Justin is correct in his explanation of evolution. I just want to add this caveat from Dawkins. He argues, and I think he is correct, that in a small way, measured by specific adaptations over time, evolution should be called "progressive."
This is from Dawkins' review of Stephen Jay Gould's "Full House". http://www.simonyi.ox.ac.uk/dawkins/WorldOfDawkins-archive/Dawkins/Work/Reviews/1997-06fullhouse.shtml
"To evolution: is it progressive? Gould's definition of progress is a human-chauvinistic one which makes it all too easy to deny progress in evolution. I shall show that if we use a less anthropocentric, more biologically sensible, more 'adaptationist' definition, evolution turns out to be clearly and importantly progressive in the short to medium term. In another sense it is probably progressive in the long term too.
"Gould's definition of progress, calculated to deliver a negative answer to the question whether evolution is progressive, is "a tendency for life to increase in anatomical complexity, or neurological elaboration, or size and flexibility of behavioral repertoire, or any criterion obviously concocted (if we would only be honest and introspective enough about our motives) to place Homo sapiens atop a supposed heap." My alternative, 'adaptationist' definition of progress is "a tendency for lineages to improve cumulatively their adaptive fit to their particular way of life, by increasing the numbers of features which combine together in adaptive complexes." I'll defend this definition and my consequent, limited, progressivist conclusion, later.
"Gould is certainly right that human chauvinism, as an unspoken motif, runs through a great deal of evolutionary writing. He'll find even better examples if he looks at the comparative psychology literature, which is awash with snobbish and downright silly phrases like 'subhuman primates', 'subprimate mammals' and 'submammalian vertebrates', implying an unquestioned ladder of life defined so as to perch us smugly on the top rung. Uncritical authors regularly move 'up' or 'down' the 'evolutionary scale' (bear in mind that they are in fact moving among modern animals, contemporary twigs dotted all around the tree of life). Students of comparative mentality unabashedly and ludicrously ask, 'How far down the animal kingdom does learning extend?' Volume 1 of Hyman's celebrated treatise on the invertebrates is entitled 'Protozoa through Ctenophora' (my emphasis) – as if the phyla exist along an ordinal scale such that everybody knows which groups sit 'between' Protozoa and Ctenophora. Unfortunately all zoology students do know – we've all been taught the same groundless myth. "
****
"Creationists love Sir Fred Hoyle's vivid metaphor for his own misunderstanding of natural selection. It is as if a hurricane, blowing though a junkyard, had the good fortune to assemble a Boeing 747. Hoyle's point is about statistical improbability. Our answer, yours and mine and Stephen Gould's, is that natural selection is cumulative. There is a ratchet, such that small gains are saved. The hurricane doesn't spontaneously assemble the airliner in one go. Small improvements are added bit by bit. To change the metaphor, however daunting the sheer cliffs that the adaptive mountain first presents, graded ramps can be found the other side and the peak eventually scaled. Adaptive evolution must be gradual and cumulative, not because the evidence supports it (though it does) but because nothing except gradual accumulation could, in principle, do the job of solving the 747 riddle. Even divine creation wouldn't help. Quite the contrary since any entity complicated and intelligent enough to perform the creative rôle would itself be the ultimate 747. And for exactly the same reason the evolution of complex, many-parted adaptations must be progressive. Later descendants will have accumulated a larger number of components towards the adaptive combination than earlier ancestors."
"_The evolution of the vertebrate eye must have been progressive._ Ancient ancestors had a very simple eye, containing only a few features good for seeing. We don't need evidence for this (although it is nice that it is there). It has to be true because the alternative – an initially complex eye, well-endowed with features good for seeing – pitches us right back to Hoyle country and the sheer cliff of improbability. There must be a ramp of step-by-step progress towards the modern, multifeatured descendant of that optical prototype. Of course, in this case, modern analogs of every step up the ramp can be found, working serviceably in dozens of eyes dotted independently around the animal kingdom. But even without these examples, we could be confident that there must have been a gradual, progressive increase in the number of features which an engineer would recognize as contributing towards optical quality. Without stirring from our armchair, we can see that it must be so."
Jerry