In light of the popularity of this piece, here are some things to keep in mind about 'selfish' genes:

1. The basic issue is about the unit of selection - does natural selection choose allele, individuals, populations, or species? The answer, like most things in biology, is yes, as Douglas Futuyma puts it in his standard textbook on evolution (p. 354. 3rd edition):

If, then, our concept of levels of selection includes causality, natural selection can act at the level of the gene (as in meiotic drive), organism, and at least in principle, population and species.

In light of the popularity of this piece, here are some things to keep in mind about 'selfish' genes:

1. The basic issue is about the unit of selection - does natural selection choose allele, individuals, populations, or species? The answer, like most things in biology, is yes, as Douglas Futuyma puts it in his standard textbook on evolution (p. 354. 3rd edition):

If, then, our concept of levels of selection includes causality, natural selection can act at the level of the gene (as in meiotic drive), organism, and at least in principle, population and species.

Futuyma lays out the general view of group (which can mean species or population) selection (p. 351-352):

No one denies that group selection, if sufficiently intense, might prevail...[in fact, it has been demonstrated to be possible by experiment] but many, perhaps most, evolutionary biologists believe that it is only rarely an important force in evolution.

2. When it comes down to genic selection vs. individual selection, the issue is more a matter of useful perspective. There is so much diversity in biology, and some situations are more usefully considered from a gene's eye perspective, while other's aren't. There are no absolute laws of selfishness here. There are clear examples of genic selection, that can't be explained by the advantage an allele confers on an individual. These are classic selfish elements, as Futuyma notes:

An increase in the frequency of a "selfish" genetic element, such as a t allele in the mourse or Medea in flour beetles, is evidently due to genic selection: selection of an allele because of its capacity for distorting segregation, irrespective of its effect on the organism that bears it. But if an allele increases in frequency because it enhances the organism's survival or fecundity, shall we call that genic selection or individual selection? Again, opinions differ.

Ford Doolittle and Carmen Sapienza, in a classic piece on selfish genes in the April 17, 1980 issue of Nature make the same point:

If there are ways in which mutation can increase the probability of survival within cells without effect on organismal phenotype, then sequences whose only 'function' is self-preservation will inevitably arise and be maintained by what we call 'non-phenotypic selection'. Furthermore, if it can be shown that a given gene (region of DNA) or class of genes (regions) has evolved a strategy which increases its probability of survival within cells, then no additional (phenotypic) explanation for its origin or continues existence is required.

Leslie Orgel and Francis Crick made the same point in the same issue of Nature, comparing selfish DNA to parasites:

In the case of selfish DNA, the sequence which spreads makes no contribution to the phenotype of the organism, except insofar as it is a slight burden to the cell that contains it... The spread of selfish DNA sequences within the genome can be compared to the spread of a not-too-harmful parasite within its host.

Experience has shown that these ideas are right: our genomes are filled with parasitic, virus-like pieces of DNA that, for the most part, make no phenotypic contribution. (Actually, their potentially negative phenotypic consequences are kept to a minimum by active surveillance and suppression.)

Bottom line: there are clear examples of 'selfish' genetic elements, with no positive fitness contribution, and which exist simply by virtue of the fact that they can perpetuate themselves.

3. Now what about 'regular genes' - the ones that contribute positively to our phenotype? Should we consider them 'selfish'? Selfish is actually a distracting term (as Dawkins himself has said). Again, the real issue, as both Futuyma and Dawkins point out, is at what level selection operates - allele or individual.

Obviously, the fitness effect of any given allele depends on its context - the other alleles around it in the genome. An allele in say, a myglobin gene, with the potential to make a faster-running cheetah, is useless if that allele is coupled with an allele that causes blindness. That's an extreme example, but the basic point is true: the effect of most alleles depends, more or less, on context.

From a gene's eye view however, it's not so necessary to consider context in order to make useful predictions about the evolutionary future of a gene: you can simply calculate the average fitness contribution of an allele, over all genetic backgrounds, and predict the evolutionary consequences. As Futuyma summarizes Dawkins' ideas about selfish genes,

Natural selection within population can be understood simply as a competition among alleles, the winner being the one that confers some characteristic on organisms that provides that allele with the highest rate of survival and reproduction, averaged over all the gene combinations in which that allele occurs.

This perspective works well in some cases, but the gene's-eye view doesn't work as well in other cases - such as the phenomenon of overdominance. The classic case is the sickle-cell allele and malaria resistance. In this case, people who are heterozygotes, carrying one normal beta-hemoglobin allele and one sickle cell allele, are more resistant to malaria than people with two normal beta-hemoglobin alleles. This means that, in one particular combination, a normally harmful allele (the sickle cell allele) has a beneficial effect. But averaged over all combinations, the sickle-cell allele is harmful. Overdominance is not a situation that is not easily predicted from a gene's-eye perspective.

On the other hand, overdominance seems to be rare. And we're again reduced to noting that biology is so diverse, that in some cases one approach is fruitful, while in another case, we should use a different perspective.

4. The most controversial aspect of Dawkins' selfish gene argument is related to altruism. Dawkins lays it out in the introduction to the 30th anniversary edition of The Selfish Gene:

We should not be surprised to find individual organisms behaving altruistically 'for the good of the genes', for example by feeding and protecting kin who are likely to share copies of the same genes. Such kin altruism is only one way in which gene selfishness can translate itself into individual altruism...

I'm going to sound like a broken record: it depends. There are many, many clear cases of kin selection, and calculations of altruism based on genetic relationships, like this one:

In eusocial Hymenoptera such as ants, wasps, and bees, diploid females develop from fertilized eggs and haploid males from unfertilized eggs. As a result, queens are equally related to their sons and daughters, whereas workers are more related to their sisters than to their brothers. These asymmetries in relatedness suggest that queens should favor an equal investment in both sexes, whereas workers should favor greater investment in females than in males. Hence, a sex ratio conflict arises between queens and workers, because workers may enhance their inclusive fitness by altering colony sex ratios in their favor, and in so doing act against the interests of the queen. The resolution of such conflicts provides important insights into the role of kin selection in social evolution.

The conclusion of this paper: kin selection happens.

By eliminating males, workers preferentially raise the sex that yields the largest marginal fitness-return per unit investment, thereby enhancing their inclusive fitness. This implies accurate discriminatory abilities at two stages: First, accurate assessment of queen mating frequency, which suggests great diversity of genetically determined odor cues. Second, discrimination between male and female brood relatively early in development, before males are so large that it would be too costly to kill them. These findings emphasize the sophistication both of worker reproductive strategies and the recognition abilities on which they often depend. More generally, they illuminate some of the complex dynamics between cooperation and ongoing conflicts among members of insect societies.

Kin selection has be extremely successful at predicting the social dynamics of many animal social behaviors - in birds, mammals, reptiles, amphibians, and insects. (Futuyma goes over a long list in his textbook, p. 597-599). Biologists have tested the theory with experimental populations, and with natural experiments. In case after case, we can explain cooperative behavior in the animal world by calculating genetic relatedness.

Calculating relatedness and making predictions isn't the same thing as finding a gene that influence altruism. Now researchers are starting to find those genes. In a paper in last year's Genetics, researchers identified regions of DNA that control worker bee behavior. In the honeybee species Apis mellifera, female workers give up their chance to reproduce by not laying eggs. Reproduction is left to the queen. This behavior is what kin selection predicts, but what genes are involved? Can we find 'cheater' mutants that cause female workers to go ahead an lay eggs?

As the authors write:

In societies where individuals act to benefit other members of the society at a cost to their own direct fitness, there is a selective advantage for individuals that "cheat," reaping the benefits of group living while avoiding the cost of contributing personally. Where genes influence cooperation among individuals, single mutations at key loci may permit selfish behavior to arise that advantages the carrier, but reduces the fitness of the group. Thus, identifying mutations for cheating behavior provides the opportunity to characterize the genetic architecture of cooperation—a key goal of sociogenomics.

Such cheater strains of bees exist, called 'anarchist lines'. By crossing anarchist males with normal queens, the researchers were able to find an 'anarchy locus' that influences altruistic behavior.

So what we have is this: a theory (kin selection) that predicts that, in many cases, natural selection will favor altruistic behavior towards kin, and we have a definite locus that does in fact confer altruistic behavior - a region of DNA that contains a variant of a gene that causes female workers to cease laying eggs in the presence of a fecund queen.

Kin selection is a powerful theory.

But biology is not like physics, and kin selection is not an absolute law - there are clear cases where other factors can take precedence of genetic relatedness, most notably cultural pressures in human societies (although clearly relatedness has an influence on altruism even among humans).

The main lesson here is that, under the heading of 'selfish genes', we can find some exteremely powerful ideas that have found substantial support in the lab and in the field. Whether or not the term 'selfish' is the best one to use (and Dawkins' has expressed ambivalence about the term, and the everyday connation of the term has little relevance to the actual, rigorously developed scientific ideas), thinking about evolution from the perspective of alleles is extremely useful.