Phil Plait, of Bad Astronomy at Slate, is usually a great read, and Bill Nye a great science entertainer, but I don't know what's up with them, with their calls to spend major money defending against Killer Asteroids. Plait even acknowledges that there isn't one confirmed human death from asteroid impact in history.
So how do asteroids turn into such a danger—to the extent you are (allegedly!) more likely to die from an asteroid than be falsely accused of rape? With some mathematical trickery, made clear on the other side of the Orange Oort Cloud.
Remember the observation that Bill Gates walks into a room and the "average" salary goes up by several orders of magnitude? That shows the sensitivity of a mean to outliers. Bill Gates is an outlier. Killer asteroids are an outlier. The risk comes almost entirely from amortizing deaths from catastrophic asteroids, or comets, like the one that is widely believed to have totaled the dinosaurs. Take an asteroid miles in diameter that kills 8 billion people, and assume an impact like that once every 1 million years. As a mean, that's 8000 people a year. But what is the frequency of loss?
There's a lot of uncertainty in asteroid-collision estimates [see especially tables 3-2 and 3-3], but they all have one thing in common: More than 99.9% of years, the expected number of deaths is zero. And that's a report for NASA by space scientists, looking to justify the anti-asteroid program! Somewhere like 99.99% of those less-than-once-per-millennium years where there is a fatality, there will be localized collisions whose death toll is less than the 2004 Indian Ocean tsunami [circa 230,000] and nowhere near manmade disasters like Great Chinese Famine of 1958–1962 [official, likely lowballed, estimate 15,000,000 deaths over three years]. Even Mao, however, is a piker compared to the Flu Pandemic of 1918–1920, perhaps 100,000,000 dead in three years. Tsunami warning systems and better flu shots (MERS and SARS?) look like much better bang for the buck. We aren't going to wait a million years for the next epidemic.
A 12th-grade statistic book will tell you to use a median instead of a mean to reduce the influence of outliers. (Median: half above, half below.) That is why we hear about the median salary, so that including Gates and Jamie Dimon don't provide a misleading view of the economy. However, the median number of anticipated annual asteroid deaths—indeed, the 99.9th percentile of annual asteroid deaths—is zero.
Silly use of the mean has real-world implications. By the most amazing coincidence, the W Administration did, dishonestly, report mean savings from his tax cuts, exactly to disguise they almost all went to fat cats. (Just to make sure of a good number, they also omitted people whose tax cuts were zero from the computation!) These likelihood calculations also distort our understanding of improbable events that nevertheless occur on a fairly consistent basis, such as false rape accusations. Our experience does not include any asteroid deaths, but that doesn't imply the guilt of the accused in the Duke lacrosse travesty. Indeed, Slate also ran a piece this week on the high likelihood of encountering improbable events, as long as you don't specify what unlikely event occurs, or to whom.
Huge lasers or a nuclear-armed space vehicle to destroy incoming asteroids are primarily expensive public works project for space scientists. Worrying about the killer asteroid ahead of so many other possible disasters that may occur in the next million years is a little like Ugg and Glugg worrying about the impact of Earth's first manmade fire, which took place about a million years ago, on Global Warming.
4:45 PM PT: Although it isn't, perhaps, clear, my main problem with the Plait/Nye line of reasoning is that imputed deaths/year for very low probability events that we will never experience or even hear of confuses our understanding of deaths/year of low probability events that are nevertheless frequent enough that we have personal or historical acquaintance of them (airline fatalities, confabulated rape charges, 1918 flu epidemic, Black Death, etc.). These two types of computations are not psychologically compatible, and they are used in this example to drum up support for expenditures that would otherwise be a tougher sell.