As a scientist (and a computer scientist at that) my view is that if we cannot simulate a process we have not understood it properly. I have been following the interesting field of Artificial Life for quite some time and the results are sobering - let me just quote two paragraphs from current overview articles:

One thing that always seems to happen with such projects is that, after they achieve their intended aim, if the ‘evolutionary’ program is allowed to run further it produces no further improvements. This is exactly what would happen if all the knowledge in the successful robot had actually come from the programmer [...]

That is why I doubt that any ‘artificial evolution’ has ever created knowledge. I have the same view, for the same reasons, about the slightly different kind of ‘artificial evolution’ that tries to evolve simulated organisms in a virtual environment, and the kind that pits different virtual species against each other.

Source: David Deutsch (2011): The Beginning of Infinity

One of the earliest networked artificial life experiments was based on the well-known A-Life system, Tierra. This was created in the early 1990s by the ecologist Tom Ray to simulate in silico the basic processes of evolutionary and ecological dynamics. After Ray began his work, he soon recognized the potential of the Web to create a large complex environment in which digital organisms could freely evolve. So he set up a project called Network Tierra to exploit this potential

The results of this experiment were mixed. One goal of Network Tierra was to reproduce the Cambrian explosion in which single-celled organisms on Earth evolved rapidly into multicellular ones and then into more complex animals.

The in silico experiment began with a human-designed multicellular organism consisting of two different cell types. This survived under natural selection, a significant success in itself, but the number of cell types never increased beyond two.

Source: MIT Technology Review (2014): The Curious Evolution of Artificial Life

The point is that I have myself successfully worked a lot with genetic algorithms and genetic programming (I am also teaching this stuff) but what bothers me is that we are still not able to create some abstract form of (co-)evolution inside a computer where some real dynamics take place to produce ever and ever more sophisticated "species".

My question
Are there hints from the biological sciences what this mysterious ingredient could be which we still seem to be missing? Is it physics? Is it chemistry? Is it something else?

Obviously the question is not clear as it stands, so I try a clarification: I refer to complexity of the resulting "species" in artificial life simulations. For example their behavioural or structural complexity. Why do these simulations always get stuck at some very low level (e.g. following food) and never ever even create something as complex as a bacterium? The computing power should be more than sufficient by now - and still, nothing... It seems that only what has been put into the simulation comes out but real evolution produces something really new (this is what the renowned scientist and polymath David Deutsch (University of Oxford) means by "I doubt that any 'artificial evolution' has ever created knowledge.")

Nathaniel gave me a decisive hint in the comments that this problem is called "open-ended evolution (OEE)" in the Alife community and it is one of the biggest research challenges there - unsolved yet! As a starting point see here: https://www.google.de/search?q=%22open-ended+evolution%22&artificial&life

Very interesting that it doesn't seem to bother the biological community and is met even with hostility here (some even lecturing me that the evidence for evolution is overwhelming and thereby implying that I might be some kind of crackpot creationist - unbelievable...)

...and no, the answer is not a matter of opinion (why this question was closed) but a valid research question (hopefully with some good answers someday)!

Last year there was even a big conference on this topic with many interesting results (although the problem itself is still unsolved):

See also my follow-up question here:
If evolution is not about increased complexity, why does so much complexity evolve?


closed as primarily opinion-based by AMR, AliceD, MattDMo, March Ho, Chris Jan 6 '16 at 9:10

Many good questions generate some degree of opinion based on expert experience, but answers to this question will tend to be almost entirely based on opinions, rather than facts, references, or specific expertise. If this question can be reworded to fit the rules in the help center, please edit the question.

  • 3
    $\begingroup$ @Downvoter: It is good practice to state your reasons. How can I improve the question? Thank you. $\endgroup$ – vonjd Jan 5 '16 at 19:12
  • 6
    $\begingroup$ I am voting to close as this question will only generate answers based on personal opinion. I will also be the second downvoter, as we understand and have 150 years worth of scientific evidence to support Evolution and the process of Natural Selection. Furthermore you have cherry-picked two articles that back up your point, and it is not the case that "if we cannot simulate a process we have not understood it properly." Galileo could not simulate the solar system, yet he could prove heliocentrism. Newton and Leibniz could not simulate the infinitesimal, but they could understand it. $\endgroup$ – AMR Jan 5 '16 at 20:46
  • 3
    $\begingroup$ You've answered the question in your first lines: .. my view is that if we cannot simulate a process we have not understood it properly. How can you make a model whilst it is unknown how the first cells appeared and why and how multicellular organisms arose? It ain't magic; if you don't put the parameters in, a model won't generate magic $\endgroup$ – AliceD Jan 5 '16 at 22:04
  • 4
    $\begingroup$ @shigeta: You write in your profile: "I believe that biology and other stackexchange sites should welcome newbies and encourage dialog about science, evolution, technology and etc." Just to let you know I don't feel very welcome here (just look at the comments) and I know how stackexchange works... see e.g. my profile here: quant.stackexchange.com/users - I know it is not your fault but it is a shame for the biology.SE site. Sorry to say... $\endgroup$ – vonjd Jan 6 '16 at 10:44
  • 3
    $\begingroup$ @vonjd Extensive discussion in the answers or below in the comments is indeed discouraged. stackexchange is set up as a question and answer site, not as a discussion site. This is something you can regret in terms of scientific discussions, but these are the rules. You could look into this stackoverflow post for example. Opinion-based questions are seen as unscientifically and not welcomed here on biology. So please, respect the rules of the community. And please stop discussing in the comments and move this over to chat. $\endgroup$ – Chris Jan 7 '16 at 18:48

The question appears interesting and made me think but I might not fully understand it. Let me know if I am answering your question.

Genetic algorithm vs simulation of evolutionary processes

I think that the whole issue comes from a confusion between the concept of simulating evolutionary processes and the use of genetic algorithm (type of optimization algorithm) for various purposes.

Genetic algorithm

Genetic algorithm is a type of optimization algorithm (and the OP knows much more than I do in this field) aiming to find solutions to search problems. The accuracy of the analogy between a genetic algorithm and the biological reality that inspired such algorithm is completely irrelevant to the usefulness of the algorithm at doing a specific task (such as the NP-hard travelling salesman problem for example).

Numerical simulations in science

I think your question is not specific to evolutionary biology but rather to science as a whole (this leads me to think that Philosophy.SE would be a good place to ask such question).

In natural sciences (Physics, Chemistry, Biology and others), we model things! We abstract the essentials from a complicated world and model it. When we model, we assume a number of properties of the system of interest. These assumptions might be extremely well documented and verified or not. When the assumptions of a model are not well documented, it is of course essential to study a posteriori the robustness of the model to violation of the assumptions and to consider the results of the model with a pinch of salt. A model can be purely verbal or most often expressed in mathematical formulations. However, many complex systems cannot be modelled mathematically (even for the most brilliant mathematicians). This is where numerical simulations come into play. Note that once a process has been modelled, we empirically investigate the accuracy of our model by formulating predictions and testing them.

You say:

if we cannot simulate a process, we have not understood it properly

If we already understood a process, there is no point spending time and money to simulate it anyway! So again, this sentence suggests that numerical simulations is worthless in science. It is true though that we can only simulate the processes for which we know the basic components (but we might not understand the dynamic of a system of interest).

Simulations in Evolutionary biology

You cite one work (which I am not familiar with) which fail to reproduce the observed pattern. In other words, the predictions of the model are not met/observed in reality.

As I said above, one needs to understand the basic components of a system in order to be able to simulate it. We happen to already know a faire amount of stuff! Of course, it is impossible to address the question "what do we know in Biology" as it would be way too broad. There are thousands of studies that have used numerical simulations (and also mathematical simulations) to study evolutionary processes.


Imagine for example, you are interested to know the probability for a given new neutral mutation to rise in frequency in a diploid population to reach "fixation" (that is a frequency of 1; everybody then carry this mutant allele). There exists a number of mathematical models (Wright-Fisher (binomial) model of genetic drift, Moran (Birth-death) model and Coalescence (branching process) model) to calculate this probability but let's assume we fail to develop such mathematical/analytical model and and we need to simulate it. We could simulate this process a lot of time (using a ABC kind of approach) and calculate the expected probability of such mutant allele to get fixed. Btw, this probability is $\frac{1}{2N}$, where $N$ is the effective population size.

Want to know more?

I am not a philosopher of science (but a PhD student using numerical tools to model evolutionary processes) and I think the question is not specific to evolutionary biology. I would recommend to ask the question What is usefulness of numerical modelling in science? or Are numerical modeling worth as much as analytical modelling in science? on Philosophy.SE.

If you do so, can you please link to your posts here, I would love reading the answers. If you don't ask these questions on Philosophy.SE, I probably do it at some point and will add the links here.

  • $\begingroup$ Thank you, I think your answer goes definitely into the right direction (and I upvoted it). My main question is why has no artificial simulation ever been able to really create something sophisticated but seems always getting stuck after reaching some low level of complexity? $\endgroup$ – vonjd Jan 5 '16 at 20:57
  • $\begingroup$ [...] reaching some level complexity. Do you refer to the complexity of the model? $\endgroup$ – Remi.b Jan 5 '16 at 21:01
  • $\begingroup$ I refer to complexity of the resulting "species" in those artificial life simulations. For example their behavioural complexity. I mean why do these simulations always get stuck at some very low level and never ever even create something like a bacterium? The computing power should be more than sufficient by now - and still, nothing... $\endgroup$ – vonjd Jan 5 '16 at 21:18
  • $\begingroup$ See also my Edit at the bottom of the question. Thank you again. $\endgroup$ – vonjd Jan 5 '16 at 21:25
  • $\begingroup$ Oh, so your question is less general as I thought. You are talking about a few quite specific type of simulations apparently. You should clarify the definitions of these simulations. Are the authors of those simulations particularly interested in the evolution of cell cooperation in a multicellular organism or maybe they are interested in the evolution of a complex genetic network... or maybe something else. $\endgroup$ – Remi.b Jan 5 '16 at 21:33

Agree with the previous answer.

Are there hints from the biological sciences what this mysterious ingredient could be which we still seem to be missing? Is it physics? Is it chemistry? Is it something else?

The OP already seems to support evolution theory, as anyone with basic biology knowledge would do.

Since he is asking about possible "mysterious ingredient", the question is very likely to be regarding the stimulation of the evolutionary process rather than generic algorithm.

Even more specifically, he wants to stimulate the evolution to know if "Probablity theory" will support evolution theory without any need for the "mysterious ingredient".

As answered above, without fully understanding all the components of the system, it may be difficult to stimulate an evolutionary process. There is not even need for that.

But if you want to test if complex characteristic can be achieved by chance you can stimulate it easily by some other method.

Develop a program which has "face detection" (from the image)function, and add some other functions such as self replication, forced "mutation", and an enviroment which will select the fittest. Try super computers where your software will self-replicate "unlimited" times at a second, and consider yourself successfull when your program gain a newer function such as "sex" detection from the or image after several years (assuming sex detection feature will make the program "fitter" at your enviroment)

  • $\begingroup$ "The OP already seems to support evolution theory" - why would anyone doubt it? I just want to fully understand it. Anyway, could you please clarify your last paragraph, how exactly would you go about? $\endgroup$ – vonjd Jan 5 '16 at 22:31
  • 1
    $\begingroup$ I have read some comments under your question, and thought that some people may think you do not support evolution because of lack of stimulation. I am sorry if i misunderstood this. Anyway i found your question very interesting and already upvoted it. I agree with you. $\endgroup$ – TeoFriendly Jan 5 '16 at 22:55
  • $\begingroup$ Please see my follow-up question: biology.stackexchange.com/questions/42050/… $\endgroup$ – vonjd Jan 6 '16 at 10:06
  • 1
    $\begingroup$ @vonjd nice question. i will comment. I have some computer programming background i generally share your feelings about improvments $\endgroup$ – TeoFriendly Jan 6 '16 at 11:05
  • $\begingroup$ Algorithmic depth and processing power are miniscule compared to the algorithms of life. The 70 common chemical elements have Van-Der-Valls force, flow, dissolution, 70^70 simple combinations, a google number of proteins, The best mathematicicans can't even model a single tree... it takes them months to program diatoms, leaves, and they always fail. trillions of gygabytes database and algorythms that cover many cd's are used in biology, and for the moment we rival them to about a billionth of their numerical depth. $\endgroup$ – com.prehensible Jul 22 '17 at 10:18

Not the answer you're looking for? Browse other questions tagged or ask your own question.