Darwin was wrong!!!!!

Stop the presses!

Seriously, are the ID proponents at UD ever going to wonder why Gould and Eldredge remained persuaded that common descent occurred, and that “punctuated equibrium”, although contrary the uniformly incremental pattern that Darwin envisaged, was nonetheless consistent with Darwin’s proposed adaptive mechanism of heritable variation in reproductive success?

Because Darwin was indeed wrong about uniform change.  Unlike us, he didn’t have computers with which to model the predicted output of his mechanism. Indeed he didn’t even know what the vector of heritability was.  We do.  Here’s a sample output from Eureqa, a program that uses Darwin’s proposed mechanism to “evolve” equations to fit data:

Look at the bottom left plot.  It records the best-fitting equations as they evolve.  On the vertical axis is the “error” in the evolving equations – the better the fit, the smaller the error.  On the horizontal axis is the complexity.  The program is set up so that complexity carries a penalty in terms of reproductive success but accuracy carries a reward.  The most efficient equations – those that give best accuracy for least against complexity –  are shown more green, while less efficient equations are shown more orange or red.

What happens over time is that as the equations evolve, there are discontinuities in the best error rate: note the step changes on the vertical dimensions.  From time to time a small change in the equation will occasionally introduce a large improvement in accuracy.  However, reductions in complexity tend to be more gradual.  As a result, we see “punctuated equilibrium” – step changes in accuracy followed by gradual reductions in the equation’s complexity.

And the system is entirely Darwinian.  Darwin couldn’t know that this is what his theory would actually predict.  Of course he was right that adaptation would be incremental – and it is – but it is more incremental at the genomic level than at the phenotypic level.  A small DNA change can result in quite a large phenotypic change.  Again, Darwin could not know this.  But, even phenotypic changes are gradual.  The key point is that the rate of change is not uniform – indeed, uniform rates of change turn out to be very unlikely under the Darwinian mechanism.  If one of the evolving equations in Eureqa gets a good “idea” (as in AVIDA) then there is very rapid change for a while as the population optimises itself to this newly available resource, followed by diminishing returns as a local maximum in accuracy is approached. Until something else happens – a novel mutation along yet another dimentions allows the population to exploit a whole new resource, following which,  again, stasis is approached, and is maintained as long as the resource remains, and the population is not outcompeted by another lineage.

And that’s before we even consider that small populations will tend to adapt faster than larger ones – or die.

h/t to whoever introduced me to Eureqa!  It’s brilliant, but I’ve forgotten who it was!

 

85 thoughts on “Darwin was wrong!!!!!”

  1. RichardthughesRichardthughes

    coldcoffee: Data available here
    Will try.

    Awesome – and kudos for rolling up your sleeves. And just so we’re clear up front, you can’t have any constructs of physics pre-loaded. It must find them itself.

  2. RichardthughesRichardthughes

    Neil Rickert: Joe seems to think the Eureqa coders agreed with him.However, my read is that they sidestepped his question.

    Full disclosure, I know then personally, love them and have hosted them in our offices for a working session for 2 days.

    Symoblic regression / evolutionary computation. Uses mutation and selection to evolve better and better solutions each generation. I don’t think they want to frame it as as “Darwinian evolution” as unfortunately its a bit of a hot button in the US.

    Don’t take my word for it, take Morgan Freemans!
    http://fast.wistia.net/embed/iframe/x0t6owlf6k?popover=true

    And Joe isn’t smart enough to understand how it works.

  3. RichardthughesRichardthughes

    Lizzie: Well, he doesn’t say what his question was.

    I think he may have updated:

    “Good day,

    On the ineternet people have been discussing Eureka. Unfortunately some over zealous activists of evolutionism are saying that Eureka models darwinian evolution. My question to you is does it? Does Eureka use an eliminative process to get its results? Does Eureka use undirected changes to achieve its results?

    To me Eureka doesn’t have anything to do with darwinian evolution. Could you help?

    respectfully,

    Joe G”

    “Unfortunately some over zealous activists of evolutionism ”

    and anyone who watches their video can see the answer to “Does Eureka use an eliminative process to get its results?” and “Does Eureka use undirected changes to achieve its results?”

    I’m not surprised they brushed off his poorly written and rhetoric filled email.

  4. Lizzie

    Richardthughes: Full disclosure, I know then personally, love them and have hosted them in our offices for a working session for 2 days.

    Symoblic regression / evolutionary computation. Uses mutation and selection to evolve better and better solutions each generation. I don’t think they want to frame it as as “Darwinian evolution” as unfortunately its a bit of a hot button in the US.

    Don’t take my word for it, take Morgan Freemans!
    http://fast.wistia.net/embed/iframe/x0t6owlf6k?popover=true

    And Joe isn’t smart enough to understand how it works.

    Awesome!

    Joe, if you are lurking, watch the vid!

  5. RichardthughesRichardthughes

    His post keeps evolving!

    “ETA- No Eureqa does not use undirected processes. It is all directed towards the solution, ie the final equation”

    Yes, its directed toward that thing the programmers don’t know. By selection based on fitness. How ID!

    .”And Eureqa does not use reproduction. It mutates the surviving equations.”

    Joe, you have *no clue* how it works.

    http://hplusmagazine.com/2011/03/25/eureqa-signs-of-the-singularity/

    ” The majority of these equations are sheer nonsense, but by chance some fit the data a little better than others. In an analogue of sexual reproduction, the software saves these equations for ‘breeding’, combining one half of a ‘father’ equation with one half of a ‘mother’ equation. Sometimes, it alters a term in the equation, to mimic random genetic mutation. Over thousands of generations, equations emerge that fit the data quite well.”

    If “wrong” were an event at the winter Olympics, Joe would take gold for Tardania.

  6. Lizzie

    Richardthughes: I think he may have updated:

    Cool.

    “Good day,

    On the ineternet people have been discussing Eureka. Unfortunately some over zealous activists of evolutionism are saying that Eureka models darwinian evolution. My question to you is does it?

    No. It uses it.

    Does Eureka use an eliminative process to get its results?

    Yes. It eliminates the less good equations, and keeps the better ones to breed.

    Does Eureka use undirected changes to achieve its results?

    Yes, AFAICT, the changes are random – the program doesn’t try to second guess what will work better.

    To me Eureka doesn’t have anything to do with darwinian evolution.

    It works exactly as in Darwin’s proposed mechanism.

  7. petrushka

    I think you have to realize how discomforting this is to IDists who deny that evolution can produce complex structures unless the selector contains the desired result, ala weasel. Avida is cool, but doesn’t do anything commercially useful.

    If Eureka actually invents valuable equations that are not obvious and not previously known,it is a major threat to Behe and Dembsky.

  8. Lizzie

    Well, Dembski knows this, which is why he’s moved to “search for a search”.

    He knows that the Darwinian algorithm works much better than blind search as long as genotype and phenotype are correlated, in other words if the landscape is smooth.

    He also almost certainly knows that high dimensioned landscapes (which this has, and life has) have many more routes to “targets” (silly term, but it’s his).

    The trouble for Dembski is that he also sees the problem with the fine tuning argument, which is the brick wall that his search-for-search-for-a-search argument eventually comes up against.

    His next book will be interesting. It looks as though he may have embraced teleonomy.

  9. Joe FelsensteinJoe Felsenstein

    Lizzie:
    Well, Dembski knows this, which is why he’s moved to “search for a search”.

    He knows that the Darwinian algorithm works much better than blind search as long as genotype and phenotype are correlated, in other words if the landscape is smooth.

    He also almost certainly knows that high dimensioned landscapes (which this has, and life has) have many more routes to “targets” (silly term, but it’s his).

    The trouble for Dembski is that he also sees the problem with the fine tuning argument, which is the brick wall that his search-for-search-for-a-search argument eventually comes up against.

    His next book will be interesting.It looks as though he may have embraced teleonomy.

    I don’t think we need to be concerned about a “brick wall”. Search For a Search concedes that evolutionary mechanisms may well work. It just argues that a Designer is needed for the conditions that enable that. That would be, for the laws of physics.

    I don’t think biologists need be much concerned about Who set up those laws, as long as their workings allow evolution to proceed. SFS is not an argument against the ability of evolutionary mechanisms to explain the adaptations we see.

    Others concerned with cosmology or theology can worry about the origin of the laws of physics.

  10. Allan Miller

    Richardthughes,

    Morgan Freeman – he’s God!

    Population, mutation, selection, breeding, multiple generations … yep, it’s all there. I wonder what information is smuggled in … oh yeah, the attractor. The actual point of the exercise.

    Hilarious that ‘Darwinian’ processes, modelled entirely on what goes on at the ‘micro’ scale in nature, cannot ‘do anything’ (gain information, cause evolution, adapt, increase fitness etc) in those models! And if they can, they don’t work in Nature, ‘cos it’s different.

  11. RichardthughesRichardthughes

    petrushka:
    I think you have to realize how discomforting this is to IDists who deny that evolution can produce complex structures unless the selector contains the desired result, ala weasel. Avida is cool, but doesn’t do anything commercially useful.

    If Eureka actually invents valuable equations that are not obvious and not previously known,it is a major threat to Behe and Dembsky.

    I think it’s great. Their very fair with academic licences and it can help make scientists of all of us. They even have an excel front end / plug in so you can put your Ticks & Watermelon data in to a table and watch the magic happen.

  12. Lizzie

    I’ve been experimenting with it to find out which combinations of neuroimaging variables best predict which diagnostic group patients fall into.

    Support Vector Machines are the usual approach, but the beauty of Eureqa is that it potentially finds an underlying principle, rather than the hyperplane, which might give you a good separator, but doesn’t tell you anything (well much) about why it works.

  13. RichardthughesRichardthughes

    That’s cool. So will you be published? Would you like to be one of their case studies?

    *editz*

  14. coldcoffee

    Richardthughes: And just so we’re clear up front, you can’t have any constructs of physics pre-loaded. It must find them itself.

    two hours on, Mathematica is unable to give me a sensible model which predicts the next 20 points of axes. I give up. Eureqa is better in solving these types of problem.

  15. Lizzie

    coldcoffee: two hours on, Mathematica is unable to give me a sensible model which predicts the next 20 points of axes. I give up.Eureqa is better in solvingthese types of problem.

    It’s interesting, isn’t it? Evolutionary processes are extraordinarily powerful, as long as you have time to go through the necessary number of generations.

    In some ways our own brains work on the same principle, but because it is so lengthy, we take shortcuts by second guessing the next tweak. This means that we avoid unpromising lines of investigation. But clearly, sometimes the kinds of lines of investigation that an human person would reject as too unlikely lead anywhere useful are the very lines that get to a brilliant solution. The very blindness of evolution – its capacity to pursue any line of investigation however unpromising an intelligent designer would think it means it can access areas of “solution space” that we would reject.

    Which is why my own view of the “ID” argument is that in one sense ID is correct – certain patterns, phenomena, have the hall mark of something pretty remarkable. We could call it “intelligence” or we could call it something like “recursive optimisation process”, and both human beings and evolutionary processes have it. What we have, and evolutionary processes don’t, is foresight, and therefore the capacity for intentional behaviour. But it turns out, I’d say, that what foresight chiefly provides is speed. If speed isn’t an issue, the non-foresighted version does as well, and sometimes better.

    Although it does have the disadvantage of not being able to reuse solutions from one lineage in another. It can only retrofit. Which is one of the hallmarks, I would say, of design lineages that are produced by non-foresighted processes (and examples are the giraffe laryngeal nerve, or the human birth canal) as opposed to foresighted intelligence like our own, which can add a camera to a phone, or floats to a plane.

  16. RichardthughesRichardthughes

    ColdCoffee, All I want to do at this point is congratulate you for doing the work. That deserves respect. We may never agree on ID, but you put a shift in. Kudos.

  17. Lizzie

    Meanwhile, Eric Anderson writes:

    The proof is in the pudding.

    Naturalistic evolution — allegedly — is able to produce not only intricate, functional machines, but the most sophisticated coding and storage system known. Indeed, it is alleged to have produced virtually everything we see.

    Yet the evolutionary algorithms aren’t being used in a widespread manner to create wonderful new things. The retort that “evolution takes lots of time” doesn’t cut it; time is precisely the thing that computers help us to deal with, running through generations in a matter of seconds.

    No, the real issue is that evolutionary algorithms are useless for generating real novelty. They are only utilized in very narrow situations in which careful parameters and constraints have been put in place, based on intelligent input.

    We have a real-world situation crying out for new innovation, new code, new inventions. Yet there the evolutionary algorithms sit — those programs that allegedly prove the power of evolution and are themselves an example of evolution in action — bringing essentially nothing to the table.

    Someone wake me back up when an evolutionary algorithm, one operating on Darwinian principles and not under the careful tutelage of an intelligent person, writes a meaningful novel or produces a play or writes a useful Android app.

    The proof is in the pudding.

    But you have to actually eat the pudding, Eric 🙂 No good sitting there at UD claiming “there is no pudding”.

  18. Allan Miller

    Lizzie,

    Yurgh. Since the selective agents adjudicating upon plays and apps are human, but we are not allowed to include them in the process, the challenge is (of course) beyond the scope of the process.

    Eric may wake me up when he’s designed an entity that can do those things. Preferably one made out of CHNOPKS and a few trace metals.

  19. petrushka

    A couple of points.

    I suspect there are problems that Eureqa will not solve. I have no evidence for this, but even the best human scientists and mathematicians are limited. So I would not expect Eureqa to solve everything.

    It is true in some sense that science advances by funerals. Not because scientists can’t be persuaded by evidence, but because some kinds of innovation are historically done by young people. It’s not a matter of stubbornly holding on to obsolete ideas. it’s more a matter of young minds being better at innovating.

    AI is in its infancy. I suspect that there will never be a universal problem solver. I suspect that any device or software that can solve problems traditionally solved by humans will be susceptible to the same failings as humans.

  20. Lizzie

    Yes, it seems like that Eureqa won’t solve some problems – just as evolution can’t.

    Or rather, evolution only can if what evolves are problem solvers! So yes, Eric won’t get his meaningful novel or theatre production, because Eureqa isn’t set up to evolve people, and you need people to do these things.

    But he might well get his Android app. Not from Eureqa but from other code-writing evolutionary algorithms.

    The goal posts are moving at an alarming rate!

  21. petrushka

    We already have passable music generated by programs.

    The definition of a copyrightable melody is such that it is possible for a program to create all possible melodies, and in fact, I think it has already been done.

  22. BruceS

    On a local educational TV station a while back, I saw Hod Lipson give an entertaining talk about how his team created Eureqa and how the program works. It’s an hour-long, semi-technical presentation. It’s here for anyone who is interested:
    Hod Lipson on Eureqa

    Worked better for me if I max the slides into a separate window.
    Interestingly, Eureqa involves co-evolution: as the model evolves, Eureqa also evolves separate experiments which try to challenge the model where it is weakest.

  23. velikovskys

    Petrushka

    It is true in some sense that science advances by funerals. Not because scientists can’t be persuaded by evidence, but because some kinds of innovation are historically done by young people

    Then science advances because of births which provide a source of new “data”

  24. petrushka

    Births of new scientists and deaths of old ones are approximately equal.

    My point is that there not a lot of documented cases of science being held up by crotchety old men who refuse to accept evidence. It’s more about crotchety old men having fewer great new ideas.

    I read somewhere that physicists peak at about age 25.

  25. petrushka

    The young mathematicians’ experiences are representative of a larger trend, according to Mr. Simonton. In a study of nearly 2,000 famous scientists throughout history, he found that mathematicians were the youngest when they made their first important contribution. The average age at which they accomplished something important enough to land in history books was 27.3. By contrast, biologists were 29.4 years old, physicists were 29.7, and chemists were 30.5.

    But starting at a young age doesn’t necessarily mean one’s career will end early or that later contributions will pale in importance — the second half of the legend. In fact, Mr. Simonton found that mathematicians make their best research contributions (which he defined as the ones mentioned most often by historians and biographers in reference books) at what many might consider doddering old age: 38.8. That age is very similar to those he found in other sciences: 40.5 in biology, 38.2 in physics, and 38.0 in chemistry.

    When you’re my age, 40 looks juvenile.

  26. Lizzie

    Quite a lot of schemes have changed their name from “Young Research …” to “Early Career ….”

    Which is helpful to late starters like me 🙂

Leave a Reply