In the ancestral population, if some flower, just by chance, happened to look a little like a bee, bees might be more likely to visit it. They’d pollinate that flower, which reproduces and passes its genes on to the next generation of flowers. Again, if some flower in the new generation looked a little more like a bee, it’d be even more likely to be visited and pollinated. As this process continues over many generations, the flower population as a whole gradually looks more and more like the bees that pollinate them.

Note, this has nothing to do with what the flowers “want”. Flowers don’t want anything. Like all living things, each generation brings forth new varieties of individuals. just by chance. Those characteristics which make it more likely that their possessors will reproduce get passed on. The ones that don’t assist in reproduction tend to die out. Over time, if there is a consistent selective pressure towards certain kinds of characteristics, those are the characteristics that will eventually predominate.

As you might notice, the question is backwards. The complex mechanism that forms a phenotype (what a living thing appears to be) depends on its genetic ancestors, not the other way around.

Those blind ancestors didn’t make themselves over to resemble their pollinators; they were pollinated by those who recognized themselves in the random patterns of our subject. Those flowers that didn’t attract pollinators simply did not propagate. So the pollinators selected the flowers, that then appealed more to particular pollinators, and round we go. Co-evolution in action.

Another example of evolution’s predicitive ability can be found in paleontology: many of the ‘missing link’ fossils that have been found were predicted well before they were discovered. The recent discovery of Tiktaalik (a fish that could walk on land) is but one example.