Could you clarify/expound on this?
Evolution is both very powerful, with many people underestimating it, and profoundly stupid, with many people also overestimating it. Once humans really get cracking with the gene engineering (which has proved harder than we had initially hoped, but I'm still confident we'll get there), I'm sure we'll find a plethora of relatively simple changes we can make to natural organisms that may involve changing three or four genes simultaneously where no single change is advantageous, which is a thing that evolution essentially can't do.
[1]: Or at least, not yet... one bizarre-ish, "Selfish Gene" way of looking at human gene engineering is evolution evolving itself a way to make those multi-gene changes. Still not in a directed fashion, of course. But evolution will still apply to our own gene engineered creations, just possibly with a very different set of constraints, such as potentially severing the billion-year-old connection between fitness and direct biological reproduction... as I've said on HN before, you don't have to be a full-on "Rapture of the Nerds"-style Singulatarian to see that the future may get very weird....
So we're evolving in such a way to be able to make combinatorial multi-gene changes- by evolving brains big enough to be able to develop gene engineering?
But observationally, humanity is currently in a place where that level of gene engineering is a decent bet, yes.
It's a view on how ideas spread and use hosts (e.g., human brains) as carriers. They mutate, they're selected, they compete for "brain space". One funny perspective is that we're the slaves to genes and memes, both using us to survive. We're just carrying out their orders. :)
(In the end it's just a way to look at information diffusion with focus on the way how ideas replicate and the condition of their success or failure over time, abstracting away from social/cultural context and other hard to grasp things.)
Actually the more I think about it, the more I am convinced that multiple interacting species introduce a serious argument against the blind hill-climbing view of evolution. Essentially what you have is not a single point on a landscape blindly climbing in the direction of the gradient, but a probability distribution with a given mean point that could jump to another mean currently at its periphery given enough pressure / interference. E.g. a newly migrated predator eating away 90% of variance of the distribution of some existing sheep-like species, leaving only its tough "tail", thus severely relocating the centroid. Maybe this is something trivial I'm just realizing, I'm not a biologist.
However, in your first case, it sounds to me like you're talking about two changes, each of which are beneficial, or at least neutral. That wasn't what I was talking about, that happens all the time. I'm talking about changing two or three or four (or more) genes simultaneously in a way that each individual change, if applied on its own, would be detrimental, but the whole is an improvement.
In this representation of life it is directed and goal based. This is a misunderstanding of evolution that was common in a world that saw humanity as the peak of the process of evolution. For this reason Darwin perferred to think of it as a directionless bush rather than a directed tree.
1: http://en.wikipedia.org/wiki/File:Tree_of_life_by_Haeckel.jp...
What I mean is, a more adequate picture of the process is not a simple "path" analogy. It's more like you have a high dimensional optimization space and a wildly varying survival function, and the organisms are bouncing around this space in a random motion. This allows them to occasionally cross valleys and take very distinct paths to each far away maxima, specially if there are a few minor maxima along the way.
My point being organisms have a given capability to bounce around: they don't have to take a monotonically improving path, although this capability is limited because otherwise (if the mutation rate is too fast) the results are too chaotic.
They're somewhere in the middle of monotonically improving, guaranteed optimization (low mutation; gets stuck at local maxima) and completely random (high mutation; does not retain advantages well and leads to many deleterious results).
It can do amazing things, but only because it's had millions/billions of years and an entire planet to run on.