interesting things dennett says

interesting things dennett says in "darwin's dangerous idea"

the "baldwin effect" uses reinforcement learning to construct a version of Lamarckian evolution within Darwinian evolution. The idea (explained further by Robert Richards, "darwin and the emergence of evolutionary theories of mind and behavior") is that natural selection has a hard time finding a "telephone pole in the desert" - an isolated peak in the fitness landscape, a "special trick". but if creatures have the capacity to discover that peak over the course of their lives through a random walk in design space, then being born closer to the peak is better than being born further away, since you're more likely to discover the "trick" and discover it sooner. Thus the ability to learn smooths the fitness landscape, even if learned knowledge is not directly passed on.

how expensive would it be to build a machine to unscramble an egg (reverse entropy)? such a machine would be a box into which we could pour scrambled eggs, and a whole egg would come out of the other side. it turns out this is trivial - just put a live hen in the box!

distinction between good and greedy reductionism. good reductionists think we can explain everything without skyhooks. greedy reductionists think we can explain everything without cranes.

different levels of possibility - logical, physical, biological, historical, actual. It is logically impossible to fly faster than the speed of light without moving. It is physically impossible to fly faster than the speed of light. It might be biologically impossible for there to be a flying horse, even if it is physically possible (biological impossibility is sort of slippery because it seems to basically be asking whether something is in a reachable-by-evolutionary-processes portion of design space, but that depends on how you choose your starting point, and how you define design space [dennett uses the library of mendel but acknowledges that this ignores the issue of how DNA is read]. and you run into complications like, what if a flying horse needed to be carnivorous to work? is that possible, and if so, would it still be a horse?). Historical possibility confuses me - Dennett seems to claim President Goldwater is historically possible, while President Clinton is actual, but that doesn't answer the question of how something could be historically impossible while still being biologically possible.

The idea of collapsing all of these levels into one is called actualism - the "dread hypothesis" that only the actual is possible. Dennett doesn't like it (see his book "elbow room" for more). (from looking at the stanford encyclopedia it seems as though this is something of an oversimplification).

not new, but a useful device - using Borges' Library of Babel as a device to explain multidimensional spaces. Suppose you wanted to alphabetize the books of the Library - you could organize their first letters by floor (so, 100 different floors in Dennett's formulation). Then you could have 100 corridors per floor each containing all of the books with a particular second letter, and 100 shelves per corridor, to organize the books by their third letter. You'd need hyperastronomically long shelves, then, but in a three-dimensional world there's no other way, you can't line things up at right angles past the first three letters. But imagine that you could, and therefore that you can store the entire library in a million-dimensional space, each dimension of which is only 100 units long.

in the case of the Library of Mendel / Design Space, Richard Dawkins calls this "genetic hyperspace".

general idea of "Good Tricks" and "forced moves" in Design Space.

dawkin's idea that "there are many more ways to be dead than alive" is one way of saying that a very, very small portion of Design Space is, ever has been, or ever will be instantiated in the physical world. Quine sort of gets at the same thing when wondering how many "possible men" were in his doorway, and whether any of them were the same man.

is it possible for evolution to have done all that work in the time available? we know the mutation rate (the rate of "genetic drift"), so we can say unequivocally that there has been far more than enough time for the necessary changes to have happened. of course, the actual rate of evolution is much, much slower than this theoretical "speed limit".

minsky's "sparseness principle" - two simple processes which produce similar results are likely to be identical. explored in "why intelligent aliens will be intelligible"

Dennett demolishes the argument for "faith" as a special category of belief. Not in any new way, but it's still pretty spectacular. He quotes Ronald de Sousa as describing philosophical theology as "intellectual tennis without a net", the net being the constraints of rational argument. If you want to play with the net down, then anything goes - I can respond by saying "your argument implies that God is a ham sandwich" and you can't call me on it, because we've abandoned rational principles. You can't just take the net down for your serves, but not my returns. Suppose that you are in a foreign land and a loved one is brutally murdered in front of your eyes; at the trial the accused's friends each bear witness to how much faith they have in his innocence, and the judge believes their faith over all evidence the contrary. Would you want to live in that land? "we're seriously trying to get at the truth here, and if you think this common but unspoken understanding about faith is anything better than socially useful obfuscation to avoid mutual embarrassment and loss of face, you have either seen much more deeply into this issue than any philosopher ever has (for none has ever come up with a good defense of this) or you are kidding yourself."

another theme of dennett's: whatever the merits of any particular theory of how life evolved (they could all be false!), it's inescapable that a theory of the origin of life must have the logical shape of darwinism, in that it must explain how complexity can develop from noncomplexity. any theory which doesn't do this, is a priori unable to explain complexity (the "if God created man, who created God"? circle). Therefore, whatever the difficulties in our current understanding of evolution, we know they can be patched somehow without disrupting the overall logical shape of the theory, because a true explanation of life must have that logical shape.

one thing I realized: dennett describes the game of Life. Individual cells flickering on and off are often described as "births" and "deaths", in that the fundamental physics of the world are supposed to mimic some abstraction of biology where things grow if they have just the right number of neighbors but die if they're too lonely or too crowded. But there is another level of "life" within the game, namely Conway's self-replicators (as well as smaller entities like gliders and eaters, etc., which do not replicate but are higher-level abstractions, maybe the "chemistry" of the Life world). These are "life" on top of life, living things composed of other living things. this is sort of analogous to a body being made out of cells (which are also alive), but maybe the game of Life is a better example for being more mathematically pure.

Dennett mentions that the implementation of a universal Turing machine in the Life world shows that computation can be done in a two-dimensional world. Apparently he explores theoretical implications of this in "the intentional stance" (1987).

even if we think we need God to explain why all of the physical laws of the world are exactly the way they are, so perfectly coordinated to allow life, the Game of Life though-experiment changes our required conception of God from "God the Lawgiver" to "God the Lawfinder", just as John Conway had to search for the rules of the Life among all of the other possible universes.

Lee Smolin claims that universes reproduce through black holes, with some mutation. This selects for universes which have the necessary conditions for black holes, which might include a prevalence of carbon (which plays some role in the collapse of gas clouds). This is of course a somewhat out-there theory, but it eliminates the "what else but God could possibly explain the universe's physical constants being the way they are?" line of argument, because there is at least one other possible explanation. Another possible response is the Hume/Wheeler idea that there is an infinite succession of universes, Big Bangs followed by Big Crunches, and every possible set of settings for the universal constants is "tried" an infinite number of times.

evolution is faster than hill-climbing because it proceeds as a sort of cloud (according to manfred eigen, and I think I'm oversimplifying). Benign mutations mean that there are lots of mostly-similar organisms occupying a region of design space, surrounding  local a "peak", but if they come near another peak it's likely that there will already be some organisms near that peak, and those will become more successful and thus re-center the process around that peak. this is one way that evolution can "look ahead".

there was no "first living" thing because the definition of life is so fuzzy. and "adam the protobacterium" (ancestor of all life on earth) differs in no interesting way from "badam", his relative who happened not to have founded anything of note.

in light of this, what happens when we try to define "function" or "purpose"? when did "purpose" appear in the world? did the first nucleotides have "purposes"? etc. however you legislate the answer, you will create a cutoff, and there will be things on both sides of the cutoff that differ in nonessential ways from each other.

a similar argument can sort of be run for "when was the first error?", but Dennett claims the first error was a copying error - this is objectively an error since it affects fitness, and before reproduction there can have been nothing that we could even hope to call an error. Dennett calls this first error "original sin".

Dennett identifies Arthur Samuel's 1955 learning checkers program as a possible "AI-Adam" and recommends reading his 1959 paper (reprinted in 1964 Computers and Thought).

the "coin-toss-champion" trick is characteristic of the sort of explanations that evolution offers. The champion may demand an explanation for why he just won ten coin tosses, but all you can give is an explanation of why someone must have won ten coin tosses. You can explain why there must exist creatures with certain features, but not why those creatures have those features (as opposed to the infinity of nonexistent creatures which could also have had those features). All of the selection events happen "offstage" - your ancestors gradually improve in fitness even though none of them ever failed to reproduce, because they were selected from a wider pool on exactly that basis.

dawkins "the evolution of evolvability", 1989. Dennett cites this in the context of talking about Stuart Kauffman's The Origins of Order: Self-Organization and Selection in Evolution, which apparently grounds evolution as a sort of special case of dynamic systems theory, or something like that (a bit unclear).

quoting ulam, the necessity of rhyme in poetry is sort of an automatic mechanism of originality, since it forces you to look for something you might not otherwise have said in order to fit the rhyme. The constraints of sonnet-writing are like the constraints of engineering anything else with interlocking parts - if you change one thing you may need to then revise everything else (to fit the new rhyme, or concept, etc.), and continue the cycle until you iterate to something satisfactory. Possibly the same is true in evolution; the constraints (of the laws of biological necessity? - analogy with cathedrals, and how many of their apparently artistic features are actually just necessary artifacts of the construction process) lead to greater creativity.

the "laws" of biology are really just deep regularities, but that's all we should ask for. the placement of the mouth at the bow rather than the stern is like the presence of key-ignition in current automobiles - there are tons of good reasons for it, it's a good solution which can be arrived at independently from many directions, etc., but it's not a law. When Kauffman and others talk about the "physics of biology", it's misleading for that reason - they're not discovering laws, just engineering principles.

john von neumann - "It is just as foolish to complain that people are selfish and treacherous as it is to complain that the magnetic field does not increase unless the electric field has a curl."

the fundamental insight of game theory is that it's not good enough to have statistical models of the other entities in your environment, if those entities are in fact guided by rationality, because those entities will in fact respond to your moves in ways that creates complications.

not an insight, but a reminder: the "survival machine" chapter (ch. 2) of the selfish gene is really fucking good ("now they swarm inside giant lumbering robots, sealed off from the outside world, communicating with it by tortuous indirect routes, manipulating it by remote control…")

dennett explains why humans are "special" by analogy with the eukaryotic revolution. prokaryotes (having developed the necessary apparatus) were invaded by symbiotic parasites to create eukaryotic cells, capable of vastly greater complexity. similarly, human beings (having developed the necessary apparatus, for other purposes) were invaded by symbiotic parasites known as "memes". A person is a human infested with memes.

the algorithmic model of evolution adopts substrate neutrality towards the difference between genes and memes.

a memetic slogan: "a scholar is just a library's way of making another library"

differences between memetic and genetic evolution: memetic evolution is directed, not random. memetic evolution is lineage-crossing, in that separate gene pools can recombine very easily (the species barrier is not quite as omnipresent; seemingly very different memes can easily reproduce with each other). memes have no fixed loci in the brain in the way that genes do on a DNA strand (two memes can't compete for "the same spot", but all memes compete for some sort-of-but-not-really-fixed (but certainly bounded) number of slots).

dennett dispatches Twin Earth beginning with the thought-experiment of the two-bitser, a device that accepts quarters but also happens to accept panamanian slug-balboas. It was originally manufactured to accept quarters (and accepting slug-balboas would be considered an error in that case), but suppose it were transported to Panama and began to be used to accept slug-balboas. Would we say that the "meaning" of its acceptance state had changed? A similar thought-experiment can be run with frog brains - suppose they are configured to recognize fast-moving dark splotches, which in the wild generally correspond to flies (so we want to say thier recognition state means something like "there's a fly!"). Now say we move some of them to a zoo, where there are no flies but the zookeepers regularly shoot fast-moving food pellets for the frogs to zap up with their tongues. Has the meaning of the brain-state corresponding to fast-dark-splotch-detection changed? What if, over time, there is selection pressure in the zoo towards frogs that are better at detecting the peculiarities of fast-moving pellets, in a way that makes them worse at detecting flies? Has the meaning changed then? If so, there's surely no single point in time at which we could place the change.

Dennett's version of Twin Earth is a place exactly similar to Earth but where horses are replaced by schmorses, identical creatures as far as we can tell that are actually a different species. If am unknowingly transported to Twin Earth and I say "that's a horse" upon seeing a schmorse, am I wrong? According to Dennett, my entering the horse-recognition state in response to schmorses is no different from the frog entering the fly-recognition state in response to pellets. The fly's natural kind was really "fly-or-pellet", and mine was really "horse-or-schmorse" - just the set of all things that active a horse-recognition state in my brain. Similarly to Siamese people who knew of no other cats than Siamese cats (and called them "kats", say), when they found about other cats they needed to decide whether their word "kat" meant just Siamese cats (which is what they always had used it for) or everything that we would call cats. If they didn't have a biological theory to provide some principled notion of cat-hood, how could there possibly be a fact of the matter? If a Siamese person said "lo, a kat!" when encountering a non-Siamese cat on the street, are they asserting an error or the simple truth? There is no fact of the matter. The answer has not been fixed, because the question has never arisen before. Our concept of "horse" is similarly indeterminate.

dennett makes the Godel argument fairly compelling before he destroys it. The argument is that no mechanism can decide the truth of statement of arithmetic, but humans can just "look" at the statements and, through understanding, decide which are true. The problem is that what Godel actually showed is that some statements have no proofs, and therefore no mechanism can find proofs for them. But neither can any human, because there are no proofs! All a human can do is look at statements and say "yes" or "no" according to some complicated internal heuristic, but we have no way of putting that into correspondence with anything like "truth" (and indeed, mathematicians are often wrong when they reason informally). We could train a parrot (or a robot) to look at theorems and say "yes" or "no" as well, according to some principle - what would be the difference? analogy - there is probably no feasible algorithm for playing perfect chess, but there are still computer programs better at it than any human. Similarly there is no possible algorithm for proving all true statements, but there may still be computer programs better at it than any human.

it must be true that there is an evolutionary explanation of how genes and memes interacted to "create the policies of human cooperation that we enjoy in civilization" (even if we haven't figured out the details yet), but there's no reason that this interaction would be to the benefit of the genes! genes do not "hold culture on a leash" - they are not intelligent designers of our minds, they could not predict the sorts of cultural interactions we engage in today, and so they are probably not strong enough to bend them to their benefit (over time, of course, genes which do will be selected… but they still compete with memes and other replicators in the culture wars).

dennett's ideas about a naturalized morality spend a lot of time focusing on satisficing/bounded rationality as a fundamental problem. utilitarianism doesn't work because it's intractable to compute (even given perfect information), and in general we have to settle for imperfect and unfair solutions (thought experiment: the search process that yields 250,000 candidates, forcing you to cull the pool arbitrarily without considering everyone. but we can't even debate about how to best cull the pool since there are an infinity of points to consider - at some point we have to cut off debate. but how do we decide when to do that? etc. - an infinite regress unless it's stopped hard somewhere, presumably on a sub-conscious level).

hofstadter's problem of "reverberant doubt" - if you give lots of a people a button which gives each of them $100 if they push it, but $1000 if no one pushes it, of course the optimal strategy is to refrain from pushing, but it's easy to doubt that everyone will play optimally…

this is the problem with hyperrationality, and one reason memes like religion are useful.

the basic point is, ethics is hard.

cultures and languages occupy positions in Design Space just like species do, and are irreplaceable in the same way, so it's just as sad to see them go extinct.