The coming memetic takeover will be a kind of genetic takeover - and there are good reasons to believe that these are likely to be disruptive evolutionary events - accompanied by mass extinctions and the loss of significant quantities of adaptive information. A rapid transition seems as though it would be undesirable - with an increased chance of things getting lost or damaged.
One of the conclusions my my studies so far so far has been that the geologically-recent explosion of cultural evolution that we are now witnessing was triggered more by social networking skills than by factors associated with brain size or intelligence.
Human behavioural imitation apparently required the complex ability to mentally put yourself in another person's shoes while watching them perform tasks. For our ancestors, this appears to have required relatively advanced cognition - as ably explained by Susan Blackmore in The Meme Machine. So, there is a limited sense in which the conventional wisdom that "intelligence did it" is correct.
However, in memetics, large brains are seen more as a consequence of cultural evolution than a cause of it. Big brains have evolved to be meme nests. Large brains are the nervous system equivalents of ant domatia. They are homes for memes. Psychological support for cultural transmission - rather than intelligence - was really the key here.
This strongly suggests that sub-human level machines could effectively reproduce the human explosion in cultural evolution. Machines can copy each other easily. We can engineer them to be social. I think this means that we can forget about attempts to directly reproduce "human level" machine intelligence, and work instead on swarms of relatively stupid minions. Then the power of collective intelligence and the wisdom of crowds can be used to get them to perform useful work for us. It will be a new kind of society of mind.
Machine progress has occurred largely by them being strong in domains where we are weak. If the aim is to reproduce human cultural evolution in a machine-based substrate (in order to better make progress) then a frontal attack on directly reproducing human cognition in machines doesn't seem to make very much sense.
To a large extent, I think that using swarms of minions is largely what humans have been doing anyway. We do already have huge numbers of not-too-smart computer systems - and we have been a putting considerable amount of effort into networking them together.
Today, much of the main action is in the process of moving out of brains and into data centers. I expect data centers to become the main social centers for machines. The coming explosion of machine intelligence looks set to take place in the computing cloud.
Interestingly, data centers are usually out of town - where land is cheap. The centers of machine civilization and human civilization thus look set to be geographically separated - although intimately connected by high-speed networks. This will create an interesting dynamic.