In natural selection, any organism that can reproduce contributes at least part of its Instruction Book to some members of the next generation of that organism. How many of the next generation get a copy of its Book depends (loosely) on how fit the organism was relative to others in its generation. Or, as we said, "less survival of the less fit." When people breed plants or animals, they can (and usually do) choose not to let just any of the survivors reproduce, but only the very best. For example, a plant breeder trying to increase the yield of corn will choose seeds from only the plants that produce the most corn. Usually, natural selection doesn't choose that strongly.

Let's take a look at two examples of artificial selection, starting with the the easiest possible case and then looking at a more complex, but more typical case. Of course, during nearly all of the thousands of years people have used artificial selection to improve plants and animals, they've had no idea which kinds of selection might work, since Instructions and Instruction Books and their relation to traits weren't known.

It's easiest to use artificial selection to choose traits caused by a single allele that produces a worker that does its job worse than the worker made by another allele -- so we want the organism we're selecting for to do less of something, or not to do it at all. By doing this, farmers and plant breeders have selected against many traits we don't like, such as production of poisons, indigestible parts, and molecules that yield a bad taste. While all of these traits are doubtless produced by a large pathway or cascade, they can easily be selected against because slowing down the worker at the head of the cascade or pathway (by selecting for workers who aren't good at their jobs) slows down the whole process.

In this our first example, a gardener wants to eliminate thorns from a kind of plant he likes. (Plant breeders have recently produced thornless roses, for example.) Production of thorns is likely a huge cascade, but there is likely a single worker that regulates the the number of thorns a plant has. (Well, more likely, the worker senses something that tells it the distance to the last thorn, so it really measures thorns per inch of stem.) In this case, the alleles for more thorns dominate those for fewer, so the gardener will be selecting for recessive alleles for fewer thorns.

When the gardener starts selecting for fewer thorns, his garden contains plants with a variety of alleles for the thorn-count worker Instruction:

The plants and their thorn-count worker Instruction are color-coded red-to-green for fewer to more thorns, and the plants show the number of thorns they have. Since Instructions for more thorns dominate Instructions for fewer, the plants' thorniness trait is simply a result of the allele for more thorns.

The breeder selects the two plants that have the fewest thorns, pollinates one with the pollen of the other, and collects their seeds. He throws away the seeds of the other plants.

Let's see what results he gets:

None of the offspring inherit Instructions for more thorns than their parents, so the gardener has at least not lost any ground. But until mutations create new alleles, he can't do better than this. For simplicity, let's assume a few mutations did occur in this generation. (If we wanted to be very precise, we'd just repeat generations of 2-thorn plants over and over until we got some mutations.) We've shown two mutants of the thorn-count worker Instruction: one (on the left) increases the number of thorns slightly, the other (next to rightmost) eliminates thorns completely, due to a nonsense mutation. The breeder can leave the more-thorns mutant out of the next generation, since he can see that it has more thorns. However, there's no way he can select for the plant containing the no-thorns allele, because its phenotype is identical to the others (since the no-thorns allele is dominated by the 2-thorn allele). So, he doesn't let the more-thorns mutant reproduce, and lets all the others reproduce at random, since there aren't any differences he can see among them.

In the next generation, there still won't appear to be any improvement, since no plant can have two copies of the no-thorns allele. (A plant with only one copy of the no-thorns allele won't look any different from any of the others in this generation.) In the generation after that, it's likely some plants will have both no-thorns alleles due to recombination:

So now the gardener can breed thornless plants, since he may choose to keep only the offspring of the two plants without thorns. Barring further mutations that produce thorny plants (revertants), none of their descendants will have thorns, so the gardener succeeded at what he wanted: making thornless plants. Even if later mutations produce revertants, he can easily select against them. A group of organisms that always produces offspring with the same trait as their parents is said to breed true. They breed true because every member of the group has the same allele in both of its Instruction Books, so future offspring always get that allele. As you might guess, making plants and animals breed true for useful traits is very important to farmers, gardeners, and animal breeders.

After a while growing his new thornless plants, the gardener notices that they get eaten by bugs a lot more than the original thorny ones. He wonders why. Let's take a look at what happened. (Unless the gardener lived very recently, there'd be no way for him to figure this out.)

Like most plants, the gardener's initial thorny plants produced an insecticide to deter bugs from eating them. They did so by a short pathway consisting of two workers. The thorny plants had a variety of alleles for these workers, most of which were good -- natural selection has worked for a long time selecting for plants that could fight off bugs -- but a few bad alleles persisted:

This diagram is a bit more complex than the last few, since we want to show two traits and two sets of Instructions: those for the number of thorns, which the gardener was selecting for, and those for insect-resistance, which he wasn't worrying about at the time. The plants are now color-coded for how well they resist bugs, and their insecticide-maker Instructions are color-coded to match. The plants' thorn-count Instructions are also shown, still color-coded for how many thorns they make, and the plants show the number of thorns they have.

The gardener selected the two plants with the fewest thorns and bred them:

Although these plants still have some good alleles for their insecticide-maker Instructions, they don't have as many as the previous generation, and many display poorer insect resistance. This wasn't the gardener's intention, of course, but wasn't avoidable, since he bred only the least thorny plants he had.

Two generations later, the breeder produced the first two thornless plants. Their insect resistance is a mixed bag: one is still making reasonably good insecticide workers (next to left), but the other one isn't doing well at all (rightmost):

Since all the thornless plants descend from these two plants, let's take a look at their offspring:

Some of these plants still have fairly good insect resistance, but some don't. Although there are still a few good alleles for insecticide-makers around, many poor ones are present, as they haven't been selected against. This tendency for good alleles to be lost due to successive selections from a small population of living things is called inbreeding depression, and is a common risk in all artificial selection breeding programs.

Another point to note: during the period of selecting for thornless plants, no mutations occurred to create worse alleles of the insecticide-maker Instructions. It's certainly possible that they might have, which would have led to even worse alleles for this trait in the generation shown above. The effect seen thus far is simply due to lack of selection for one trait while strong selection was applied to another.

This is the generation in which the gardener noticed that his thornless plants were being eaten by bugs. The second example looks at his attempt to breed better insect resistance back into his plants.

The gardener chooses the plants that have the best insect resistance, and lets them breed. He has several to chose from, so we'll only show the offspring from one pair. This doesn't matter that much, since the most resistant ones are all but identical in their resistance alleles.

Selection has improved this generation a little compared to the previous one. Still, a number of bad alleles for insect resistance are present, and reducing their numbers will take time. A single good mutation for one of the insecticide-makers has shown up in the rightmost plant (better allele in the top Instruction Book, left hand page), but due to the other alleles in the plant, the effect isn't yet visible. Still, since this plant is among the most resistant in this generation, the breeder will let it pass its Instructions on to future generations, where the new, better allele can be selected for:

The gardener again selects the two most insect-resistant plants and breeds them:

At this point, unless there's another beneficial mutation, there's nothing more the gardener can do to improve the insect resistance of his plants. He's selected for the plants with the best traits he can find. Note that these plants don't breed true for insect resistance as they do for thornlessness. Some of their offspring are carriers for less-resistant recessive alleles. Eliminating these by selection is all but impossible (since all the plants look alike), so in every generation a few plants will be eaten by bugs. (And of course, if the gardener wasn't careful, he may have inadvertently let another trait get worse.)

Modern breeding programs (early 20th century and later) are much more sophisticated than the examples we gave here. Usually, rather than working on small numbers of organisms, they attempt to improve large populations of plants and animals at once. This allows modern breeders to rapidly improve traits, while preventing useful traits from being lost. On the other hand, the problems these programs are trying to solve are usually harder than either of the ones shown in the examples: they often try to select for traits produced by many Instructions and try to produce traits that previous generations of breeding couldn't produce, like animal feeds that reduce manure production, or plants that are very tolerant of salty fields. Nonetheless, modern breeding can't get around the basic problems of artificial selection.

While some improvement in traits can be achieved simply by selecting for desired traits in an existing population, we sometimes have to wait for mutations to create traits we want. Since mutation is random, we may have to breed plants and animals for a long time before we see a trait we want. Keep in mind that we shortened the examples by skipping generations where nothing happened, and assumed a rate of mutation much higher than normal. In real life, germline mutations producing visible new traits -- both good and bad and both desirable and undesirable -- occur in about one in a million individuals. Even where good alleles are available for a trait we want, distributing them in a population without increasing the proportion of other bad alleles or lowering overall genetic diversity requires many generations.

Selecting for several new traits at once is often hard because mutations producing new traits are very rare. It's sometimes even difficult to select for a single new trait while preserving existing desirable traits, as sometimes the new beneficial trait occurs in an organism that has less-desirable alleles. Modern plant and animal breeders sometimes try to increase mutation rates by treating organisms with mutagens; this helps some, but doesn't entirely solve the problem.

Sometimes, artificial selection produces bad traits, as breeders deliberately tolerate losing one good trait in order to choose another, with the hope of recovering the first trait later. The rarity of new traits encourages this. Occasionally, bad traits slip in undetected, either because they were masked by dominant alleles, or because nobody thought to look for them -- the latter is especially true with traits we now want that were lost to historical breeding.

Over the last several thousand years, people have used artificial selection to produce plants and animals with many traits they wanted, like high-yielding grains and docile domestic animals. However, during this time, they lost some desirable traits we now want. Pet lovers know two of the best examples: Persian cats tend to be deaf and Golden Retrievers often have hip problems; both of these traits were produced accidentally via artificial selection. It's also possible that people allowed bad traits to slip into crop plants: some species of crop plants are less insect and disease resistant than their wild relatives. It's not clear that the people selecting for traits throughout history could have done much about this. Since they had no knowledge of chemistry or biology, there's no way they could have known that plants actively fight off insects and pests. It's also possible that the plants they started with simply lacked the desirable traits.

There's one subtle drawback to artificial selection: it's not possible to select for or against things that aren't visible. Since artificial selection relies on random mutation to produce new visible traits, it also suffers from mutations producing new invisible traits, like new allergens. Even today, new allergens and poisons occasionally appear in food as a result of artificial selection.

One effect of artificial selection has been to exaggerate the differences between Instruction Books within a species. Artificial selection has produced several distinct Instruction Books for each species it's modified; depending on the type of organism, these kinds of Books are called lines, breeds, or varieties. (The terms mean essentially the same thing, but "breed" is usually used for animals and "variety" is usually used for plants.) For example, there are several breeds of dogs: German Shepherds, Golden Retrievers, poodles, and many more. They were bred for different traits, like helping with the hunt or looking cute. Organisms within a variety have Instruction Books that are more similar to each other than they are to Books in organisms in different varieties of the same species. Of course, even within a variety, the organisms differ somewhat due to mutation, and even between varieties, organisms of the same species are more similar to each other than they are to organisms of different species. That is, two German Shepherds are more alike than a German Shepherd and a Golden Retriever, and a German Shepherd and a Golden Retriever are more alike than a German Shepherd and a fox.

Varieties' different traits breed true only if both parents come from the same variety. The if the parents are of different varieties, the offspring receive two slightly different Instruction Books, and often display traits different from either parent. These offspring are called hybrids and their traits are often "better" in some way than that of either parent. For example, many corn hybrids produce more corn and resist disease better than either parent. Unfortunately, hybrids themselves don't breed true, because their Instruction Books have differing alleles for many Instructions.

Despite the limitations of artificial selection, until recently it's all we had. All crops and domestic animals are the result of mass breeding over thousands of years. (Most garden plants have been artificially selected for prettier flowers, too). Even "organic" crops result from the same thousands of years of artificial selection; wild plants simply don't produce enough food to support large human populations. (There are some exceptions: some "wild rice" harvested is in fact wild, and in many parts of the world, people hunt game animals or catch wild fish for food.)

We just said that artificial selection was all we had "until recently". In the last few decades, people have developed a new technique called genetic engineering. Genetic engineering addresses many of the limitations of mass breeding.