In Kurzweil's projection, The Singularity arrives in a three-stage revolution, the GNR revolution - genetics, nanotechnology and robotics.

We are just entering the first, genetic phase. The nanotechnology phase will not really kick in till the 2020s.

The obvious problem with this projection, partly acknowledged by K, is that the genetic phase opens up the potential for such catastrophe (e.g. bioengineered viruses) that we might not get to the nanotechnology phase. This is a nuisance because nanotechnology could (theoretically) equip us with the weapons capable of staving off any biologial attack.

So according to K's narrative, nanotechnology is 'our' only hope, but we can't really reach it until we are in the shadow of the utmost peril. Then nanotechnology itself presents new dangers (which doubtless cannot be overcome until the robotics revolution is in full swing, but haven't got to that bit yet...)

The motor for the scenario is exponentially escalating peril and panic.

sd - Kurzweil might still be erring on the side of conservatism with some of his dates. Given not only increasing returns but cross-linkages and unpredictable thresholds, it seems bold in the extreme to forecast medium-term (decade-plus) delay schedules (with nanotech for e.g.). 'Existential' security threats also accelerate things remarkably, so a surge in the global Hobbes-factor could scramble everyone's expectations. I'd go out on a limb and say pretty much anything could happen in a decade, excepting only large scale extraterrestrial engineering projects.
Change in the epoch of Singularity Revolutions is not only nonlinear (logistic) in aggregate, but also polycatastrophic: disintegrated, erratic and aleatory - a breakthrough in any number of technoscientific lineages could have consequences that are impossible to circumscribe - molecular gigacivilizations triggered in a petri dish or self-assembling artificial intelligences speading like wildfire through the Internet.
Since the earth already hosts incomprehensibly vast quantities of DNA, all basically computationally inert, an intelligenic technoviral trigger that switched significant chunks of it into productive cognition would probably be the best radical accelerator. There are innumerable alternative scenarios however, and more every year (every week?).

Anyway, this seem right:
"The motor for the scenario is exponentially escalating peril and panic."

Being realistic about these speed questions requires some recognition of how incredibly slow neuronal processing is, with nerve cells spiking at about 100 hz. The only mitigating factor is the vast parallelism of biological brain tissue - which technological arrays can match or surpass in principle. Yet the components of a bionanotronic array would function at speeds at least a billion times faster, with terahertz processor speeds already entirely plausible. Cognitive traditions equivalent to the entire span of human history could be compressed into days, at most. Now all that's needed is a speck of self-propagating nanoneuronal goop mating with the fungal infection under a lab technicians fingernail ...

Nick - "erring on the side of conservatism with some of his dates"

Yes - he does hedge his bets a bit, but when he means nanotech really kicking in, he means internal nanobot brain scanning - neuron by neuron - to complete reverse engineering the brain, and nanotech organ replacement. According to his calculations, which take into account exponential doubling etc., these thresholds are not going to be reached much before 2020.

I agree that breakthroughs in other fields could trigger breakneck acceleration in nanotech - there are many 'whole new ball game' scenarios ahead, but there are clearly massive hurdles that simply stand in the way.

For example, spindle cells were only discovered a few years ago, so modelling and simulating them is a (relatively) long way off. Due to their complexity, they won't really be modelled until they can be scanned by nanobots - not really on the cards in the next ten years.

[high in babble factor - still trying to process TS]