IBRI Colloquium, 22 Jan 1991

Dr. Robert C. Newman

Biblical Theological Seminary


                                   COMPUTER SIMULATIONS OF EVOLUTION




            Not a literature search

            Not covering origin of life question

                        tho 2 programs on diskette are self-reproducing automata

                                    REPRO - Langton's automaton in my JASA Spr 88 paper

                                    BYL - Byl's in his JASA Spr 89 paper

                        both these can be "mutated" by manipulating data for             structure & transition rules

            Not dealing with competition & spread of varieties

                        good deal has been done on ecology/population genetics

            Rather, a description & investigation of three programs that relate to the mechanism of evolution:

                        -- two described by Richard Dawkins in his Blind Watchmaker (1986), 46-75

                        -- one devised by self

                        these 3 programs also on diskette available from IBRI for $5





                        program, slightly simplified from Dawkins, for building "organisms" from genetic information, selecting among mutants

                        gene is sequence of eight small integers

                        generates "tree" controlling branch length, angles,

                                    # of levels of branching, with mirror symmetry

                        given original gene, program constructs all "one-step" mutations, displays on screen

                        operator selects which mutant to succeed parent


            Lessons from BIOMORPH:

                        shows how:

                                    mutation operates on DNA

                                    selection operates on developed form, not on DNA

                        see that:

                                    identical forms can conceal diff genetics

                                    leaving room for neutral mutation


Program SHAKES



                        Dawkins seeking to circumvent "monkeys typing Shakespeare" problem of enormous times involved

                        choose target sentence/phrase

                        start with gibberish of same length

                        mutate gibberish, selecting mutant/parent which is closer to target to be new parent

                        gibberish converges to target much faster than if monkeys were typing randomly

                        Dawkins gets convergence in typically 40-70 generations


            Dawkins' version:

                        Not described in detail, so can't tell how he generated mutants, how many mutations per generation

            My version:

                        One mutant each generation, compared w/ parent

                        Better of mutant/parent survives

                        I get much slower convergence, taking over 1000 generations


            Lessons from SHAKES:

                        shows that a "rachet mechanism" does work

                                    important reason why many convinced evolution must

                                                be correct

                        but this is "guided evolution,"

                                    which is considerably more efficient than even artificial selection,

                                    to say nothing of natural selection!

                        does not solve time question

                                    which version is more realistic?

                                                mutation rate in eukaryotes is 10-8 per             replication

                                    both ignore time involved for mutant to take over


                        my version suggests a problem

                                    for mutating into complex or optimal structures:

                                    last pieces of puzzle are highly constrained


Program MUNSEL



                        simulate mutation & natural selection by analogy with human language

                        letter string is both gene and organism

                        mutation is random change in content and/or length

                        selection is "naturalized" by requiring that each

                                    grouping in string be an English word

                        current version has operator do selecting,

                                    but comparing with a spell-checker would be more objective

                        generates words of 1-4 letters rather easily

                        relative frequency of space character (and nature of selection) tends to keep words short

                        little success in getting intelligibility in 100s of steps


            Lessons from MUNSEL:

                        complex organisms involve hierarchies of structure

                                    somewhat like intelligible writing

                                    letters > words > phrases/sentences > paragraphs

                        mutation only works at lowest level

                                    nucleotides <=> letters

                                    so becomes tougher to get anything acceptable as we move up hierarchy

                        non-selected mutation => gibberish

                        neutral mutations spread only by random walk

                        functional isolation seen here (as in terrain analogy)

                                    many combinations cannot be reached by single mutations from acceptable smaller groups

                                    what is the relative size of islands of intelligibility vs oceans of gibberish

                                                for each level of hierarchy?

                                    can you really get there from here?

                                                complex organs/organisms

                                                crossing higher levels of bio classification