God and Darwin
There is a fascinating conversation going on between Ross Douthat, Daniel Larison, Andrew Sullivan and our own Alan Jacobs on the general topic of religion and evolution.
Larison argues that many religious doctrines do not exclude evolution:
Tell the Christian that he must either endorse evolutionary theory or accept the Bible, and he will typically take the Bible, especially if he is not grounded in an authoritative teaching tradition that tells him that this choice is a false one.
Interestingly, one can show that this is a false choice when arguing from a purely scientific / rationalist perspective as well. In a somewhat technical article in National Review, I argue that one can derive the following three statements from the mathematical structure of the evolutionary algorithm:
1. Evolution can not begin ex nihilo, but instead requires pre-existing building blocks, both in the form of an initial population of objects and in the underlying rules of the algorithm itself. That is, it does not solve the problem of Creation.
2. Evolution is a device designed to find a specific needle in a haystack – the most fit potential genome. That is, it has a goal
3. All of the information required to know the goal of the process is embedded in the combination of the definition of the possible genomes and the laws of chemistry that determine the fitness of each possible genome. That is, the goal is knowable before the process of evolution begins.
One needs no reference to any religious text of any kind to understand that the idea that evolution implies atheism is false.
When it comes to applying this reasoning to political decisions, Andrew says that:
If a candidate cannot accept Darwinian evolution, then I simply lose all respect for him or her. I do not trust their empirical judgment, which means I don’t believe their political decisions will be affected by, er, reason.
While I think the second half of the latter sentence goes a little far, in the end I agree with this. I doubt that, in practice, I would ever vote for a candidate who rejects evolution.
That said, I find the rejection of evolution by a thoughtful Christian candidate to be more poignant than risible (and I suspect Andrew shares this reaction). Further, since evolution doesn’t make many everyday-observable predictions, it remains a pure abstraction for almost all politicians, and therefore I’m not sure that “rejecting evolution” will necessarily predict any especially wacky future policy decisions. This is why I agree with Larison that one’s reaction to evolution acts, at least in part, as a social marker – try asking 99% of the smirking journalists who ask candidates these questions about any technical aspect of the relevant science, and get ready for the mouth-breathing.
Nevertheless, in my view Andrew’s bottom-line is correct: rejecting evolution remains a correctable error in reasoning that should be corrected.
<i>2. Evolution is a device designed to find a specific needle in a haystack – the most fit potential genome. That is, it has a goal</i>
just a specific note, this describes adaptive evolution, that is, via natural selection. a lot of evolution may be neutral in regards to fitness (on the level of molecular evolution it is probably mostly neutral).
— razib · Jan 10, 05:42 AM · #
Razib:
Thanks for the comment. I agree. I am referring here to evolution via natural selection (in the linked article, I actually specificy the genetic algorithm that I use to make these three claims, and it is a mathematical abstraction from actual bichemical processes).
— Jim Manzi · Jan 10, 06:15 AM · #
Jim, I think you have a technical problem there. Your argument seems to show that evolution has a deterministic path, but you’re describing that as a goal. I don’t see how goal is the right word there—it’s got a bit of baggage.
I was going to write ‘endpoint’ instead of deterministic path, but as you are clearly aware from what you say in the article, it’s possible that there’s no equilibrium that is converged to, since an organism’s fitness is partially determined by it’s environment, which itself changes as evolution progresses. So that’s another way in which the idea of a goal might not be wholly apt.
— Justin · Jan 10, 06:23 AM · #
Justin:
Thanks for the great questions.
I chose the word “goal” in the article deliberately.
Think of the goal as the fittest of all possible genomes. Think of the path as the sequence of tested genomes that are produced in all generations.
I was arguing not that evolution has a deterministic goal and path, but that the evolutionary algorithm itself does not create any incremental indeterminacy over and above that which is inherent in the physical laws to which it can ultimately be reduced (e.g., quantum mechanistic uncertainty).
The point that I was trying to make about the changing fitness environment is that these changes are themselves merely the product of other physical processes, so while it is inconceivable to calculate how they interact over time, they are in principle as deterministic as those physical laws. Therefore, scientists are justified in proceeding as if there is no goal, while at a philosophical level there is a knowable goal and path (or to be more precise, that the evolutionary algorithm itself does not create indeterminacy about the goal and path over and above what is inherent in these underlying physical laws).
— Jim Manzi · Jan 10, 07:09 AM · #
I’ve always felt that perception of a contradiction between scientific and biblical accounts of creation betrays a lack of imagination. God is omnipotent, and isn’t bound by linear time. If He wanted to create a man at a specific place and time, he could alter the beginning state of the universe to make that happen. Or perhaps the universe would rearrange itself backwards in time in order to conform to His will. Either way, there would be no contradiction between millions of years of evolution and “‘let there be x’ and there was x”.
I’m not a theologian, but it seems sufficient imagination about the nature of the divine would resolve this nagging issue.
— Chris · Jan 10, 08:01 AM · #
jim, cool. as for the goal, that’s the major hinge in your argument where semantical quibbles come up. i’m roughly satisfied with your exposition, though i’m not convinced strongly on the question of historical contingency and waht not (though you make it hard not to agree in <b>principle</b> that the historical contingency only seems capricious because of our non-omniscience). do note though that in <i>ancestor’s tale</i> richard dawkins made arguments quite close to yours, and aligned himself with simon conway morris (an anglican paleontologist) in suggesting that selection tends to result in particular optimal morphs. given infinite time and infite populations makes selection very powerful indeed…but a large group of workers have started arguing for the critical role of stochasticity in genomic evolution for lower effective pop size organisms.
— razib · Jan 10, 09:34 AM · #
oh, and generally of course i don’t think that evolution necessarily implies atheism. that’s just empirically false, and i am as convinced by the apologia that dawkins makes for this position as i am by arguments for the existence of god.
— razib · Jan 10, 09:38 AM · #
I think both Jim and Razib are grossly mistaken as regards the “goal,” and I’ve got a lot to quibble with in Jim’s article and in comments above. For what it’s worth I’m a biomedical professional (and, oddly, work with a lot of biomedical professionals who don’t believe in evolution) and I’m what people generally call religious.
This “goal” stuff, from both of you, is crapola. I’ve been involved with a lof of “directed evolution” work where people actually try to evolve biological function. Now, for one thing, the “fittest” genome isn’t really a well-defined concept. For another, the process doesn’t necessarily get you to the best solution anyway — it’s in some ways analogous to a thermal annealing, you can get trapped by partial but nonideal solutions because of cross-talk among mutations. You almost certainly never get to a real optimum because in the contexts we’re talking about even Jim’s 2^100^ is podunk: Dawkins’ “damn near infinite” is actually more useful to think about it. And most importantly, you don’t necessarily get to your answer in a way you want — this is where Jim’s thinking about genetic algorithms leads him astray.
If you do directed evolution with organisms in a lab, well, evolution’s a bitch. You try to set up the problem as clearly as you can — say, the organism needs to improve how well it catalyzes a particular reaction essential to its growth, by evolving a protein that catalyzes that reaction poorly into one that does it better: this is the nut of most industrial and academic directed evolution. Now, in the case of a genetic algorithm, you can reduce and simplify screening your answers so they do what you want. But in the cell, man, it comes up with every goofy way of solving the problem you hadn’t anticipated: overexpressing the poor protein via freakishly improbable mutations, downregulating something else, or coming up with a chemical solution which works in the very very specific lab selection but not really outside it in the general way you were seeking — the upshot being it’s really, really hard to determine what the evolutionary “challenge” really is and to think about what responses to it can be. Genetic algorithms are a terrible way to think about real live genetic evolution.
And it’s worse than that, really, for talking about a “fitter” genome as a goal, because frankly evolution changes environment, or changes the parts of the environment which are relevant to the selection: your survival in a room isn’t determined by the same factors that determine whether a thrush or an ant or a bonobo can survive in it. This “fitter” genome concept is just goofy. Evolution doesn’t have “goals” that way.
And when biologists say there’s no goal, they mean something profounder than that anyway — odd that Jim goes off to the philopsophers in the preceding point but glosses over it here. They’re saying, there’s no telos: you’re not trying to evolve towards something. It sort of goes back to what I was saying above: evolution doesn’t really give a rip how it solves a challenge. It doesn’t even care if instead it solves some other challenge or changes the rules of survival for the organism. The organism is responding immediately to immediate forces: it’s analogous to the idea of a virtual photon, say: that there aren’t long-range forces in physics either. And that’s where the “scientific atheists” are biting (I have issues with it, and reasons I think they’re wrong to make theological conclusions from that: but they’re just right about the “no goal” thing and Jim’s response on it doesn’t wash.) There’s no real plan.
And it gets worse than that too because remember Darwinian style evolution isn’t exactly, well, right. You can show for example that bacteria undergo something like Lamarckian evolution, reaching harder for solutions in certain parts of their genomes nutrient starved or under phage pressure or whatever. That’s casting about like mad without all this nice steady Darwin stuff, no?
And that third point is beyond silly. For one thing it’s aphysical. In that article Jim writes in fact, even the “random” elements of evolution that influence the path it takes toward its goal—for example, mutation and crossover—are really pseudo-random. For example, if a specific mutation is caused by radiation hitting a nucleotide, both the radiation and its effect on the nucleotide are governed by normal physical laws. Umm, yep — and those laws say they’re random. That’s exactly a quantum thing.
For another thing, the chemistry of the genomes itself changes — this goes back to the really nonseparable nature of genes and environment I tried to hit on above. And that chemistry can change in really, really surprising ways, and again there is really a fundamental randomness that will get amplified in this process. Sorry, Jim (and Dirac), there’s not an equation you can write and from then on everything is billiard balls: at least, that’s not how science views the world right now. That’s just wrong.
I’m sorry, I think there’s too much to say but you both are clearly savvy enough to grab where I am. And I think what Jim’s saying here is (1) unscientific and (2) built on sand.
— Sanjay · Jan 10, 04:02 PM · #
Razib:
I think Conway Morris has argued for a point of view that is more aggressive than what I propose here (roughly speaking, that we can have some intimations of what the goal of evolution via natural selection might be), but that he only argues it is plausible, as opposed to demonstrable.
I have argued in this article for a narrower point of view (that such a goal exists, even though I make no statements about it or its relation to humans), but on the other hand, that a goal must exist, not that it is merely plausible.
Note that I don’t consider these two points of view to be mutually exclusive.
Further note that when I say “must exist”, this is subject to two caveats. First, as per the prior comments, that it is more precise to say that the evolutionary algorithm adds no incremental indeterminacy over and above what is already present in the physical laws to which it must ultimately be reducible. Second, and yet more foundationally, all human knowledge, scientific or otherwise, is contingent.
— Jim Manzi · Jan 10, 04:17 PM · #
Human consciousness is the product of evolution. Evolution does not produce perfectly fit systems; it merely eliminates, through natural selection, systems that are too unfit to survive. There is no reason to believe that evolution has (or even can) bless humans with a consciousness that is capable of perfectly grasping any ideas or concepts about the “real world” whatsoever. If it was beneficial for survival, evolution would lead the human mind to perceive reality in a way that is fundamentally at odds with the external world and its makeup. I can imagine many ways, in fact, that it would be more beneficial for the human consciousness to be deluded about what it is perceiving around it— or, at least, that it would be more beneficial for the human consciousness to have blind spots that it is incapable of perceiving or cognating.
In other words, I think it’s a good idea to recognize the very real possibility (and I would say the fact) that science will never have a perfect grasp on “objective reality”, that even those things that humans take as the absolute scientific truth may be flawed in ways that we are incapable of perceiving, and in light of that, there is no intellectual virtue greater than humility and the acknowledgment of one’s own fallibility. That is a lesson that the antitheists such as Dawkins could really stand to learn.
— Freddie · Jan 10, 04:43 PM · #
Sanjay:
Thanks for your very detailed and thoughtful comments. You make so many good and deep comments, that I’ll try to react to each of them one-by-one:
I think both Jim and Razib are grossly mistaken as regards the “goal,” and I’ve got a lot to quibble with in Jim’s article and in comments above. For what it’s worth I’m a biomedical professional (and, oddly, work with a lot of biomedical professionals who don’t believe in evolution) and I’m what people generally call religious.
This “goal” stuff, from both of you, is crapola. I’ve been involved with a lof of “directed evolution” work where people actually try to evolve biological function. Now, for one thing, the “fittest” genome isn’t really a well-defined concept.
Fitness is extremely well-defined in the chemical plant example I provided in my article. There is a spectrum of “well-definedness” that goes from this example all the way to evolution via natural selection in the wild. I think it would be more precise to say that in your expereince the fitness function was not well-defined by the experimenter, that is, the failure is one of the knowledge of the observer, rather than something inherent to the phenomenon (of course, you did these experiments, not me, so correct me if I’m wrong).
For another, the process doesn’t necessarily get you to the best solution anyway — it’s in some ways analogous to a thermal annealing, you can get trapped by partial but nonideal solutions because of cross-talk among mutations.
As I’m sure you know, based on your comments, all search algorithms are really clever algorithms that let us search a space of possible solutions intelligently by trying only a subset of all possible combinations (in this case “combinations” = “genomes”). Therefore, by definition, we can never be certain that the “best solution yet found by the algorithm” is really the best single possible solution. In fact, in a very large search space, we can normally assume it is not the single best solution.
Your analogy to thermal annealing is an excellent one; in fact in AI research simulated annealing algorithms are often used as an engineering alternative to genetic algorithms. As you have pointed out, all algorithms must balance the need to home in on precisely the best solution once it appears we are in the right neighborhood vs. avoiding the problem of settling on a local optimum (in slang terms, how “greedy” should the algorithm be?).
This is why I was very careful in the article to say that evolution is a “device designed to tend toward [italics added] a specific needle in a haystack—the single best potential result.” Further, if you think about it, with sufficient iterations, I will actually find the single best answer, and will know that I have it, because I have exhausted all possibilities. In the end, while (as I say in the article) it will move in “fits and starts”, it will always (as I say in the article) converge on this one best answer.
You almost certainly never get to a real optimum because in the contexts we’re talking about even Jim’s 2^100^ is podunk: Dawkins’ “damn near infinite” is actually more useful to think about it.
In terms of doing science (as I say in the article), it is appropriate to proceed as if the space of possible genomes is infinite, since in practice it is incalculable. This is different when doing philosophy, since it is in principle knowable.
And most importantly, you don’t necessarily get to your answer in a way you want — this is where Jim’s thinking about genetic algorithms leads him astray.
If you do directed evolution with organisms in a lab, well, evolution’s a bitch. You try to set up the problem as clearly as you can — say, the organism needs to improve how well it catalyzes a particular reaction essential to its growth, by evolving a protein that catalyzes that reaction poorly into one that does it better: this is the nut of most industrial and academic directed evolution. Now, in the case of a genetic algorithm, you can reduce and simplify screening your answers so they do what you want. But in the cell, man, it comes up with every goofy way of solving the problem you hadn’t anticipated: overexpressing the poor protein via freakishly improbable mutations, downregulating something else, or coming up with a chemical solution which works in the very very specific lab selection but not really outside it in the general way you were seeking — the upshot being it’s really, really hard to determine what the evolutionary “challenge” really is and to think about what responses to it can be. Genetic algorithms are a terrible way to think about real live genetic evolution.
I have enormous respect for the practical complexities of manipulating real live organisms. I think what you are describing here are our (current) technological limitations. Once again, this is a limitation of the observer, rather than the phenomenon.
And it’s worse than that, really, for talking about a “fitter” genome as a goal, because frankly evolution changes environment, or changes the parts of the environment which are relevant to the selection: your survival in a room isn’t determined by the same factors that determine whether a thrush or an ant or a bonobo can survive in it. This “fitter” genome concept is just goofy. Evolution doesn’t have “goals” that way.
I think that I have addressed this point in the article and in the prior comments. In effect, it’s an extended variation of the point that “it’s too complicated to know”, and therefore is subject to the same observation that “too complicated for us to understand” doesn’t equal “not true”.
And when biologists say there’s no goal, they mean something profounder than that anyway — odd that Jim goes off to the philopsophers in the preceding point but glosses over it here. They’re saying, there’s no telos: you’re not trying to evolve towards something. It sort of goes back to what I was saying above: evolution doesn’t really give a rip how it solves a challenge. It doesn’t even care if instead it solves some other challenge or changes the rules of survival for the organism. The organism is responding immediately to immediate forces: it’s analogous to the idea of a virtual photon, say: that there aren’t long-range forces in physics either. And that’s where the “scientific atheists” are biting (I have issues with it, and reasons I think they’re wrong to make theological conclusions from that: but they’re just right about the “no goal” thing and Jim’s response on it doesn’t wash.) There’s no real plan.
And it gets worse than that too because remember Darwinian style evolution isn’t exactly, well, right. You can show for example that bacteria undergo something like Lamarckian evolution, reaching harder for solutions in certain parts of their genomes nutrient starved or under phage pressure or whatever. That’s casting about like mad without all this nice steady Darwin stuff, no?
I know you’re aware that you’re speaking very loosely when you say, anthropomorphically, that “evolution doesn’t give a rip”, but this is a crucial distinction. Evolution via natural selection is a process governed by physical laws. At an algorithmic level it is a (brilliant) way to search a finite space of potential genomes. At any moment in time and space, each potential genome has a higher or lower fitness, and therefore probability of survival and reproduction, and therefore in that environment any method of search would statistically converge on the most fit solution. The changes in environemnt that occur across space and time are, in turn, governed by normal physical laws. Therefore, with complete knowledge of these physical laws and initial conditions, the convergence point is knowable. Now, this computation would be so ridiculous that no real human (for the foreseeable future) could ever execute it, so this belongs in the realm of philosophy rather than science. Further, any indeterminacy inherent in these physical laws would propagate through the evolutionary algorithm, but evolution itself would add no incremental indeterminacy.
And that third point is beyond silly. For one thing it’s aphysical. In that article Jim writes in fact, even the “random” elements of evolution that influence the path it takes toward its goal—for example, mutation and crossover—are really pseudo-random. For example, if a specific mutation is caused by radiation hitting a nucleotide, both the radiation and its effect on the nucleotide are governed by normal physical laws. Umm, yep — and those laws say they’re random. That’s exactly a quantum thing.
I think you’re missing the point here (probably because my exposition in the article was insufficient). Suppose we had no theory of evolution of any kind, but physics stood in its current state. One would draw whatever, if any, religious or philosophical conclusions he believed relevant from, say, quantum mechanics. Now imagine that we subsequently constructed the whole Modern Synthesis. Would this provide any incremental indeterminacy? No, as per the argument in the article.
For another thing, the chemistry of the genomes itself changes — this goes back to the really nonseparable nature of genes and environment I tried to hit on above. And that chemistry can change in really, really surprising ways, and again there is really a fundamental randomness that will get amplified in this process. Sorry, Jim (and Dirac), there’s not an equation you can write and from then on everything is billiard balls: at least, that’s not how science views the world right now. That’s just wrong.
This is, of course, an extension of the argument in your prior paragraph that is directly analogous to the extension of the “the search space is almost infinite” to the “the search is yet vaster when you consider that it is not static”. I think the same basic idea as in my prior paragraph applies: there is no incremental contribution of indeterminacy created by evolution via natural selection.
I’m sorry, I think there’s too much to say but you both are clearly savvy enough to grab where I am. And I think what Jim’s saying here is (1) unscientific and (2) built on sand.
Thanks again for the very detailed thoughts.
— Jim Manzi · Jan 10, 05:20 PM · #
Hi, Jim. I think you’re still missing the same crucial thing. Let me try to be clearer.
Firstly, you say that fitness is well-defined in the example you gave and in a genetic algorithm. Fine, but irrelevant. The example, or the analogy to genetic algorithms, is only useful inasmuch as it helps you talk about biological evolution: that’s where the controversy is. And that’s where the problem is perpetually poorly defined.
I talked a little about lab research and in the many, many, many directed evolution studies now in the literature this is a perpetual problem. The way you get around it — the way you see really prominent in academic research — is “screening not selection;” you dont apply vast genetic selection but do a lot (like tens of thousands) of individual precise measurements for a physical property like Michaelis constant or turnover number. But that’s not what evolution does. It’s not a problem with our lab’s experiments — it’s a problem with all of them.
And so again, the example is irrelevant. When you talk about the “goal” of real, biological evolution — the evolution that’s relevant to the theological discussion — you have something poorly defined. What’s the problem an individual finch is trying to solve when it mutates?
I think the complexity of the statistics also is getting lost. You write, In the end, while (as I say in the article) it will move in “fits and starts”, it will always (as I say in the article) converge on this one best answer. Well, no, it doesn’t. Demonstrably. You can evolve the crap out of some enzymes and show that they end up in local minima nowhere near what you could be getting. That’s experiment, not theory. Evolution hits dead ends where in theory it could back out but in practice it doesn’t; it’s in a local minimum and doesn’t have enough “temperature” (to return to the annealing analogy) to break out — does it in real life outside the lab too, so we think. Thus, maybe, no more sauropods (sob).
Your thinking about the system complexity doesn’t really work and in fact I’m pretty sure most scientists would say it’s just wrong. I think what you’re saying boils down to, (1) it’s reducible to physical laws (which a certain amount of quantum randomness) which make those “best solutions” calculable, ( (2) and we know those best solutions are achieved), and sure I’m with you that nobody could conceivably solve that equation nor pin down all its initial conditions, but (3) its writeable and a mighty enough supercomputer and/or God can solve them. Each of those ideas is flatly wrong.
I think I’ve already taken on (1); it’s just not clear to me (or you) what it is you’re reducing to physical laws. What’s the well-defined problem a snail genome is trying to solve?
Let’s take point (3) now. The problem here is, some systems (and evolution is certainly one) just ain’t linear. So very small changes in inputs — and that tiny randomness quantum mechanics imposes — have huge consequences for outputs. Small uncertainties get amplified to the point where they make consequences mathematically unpredictable. That’s not such a radical notion; it’s known and accepted for a lot of problems, and I don’t think there’s much quibbling that evolution is such a problem.
Consider if you will bacterial evolution. One little toy of cybernetics people for many years now has been pattern formation by motile bacteria — under certain circumstances, bacteria will aggregate and form wonderful sunflower patterns or the like on semisolid media. Each bug is statistically biasing its motions based on chemoattractants each bug (including itself) secretes, responding to local forces, and the result has order seen from above. It’s a nice nonlinear problem because, move a single bug out of millions and the pattern you predict — where each bug ends up — changes. So the environment each cell sees is that sensitive to positions which are themselves statistically determined on binding events where quantum chemistry gets to play, and you multiply by millions upon millions. That’s not something that math lets you predict unless you have zero uncertainty in every position and energy and science says you can’t. This idea that you’re not adding uncertainty is irrelevant because the system complexity means the uncertainty, however small, ultimately detemines the problem.
And point (2) is simply wrong. It’s wrong experimentally. It’s wrong factually: what’s a best solution to an ill-defined problem? And worst it’s wrong mathematically. This is where you’re missing the “virtually infinite” problem.
Look, when you do directed evolution of a single protein in a lab to solve a problem you’ve for which tried to restrict acceptable types of solutions as tightly as possible, there’s just ridiculous numbers of possible solutions/mutations and so it’s still a mega-bitch to see you’re sampling those solutions with any kind of decent coverage. When we’re talking about which of the mutant albatross chicks bears some mutation that’s really beneficial to the species of albatrosses as a whole, and maybe this one does but it just turned out that that was the wrong rock to build the nest on or a hurricane hit this year or what-have-you — that a particularly vastly helpful mutation to albatrosskind has a good chance of never being able to be demonstrated anyway and that presumably the vast majority of beneficial mutations never “get a chance to play” before the whole problem changes anyway and an albatross is no longer even an albatross but really a very interesting kind of squid: well, your statistics become bogus. You have no means to demonstrate that real world evolution is even sampling a local space well, let alone the space of all posiible solutions, to the problem which was never defined well to begin with!
— Sanjay · Jan 10, 07:20 PM · #
Sanjay:
I think I would (crudely) summarize your comment as “Look, Jim, sorry to say this, but the facts puncture your speculation”. (By the way, I’m extremely sympathetic with this line of argument, since it’s normally the side of the street that I work).
I’ve said over and over that none of what I’m saying is relevant for any practical scientific enterprise. The genome space through which search operates is so vast that as scientists we should see the effects that we would if no goal existed. Therefore, I accept all of the experimental results that you describe, but don’t think that they falsify my point.
Thanks again for devoting so much time and obvious expertise to this discussion.
— Jim Manzi · Jan 10, 08:48 PM · #