I reported on a story last April in which the algorithms of two used-book-listing bots resulted in a $23.7 million used textbook. Now here are a pair of recent stories about it happening again—in both directions.

First of all, Carlos Bueno wrote a self-published children’s book called Lauren Ipsum, about understanding how computers work. He priced it at $14.95 on Amazon—then he discovered a pair of third-party-vendor bots listing the book at around the $55 mark. These bots would, presumably, place an order from Amazon for the $15 book as soon as someone ordered the $55 one, then sell the $55 one and make a profit. (It’s unclear why anyone would order a $55 version when the $15 version is right there, but like the old joke about the million-dollar watermelons, they probably only need it to happen once for the bot-meisters to earn a profit on what it cost to set up the bot.)

But the truly brilliant thing is that these bots then started competing with each other, each trying to undercut the other, and then the other reciprocating, as their algorithms took a look at what each one was pricing the book for. And apparently there were no lower bounds on either’s algorithm, so they were soon pricing below the book’s original retail cost.

The punchline is that Amazon itself is a bot that does price-matching. Soon after the marketplace bot’s race to the bottom, it decided to put my book on sale! 28% off. I can’t wait to find out what that does to my margin. (Update: nothing, it turns out. Amazon is eating the entire discount. This is a pleasant surprise.)

So now Amazon is selling Bueno’s $14.95 print-on-demand book for $10.76, and paying him the full amount he would have gotten at $14.95 for each copy. (It’s available for $4.99 for the Kindle.) There are five vendors selling it new from $10.76…and two selling used at $44.99. Funny, normally being used causes the value of a book to go down. (Found via TechCrunch and BoingBoing.)

At the other end of things, there’s a story on ReadWriteWeb by Marshall Kirkpatrick about a book called How to Survive Personal Bankruptcy that can’t be bought in paperback form for less than $2,344,682.63. (Though you can buy it on the Kindle for $7.45, thus saving enough money to buy 27,584 $85 Kindles—one of the best arguments for the economic benefits of e-books that I’ve ever seen!) And that’s only one of two copies; the other costs $7,534,585.41. At a price like that, it might be better titled How to Cause Personal Bankruptcy. Weirdly, both copies are available from the same seller, so it’s basically competing with itself.

Kirkpatrick goes back to that $23.7 million textbook article I mentioned for its analysis of what happened—one bot tries to price slightly under the other, and the other bot tries to price at a greater amount more than the other bot, and they get caught in a loop that trends upward. (Or, if the lower price difference is greater, it trends downward.)

Kirkpatrick also talks to a couple of bot experts to find out what causes it and what Amazon could do to prevent this sort of problem. Machine learning expert Neal Richter, Chief Scientist at the Rubicon Project, said that the problem was that Amazon’s simple APIs and transparent pricing mean that “anyone with minimal skill” can can try to make a quick buck on this sort of arbitrage. He suggests that Amazon charging a fee, such as 5 cents, any time someone made a price change would give bot-writers incentive to improve in a hurry.

CTO Charles Wooters of AI provider NextIT suggests an alternative would be for bot coders to make it easier for humans to spot errors in pricing and modify them. (Though it seems to me that still relies on them having incentive to do so; there aren’t currently any penalties for bad bots other than the books failing to sell.)

Regardless, it’s certainly not in doubt that runaway pricing bots can make for some amusing Amazon browsing experiences. Just be careful clicking that “buy” button.