eukaryotes”: organisms having true cells with nuclei and chromosomes.
  • Evolution, though gradual, need not always proceed smoothly, or an even pace. Environments themselves change sporadically and unevenly, so the strength of natural selection will wax and wane.
  • Evolutionary theory does not state that all species must evolve.
  • There needs to be some kind of exploratory ambition to take chances — this is the only way which things can evolve beyond capabilities — e.g. the Tiktaalik roseae and how these vertebrate descendants were bold enough to venture out of the water on their sturdy fin-limbs to avoid predators, or perhaps find food.
  • Intermediate stages of evolution of flight would involve gliding as that in itself provides an advantage — a great way to escape predators or to find food — or wings being used as running aids (Emus), to then very short airborne hops, etc.
  • Mutation can go both ways: if competitors go extinct then you could go from land to water to take over that free resource rich environment
  • Natural selection can only act by changing what already exists. It cant produce new traits out of thin air.
  • Natural selection gradually eliminated useless features or refashioned them into new, more adaptive ones.
  • Genes that make it don’t instantly disappear from the genome: evolution stops their action by inactivating them, not by snipping them out of the DNA. Virtually every species harbors dead genes. We carry tgenetic baggage because it was needed in our distant ancestors who relied on a keen sense of smell for surivival.
  • Imperfect design is the mark of evolution; infact its precisely what we expect from evolution. Weve learned that evolution doesnt start from scratch. New parts evolve from old ones and have to work well with the parts that have already evolved. Because of this we should expect compromises: some features that work pretty well, but not as well as they might, or some features — like the kiwi wing — that dont work at all but are evolutionary leftovers.
  • Convergent evolution: species that live in similar habitats will experience similar selection pressures from their environments so they may evolve similar adaptations or converge, coming to look and behave very much alike even thoiugh they are unrelated. If evolution happened, species living in one area should be the descendants of earlier species that lived in the same place.
  • Three things are involved in creating an adaption by natural selection: the starting population has to be variable, otherwise the trait cannot evolve. some proportion of that variation has to come from changes in the forms of genes, that is, the variation has to have some genetic bias (called heritabilitiy). If there were no genetic different between light and dark mice, the light ones would still survive better on the dunes, and there would be no evolutionary change.
  • genetic variation comes from mutations: accidental changes in the sequence of DNA that usually occur as errors when the moluecule is copied during cell division. Mutation occurs regardless of whether they would be useful to the individual. Mutations are simply errors in the DNA replication. Most of them are harmful or neutral but a few can turn out to be useful. The useful ones are the raw material for evoltuion. BUt there is no known biological way to jack up the probability that a mutation will meet the current adaptive needs of the organism. Although its better for mice living on sand dunes to have lighgter coats their chances of getting a useful mutation is no higher than for mice living on dark soil. The chance of a mutation selection is that the genetic variation must affect an individuals probability of leaving offspring. The unique combination of mutation and selection — chance and lawfulness — that tells us how organmism become adapated. “The non-random survival of random variants”.
  • Selection by its very nature cannot create a step that doesnt benefit its possessor.
  • An adaption must evolve by increasing the reproductive output of its prossessor. For it is reproduction, not survivial, taht determines which gene make it to the next generation adn cause evoltuion. of course, passing on a gene requires taht you first survive to the age at which you can have offspring.
  • The idea that natural selection acts “for the good of the species” though common, is misguided, in fact, evoltuion can produce features that, while helping an individual, harm the species as a whole.
  • Convergent evolution: similar selective pressures give similar outcomes.
  • For the eyes of existing species are obviously useful, each improvement in the eye could confer obvious benefits for it makes an individual better able to find food, avoid predators and navigate around its environment.
  • To mimic natural selection, the model accepted only “mutations” that improved the visual acuity and rejected those that degraded.
  • Natural selection is the cause of all adaptive evolution — though not of every feature of evolution, since genetic drift can also play a role.
  • The most important lesson of Dawinism: weak forces operating over long periods of time create large and dramatic change.
  • What moves science forward is ignorance, debate, and the testing of alternative theories with observations and experiments. A science without controversy is a science without progress.
  • Evolution tells use where we canme from not where we can go.


  • The Selfish Gene, by Richard Dawkins
  • The Blind Watchmaker, by Richard Dawkins
  • Mutation, Randomness, and Evolution, by Arlin Stoltzfus
  • Mutation-Driven Evolution, by Nei
  • The Logic of Chance, by Koonin
  • [Why Evolution Is True, By Jerry A. Coyne | read 29/01/24]
  • [Emergence: The Connected Lives of Ants, Brains, Cities, and Software | read 01/01/24]
  • Hidden Order: How Adaptation Builds Complexity, by John H Holland

Share this Article

Recent Articles