Imagined!

You lost A game, but not the game you cared about playing - winning for the team. For example, I would not consider losing your life while rescuing that of another person "losing", as long as that was your intention. If your intention was to live long and prosper, then yes, you failed.

By cooperating, you guarantee that the absolutely worst outcome for the team (both prisoners defect) will not happen. It is possible to have goals involving entities greater than oneself, even if some individuals don't.

You can unilaterally decide that you are on the other prisoner's team, even if he does not realize it or agree. All that matters is that YOUR goal is "Prisoner team wins".

Humans are selfish by nature.

I think you might want to read The Selfish Gene. It explains how genes, not individuals, are selfish by nature, and will sacrifice the human carrying them if it can help their cause, replication. An example is suicidal rescue missions of one's own relatives (especially children and siblings at reproductive age). By force of natural selection, the best genes "know" that relatives, neighbors and fellow human beings are likely to carry instances of the same genes (in decreasing order of probability).

Whenever I read about the (non-repeated) prisoner's dilemma, someone claims that the "rational" choice for either party is to defect, because it yields the highest payoff for one player. This seems to ignore an important point:

The game involves three players - "prisoner one", "prisoner two" and "prison". If the prisoners form a team, it will be better for the team if both of them cooperate. There doesn't have to be any wishful thinking, but simply a goal of doing better for the team. You can never improve the score for the team by defecting.

What is irrational or not always depends on what your goals are.

I *KNOW* that I could make a simple web app launched in firefox and attached to a locally-running apache instance on a linux box NOT MISS A SINGLE VOTE.

I guess the problem was that these people also "knew", and thus didn't see the need to actually test the interface on a sufficient number of people - There's a 95% change that at least 1 out of 150 random testers would fall victim to a 2% failure rate. If you allowed the testers to leave feedback, the mistakes could probably have been discovered a lot faster. I'm basing this on #27545121, which claims this was a user interface issue.

They could have put the machine up in a mall, and let people use it to leave customer feedback or something, with a chance to win a small prize.

But most importantly, for 99,9% of the voters, it is impossible to understand the system, let alone verify the actual vote.

To verify the system, only a small absolute number (not percentage) of people needs to verify it. Assume 1% of the votes are incorrect and 500 random (from the cheater's perspective) people verify their hashes. The probability that none of these are victims of a forged vote is 0.65%. If only 0.1% of the votes are tampered with, you need 5000 people to achieve a similar percentage.

Your made up number of 0.1% of the people checking the hashes will thus be very resilient for voting populations greater than 500,000.

As for preventing the insertion of fake votes, you need to publish a list of who voted, and compare the length of this list to the length of the vote list. This list can also be verified by random sampling.

You don't need to. You just tell them to look up the hash in the local newspaper after the election (if they want to), and disregard the random numbers they got along with it. Other people will check that the hash matches the random number and social security number.

If you gave up secret voting, you could likely make a 'secure enough' voting system, since anyone could check their own vote in the system.

This is actually a solved problem. When you vote, you get a unique random sequence of characters. After the election is completed, a list of all votes is published. Next to each vote, the SHA1 sum of the voter's personal ID number concatenated with the random characters is listed. Example (truncated SHA1 sums):

64038c437f2c republicans

aea7fb41626d republicans

86895065f81f democrats

0ee79f4948b0 democrats

The random characters are never stored by the voting system, only the resulting hash. Any one person can verify that his vote has been counted, because he knows his own random characters and resulting hash. No one can find out what anyone else voted, because they don't have the random characters.