20
egy.” As hoped, the benchmark strategy proved to do a respectable job; in
21
a large number of trials, it typically reached about 69 percent of a per-
22
fect score.
23
Alternatively, we can be inspired by nature’s method, and evolve a strat-
24
egy using directed evolution. A specific strategy for Robby is like a specific
25
list of nucleotides in a DNA helix, a discrete information- carrying string.
26
We can artificially evolve it by starting with some number of randomly
27
chosen strategies, letting them run for a while, and picking out the ones
28
that do the best. Then we make several copies of each survivor, “mutating”
29
each copy by randomly altering a few of the specific actions each strategy
30
specifies for a particular state, and even mimicking sexual reproduction by
31
cutting strategies and pasting them together with other ones. The process
32
is reminiscent of evolution. Can it find strategies for Robby that are better
33
than the “pretty good” designed one?
34
Yes, it can. Evolution easily found much better solutions than design.
S35
After only 250 generations, the computer was doing as well as the
N36
2 85
Big Picture - UK final proofs.indd 285
20/07/2016 10:02:50
T H E B IG PIC T U R E
01
benchmark strategy, and after 1,000 generations, it had reached almost 97
02
percent of a perfect score.
03
After a genetic algorithm has evolved, we can go back and watch what
04
it does, trying to figure out what made it so effective. This tricky bit of
05
reverse- engineering is increasingly a real- world challenge. Many useful
06
computer programs operate according to genetically constructed algo-
07
rithms that no human programmer actually understands, which is a scary
08
thought. Fortunately, Robby’s choices are sufficiently constrained that we
09
can try to figure out what is going on.
10
Robby’s best strategies improve on the benchmark in a number of clever
11
ways. Consider a situation where Robby is on a square containing a can, and
12
the squares to the east and west also contain cans. The benchmark strategy,
13
quite naturally, instructs him to pick up the can. But think about what will
14
happen next: Robby will move either east or west, thereby losing track of the
15
can in the other direction. The genetic algorithm, though it was constructed
16
using nothing but random variations and selection, “figured this out,” and
17
came up with a better strategy. When Robby is in the middle of a sequence of
18
three cans, he doesn’t pick up the one on his square; he moves east or west
19
until he’s reached the edge of the can grouping, and only then does he pick up
20
a can. Next, quite naturally, he moves back into the grouping, scooping up
21
cans along the way. This and other bits of clever engineering turn out to be
22
enormously more effective than the “obvious” designed benchmark strategy.
23
Evolution isn’t always better than design. An omniscient designer could
24
find the best strategy every time. The point is that natural selection, or di-
25
rected evolution in this case, is a really good search strategy. It doesn’t nec-
26
essarily find the best solution, but it regularly finds impressively clever ones.
27
•
28
29
As wonderful as evolution is at searching for peaks in a complex, high-
30
dimensional fitness landscape, there are places that it won’t find. Consider
31
a landscape with a very high mountain, separated by a long, flat plain from
32
a collection of undulating hills. And imagine a population whose genomes
33
are located within those hills. The process of small variation and natural
34
selection will let the species explore around the hills, looking for the highest
35S
point it can find. But as long as the variations in the genome within the
36N
population remain small, all of the individuals will remain in the grouping
2 86
Big Picture - UK final proofs.indd 286
20/07/2016 10:02:51
S E A RC h I n g t h R O ug h t h E l A n d S C A PE
of hills. None will have any reason to make a long, unrewarding trek across
01
the flat plain to get to the isolated peak. Evolution can’t see globally across
02
the space of genomes and find a better one; it proceeds locally through ran-
03
dom variation and then an evaluation (through reproduction) of how well
04
that particular variation is doing at the moment.
05
06
07
08
09
10
11
12
13
14
15
16
17
A fitness landscape with an isolated peak that would be difficult for natural selection to find.
18
19
The failure to find an isolated solution to some problem within a long
20
list of possibilities isn’t unique to evolution. Almost every efficient search
21
strategy attempts to take advantage of structure within the list of
22
possibilities— such as the fact that nearby points on a fitness landscape have
23
similar values of fitness— rather than blindly scanning every option. It
24
could, however, enable an empirical challenge to natural selection as the
25
correct theory of the evolution of species. If someone could show that a
26
particular organism’s genome had high fitness within the landscape defined
27
by its environment, but could not be “found” by the strategy that evolution
28
employs, it would decrease our credence in Darwin’s theory.
29
Given any one particular genome, how do we know that it is an isolated
30
peak in the fitness landscape? Such peaks almost certainly exist, although
31
they might be less common than they first appear. When we draw a two-
32
dimensional landscape, isolated peaks are almost inevitable, but when the
33
underlying space has many more dimensions (like the 25,000 or so genes
34
in a human being), there can be a lot more paths to get from one peak to
S35
another.
N36
2 87
/>
Big Picture - UK final proofs.indd 287
20/07/2016 10:02:51
T H E B IG PIC T U R E
01
A possible criterion for genomes that wouldn’t be produced by evolution
02
was put forward by Michael Behe, a critic of natural selection and advocate
03
of intelligent design. In an attempt to show that certain organisms couldn’t
04
have arisen through conventional Darwinian evolution, Behe proposed the
05
notion of “irreducible complexity.” An irreducibly complex system, in Behe’s
06
definition, is one whose functioning involves a number of interacting parts,
07
with the property that every one of the parts is necessary for the system to
08
function. The idea is that certain systems are made of parts that are so inti-
09
mately interconnected that they can’t arise gradually; they must have come
10
together all at once. That’s not something we would expect from evolution.
11
The problem is that the property of irreducible complexity isn’t readily
12
measurable. To illustrate the concept, Behe mentions an ordinary mouse-
13
trap, with a spring mechanism and a release lever and so forth. Remove any
14
one of the parts, he argues, and the mousetrap is useless; it must have been
15
designed, rather than incrementally put together through small changes
16
that were individually beneficial.
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
Incremental evolution of a complex mousetrap, as designed by John McDonald. The trap starts 33
as a simple wire that can snap shut when disturbed. In a series of steps, it adds: a spring, some 34
bait, resting on its side, attached to a platform, a longer “hammer,” a tripwire, a staple to hold the 35S
tripwire, a shorter spring wire, an even shorter spring wire, a separate catch to hold the tripwire, 36N
separating the hammer from the spring, and finally a more elaborate catch to release the trap.
2 88
Big Picture - UK final proofs.indd 288
20/07/2016 10:02:51
S E A RC h I n g t h R O ug h t h E l A n d S C A PE
You can probably guess what happened next. At least two different peo-
01
ple (John McDonald and Alex Fidelibus) presented possible “evolutionary
02
paths” that mousetraps might have followed. They created a series of de-
03
signs, starting very simply and becoming gradually more complex, of work-
04
ing mousetraps. Each step worked a little better than the previous one,
05
despite differing by only a small change. And the final step was precisely a
06
modern mousetrap. Adding insult to injury, Joachim Dagg investigated the
07
way that actual mousetraps have changed over the years, showing that (de-
08
spite being designed) they evolved gradually rather than appearing all at
09
once. In Dagg’s words, “All prerequisites for evolution (variation, transmis-
10
sion, and selection) abound in mousetrap populations.”
11
12
•
13
Irreducible complexity reflects a deep concern that many people have about
14
evolution: the particular organisms we find in our biosphere are just too
15
designed- looking to possibly have arisen through “random chance plus se-
16
lection.”
17
A version of this conviction can be traced back to William Paley, of the
18
watchmaker analogy. Paley wrote before Darwin came on the scene, but he
19
put some effort into attempting to refute any future Darwin- like thinker
20
who would deny God’s central role in explaining the complexity of the
21
world. His favorite example was the eye. The word “eye” appears more than
22
two hundred times in Paley’s Natural Theology: or, Evidences of the Exis-
23
tence and Attributes of the Deity, Collected from the Appearances of Nature.
24
The many pieces that have to work together, the undeniable effectiveness of
25
the eye at its assigned task, the effort to which the body attempts to protect
26
and preserve its eyes— to Paley, these spoke strongly to the view that the eye
27
implied “the necessity of an intelligent Creator.”
28
Not only can eyes be explained through natural selection; they seem to
29
have evolved separately dozens of times over the history of life. It’s not dif-
30
ficult to trace out plausible paths for how eyes could develop. The absorp-
31
tion of photons is one of the most basic activities that living organisms do.
32
This ability can be concentrated in photosensitive patches, or “eyespots,”
33
that are found even in some single- celled organisms. Given that an organ-
34
ism can sense light, it can be advantageous to acquire sensitivity to the di-
S35
rection from which the light is emitting. A simple way to achieve this
N36
2 89
Big Picture - UK final proofs.indd 289
20/07/2016 10:02:51
T H E B IG PIC T U R E
01
ability is to locate the eyespot in a recessed cup, such as is seen in certain
02
flatworms. Deepening the cup to an almost spherical opening allows the
03
organism to employ a primitive kind of lens, similar to that in a pinhole
04
camera; this is what we find in some contemporary mollusks. Filling that
05
eyehole with a transparent fluid helps with both protection and focusing.
06
Many of the steps along the way won’t arise in single jumps; often, evolution
07
can borrow mechanisms from other functions in the organism that came
08
about for different reasons.
09
You get the idea— not only can eyes be developed in stages of increasing
10
complexity and fitness, but we actually see such development in real crea-
11
tures alive today. And the human eye, as wondrous as it is, has unambigu-
12
ous flaws that would be inexcusable for a talented designer but make
13
perfect sense in light of evolution. The nerve fibers that carry visual infor-
14
mation to the brain are, for no good reason, in front of our retinas rather
15
than behind them. The octopus eye is a better design, with the retina in
16
front and nerves in back, so that octopuses don’t have a blind spot
like hu-
17
mans do. Our anatomy reflects the accidents of our evolutionary history.
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35S
36N
290
Big Picture - UK final proofs.indd 290
20/07/2016 10:02:51
01
02
35
03
04
Emergent Purpose
05
06
07
08
09
10
11
12
t
13
ime for a multiple- choice quiz: Why do giraffes have such long
14
necks?
15
16
1. Over the generations, giraffes kept stretching upward to
17
reach leaves near the tops of trees. Gradually their necks got
18
longer and longer.
19
20
2. Long necks help you eat. Random mutations in their DNA
21
gave some giraffes longer necks than others. These individu-
22
als enjoyed a nutritional advantage over their compatriots,
23
because they could reach fresh leaves near the treetops. This
24
advantage was passed on to their descendants, and gradually
25
the giraffe population developed longer necks.
26
27
3. Long necks are sexy. Male giraffes compete for the affections
28
of females by swinging their heads at each other. Random
29
mutations in their DNA gave some giraffes longer necks
30
than others, which conferred a reproductive advantage. This
31
advantage was passed on to their descendants, and gradually
32
the giraffe population developed longer necks.
33
34
4. Given the laws of physics, and the initial state of the uni-
S35
verse, and our location in the cosmos, collections of atoms
N36
291
Big Picture - UK final proofs.indd 291
20/07/2016 10:02:51
T H E B IG PIC T U R E
01
in the shape of long- necked giraffes came into existence 14
02
billion years after the Big Bang.
03
04
The difference between options 1 and 2 is a common way of explaining
05
Darwin’s theory of natural selection. Option 1 is incorrect; changes that
06
individuals undergo during their lives, such as through exercise or learning
The Big Picture Page 49