"Smivs writes with word on a spider..." My first thought when reading this was "What kind of computer is a spider?" followed by "Why isn't he using Open Office?"

The Structure of Scientific Revolutions by Thomas Kuhn is a must read for anyone trying to understand how a theory (like evolution) is formed and evolves. It talks in depth about how new ideas challenge and eventually overthrow the established science, the difficulties involved, and how a paradigm eventually solidifies. His examples are mostly in physics and chemistry, but evolutionary biology had a very similar path to those described: a new theory is posited with powerful explainatory powers, although it certainly can't explain everything. Eventually, it is generally recognized as the most powerful and parsimonious explaination, although significant changes are made to its initial hypotheses. Something very similar happened with Einstein and physics, and Copernicus and astronomy. Of course, the problem is not that people believe stupid things about how science works, but that people in power believe stupid things about how science works. http://xkcd.com/154/

Several people have already mentioned this, but I have to add my vote for Goedel, Escher, Bach. I was 16 when I read it for the first time, and it completely blew my mind and made me want to understand mathematics as a way of understanding life. I think I'm a computer scientist today largely because of that book. After I read it, I tried to get everyone I knew to read it, but I don't think I got any takers. The large text deals with a very complex set of ideas, including Goedel's incompleteness theorem. But with a teacher's help, I think most highschoolers could get through it, and possibly have their minds expanded. I also like Chaos: Making a New Science by James Gleick, for the accessible and personal introduction to chaos and fractals.

Why Amazon? Is it just name recognition? They're probably the largest online retailer. Are they "the online walmart" like the article says?

Some things we can measure. http://www.bea.gov/national/xls/gdplev.xls http://www.census.gov/statab/hist/HS-01.pdf
GDP/person 1993 = $5043.65.
GDP/person 2008 = $47141.20.
In other words, in 1993 the average person made about $5043 a year, including adjustments for inflation. Of course it's not a perfect comparison, inflation can't account for the fact the we can buy computers and plane tickets now, so I'd say calling the average person in 2008 10 time richer than the average person in 1933 is probably an understatement.
One more number: real gdp declined 27% from 1929 to 1933. Until we get comparable declines, don't talk about approaching the great depression.

If you learn a programming language, at least you can write some useful programs. Also it's worth noting that learning a new programming language gives you insight and new perspective. Does someone who knows one religion have incentive to learn another?

People have already mentioned project euler, which is good, but most of those problems will be beyond the capacity of 12-year-olds, I think. Start really simple: write the absolute value function, the min and max functions, factorial, simple I/O interactions like "how old are you?". Build confidence in their ability to solve simple problems before you give them something more complex. A good one might be the "higher/lower" game. Write a program that picks a random number and then asks for a guess. It tells whether the guess is higher or lower than the true value. If you make them program an AI that can play the game, they may discover binary search all by themselves. I think problem solving is more important than being able to see pretty results.

I was going to post an exerpt, but /. destroys my formatting. http://poetry.poetryx.com/poems/784/

It seems like way too often I go camping and it rains, and even the "waterproof" tent lets water in, especially seeping underneath where you sleep...

For what it's worth, I've had good experiences with coq: http://coq.inria.fr/ Although I've never used it for anything large, it has the nice ability to make proofs about code, and export the code to haskell, scheme, or ML. I had a fun time proving that the min function always returns the lesser of the two values.