Figure 3
Blow up of the evolutionary activity graph in Figure 2, showing the neutral variants that cause the fitness increase just after time step 1000.
Figure 4
Comparison of the activity of real patents and shadow patents, showing the twenty most heavily cited patents and the twenty most heavily cited shadow patents. The activity waves of the real patents all rise above 100 while those of the shadow patents remain below 25. Note that the activity accrued by significant real patents can vastly exceed that accrued by any shadow patent.
Notes
1 Another interesting form of biological teleology has to do with the purported inherent “progress” exhibited by the course of evolution. For a discussion of how my objectification of value in biology bears on this, see Bedau 1998.
2 For more on my value-centered view of teleology, and for comparisons with other views of teleology, see Bedau (1990, 1991, 1992a,b, 1993).
3 Not just selection of the trait; see Sober (1984) for this distinction.
4 “Selection test” information would be more informative (and more awkward) terminology. Norman Packard and I developed and applied this method to a number of systems over a number of years with the help of students and colleagues. See Bedau and Packard 1992, Bedau 1995, Bedau 1996, Bedau and Brown 1997, Bedau, Snyder, Brown, and Packard 1997, Bedau, Snyder, and Packard 1998, Bedau, Joshi, and Lilly 1999, Rechtsteiner and Bedau 1999a,b.
5 Although the evolutionary activity method is novel, the essential logic behind it should be familiar. See, for example, Kumura 1983, Wimsatt 1987, Raup 1987, Beatty 1987. These parallels are traced in greater detail in Bedau 2002.
6 Created by C. Titus Brown, Evita is inspired by Tierra (Ray 1992) and its derivative Avida (Adami and Brown 1994), but it is much simpler because it disallows the kind of interactions that lead to parasitism and the other interesting evolutionary phenomena observed in Tierra. Its simplicity makes it an especially simple and clear illustration of how graphing evolutionary activity reveals a system’s evolutionary dynamics. A much more detailed presentation of the material in this section is available elsewhere (Bedau and Brown 1997).
7 If the mutation rate is too low, there is no significant genetic change in the population. If the mutation rate is too high, the population dies out almost immediately because no successfully reproducing creature can survive the bombardment of mutations long enough to reproduce.
8 Actually, it has an indefinite number of them, due to random sampling differences—a qualification I will usually ignore.
9 In the Evita activity graphs shown here, the Evita system parameters were all identical except for mutation rate and elapsed time. Each genotype in a given run is given a unique name of the form Nxxx, where N is a number indicating the genotype's length and xxx is a three-character string (in effect, a base 52 number) indicating the genotype's order of origination among genotypes of that length. For example, 32aac is the third length 32 genotype to arise in the course of a given run.
The grid size was 40 x 40, so when the grid filled up the population consisted of about 1600 self-reproducing programs. I have pruned out irrelevant data about transitory genotypes by graphing only those genotypes that had at least five instances in the population at some time. This removes some of the “little hairs” created by nonadaptive noise (see Figure 3).
10 Note that the activity scale (y-axis) in these two plots is roughly comparable, except that activity on the bottom is expanded by a factor of three to make the neutral model activity easier to see.
11 Notice that the fourth salient wave (due to genotype 32abl) does not correspond to a significant fitness jump. This genotype is well adapted, but it is not significantly better adapted than its main predecessor: genotype 32aaV. The waves from 32aaV and 32abl coexist for so long because the two genotpyes are nearly neutral variants. In fact, the fitness of the second wave (32abl) exceeds that of the first wave (32aaV) by about only 0.5%. The interactions among the three salient waves between updates 4000 and 5000 have a similar explanation. They are a significant improvement (5% fitness advantage) over the genotypes that they drive extinct, but they differ from one another by much less (less than 2%).
12 My use of the term “adaptationism” captures what I believe is the central issue, but I should emphasize that the term is sometimes used in other ways. To get a sense of the similarities and differences, see Maynard Smith 1978; Dawkins 1982; Dupré 1987; Sober 1987; Brandon 1990; Burien 1992; West-Eberhard 1992; Orzack and Sober 1994, 2001; Godfrey-Smith 2001.
13 It might be uncontroversial that some specific traits are adaptations, but those are the exception.
14 One can begin to appreciate just how canonical this response is by examining some other well-known responses to Gould and Lewontin (e.g., Mayr 1983, Dennett 1983, Rosenberg 1985, Sober 1987, Dennett 1995, Sober 1993). I spell out these parallels elsewhere (Bedau 2002).
15 For example, inspection of Figure 2 shows that the big wave in the middle produced by genotype 32aaV is an adaptation but it does not show what makes it better than its peers. We can usually discover a genotype’s adaptive advantage by independent microanalysis, as we did for 32aaV.
Genetic hitchhikers and genetic drift introduce some complications in this analysis. I discuss these details in a forthcoming monograph.
16 Mathematical details about the statistics can be found elsewhere (Bedau and Packard 1992; Bedau 1995; Bedau, Snyder, Brown and Packard 1997; Bedau, Snyder and Packard 1998).
17 The material in this section is explained in greater detail in Skusa and Bedau 2002. There is plenty of previous work on cultural evolution and on patents, but none quite like ours. For many years cultural change has been treated as a process of the diffusion of ideas (Rogers 1995), and the scientometrics community has been investigating scientific and technological change by analysis of bibliometric data and patent records for decades (Pavitt 1985; Garfield, and Welljams-Dorof 1992; Narin 1994; Albert 1998). But these approaches think of “evolution” simply as any change in time rather than just change resulting from differential imperfect replication and selection. Sociobiology (Wilson 1978, Lumsden and Wilson 1992) and its contemporary sibling evolutionary psychology (Barkow, Cosmides and Tooby 1992) explore one kind of connection between biological and cultural evolution, specifically, the extent to which certain psychological and cultural phenomena (e.g., homosexuality and altruism) can be explained by appeal to the operation of biological evolution itself. This reduction of social science to biology is contrasted with the approach to culture illustrated by memetics (Lynch 1996, Blackmore 1999, Aunger 2000), which considers the evolution of cultural phenomena in its own right, independent from and even competing with biological evolution. The two classic quantitative treatments of cultural evolution (Cavalli-Sforza and Feldman1981; Boyd and Richerson1985) tend toward different answers to the question whether cultural evolution is ultimately explainable in terms of biological evolution, with Cavalli-Sforza and Feldman leaning toward explanatory dependence and Boyd and Richerson leaning toward a limited autonomy for culture. My approach is neutral on this issue. I study cultural evolution as an evolutionary process in its own right, ignoring whether and how it might depend on biological evolution. My goal is to provide an empirical and quantitative picture of the evolution of culture, one which allows us to compare its evolutionary dynamics with those of biological evolution. Both reductionists and antireductionists could profit from objective empirical measurement of cultural dynamics.
18 Special thanks to my collaborators on the evolutionary activity investigations reported here: Titus Brown, Norman Packard, and Andre Skusa. For helpful comments or discussion, thanks to Phil Anderson, Peter Godfrey-Smith, Mike Raven, Tom Ryckman, and Chris Stephens. thanks also to audiences at the University of Oklahoma, the University of Washington, Washington University, the Center for Humanities at Oregon State University, the Center for Cognitive Studies at Tufts University, the Santa Fe Institute, the first Genetic and Evolutionary Computation Conference, the fourth European Conference on Artificial Life, Artificial Life VI, the Lake Arrowhead Conference on Computational Social Science, and at the Fraunhofer Gesellshaft in Sankt Augustin, Germany, where some of the material discussed in the paper has been presented. Thanks also to the University of Oklahoma, its Zoology Department, and Professor Tom Ray, for hospitality and support while some of this work was accomplished.
| 