Slashdot Mirror
Amazon 'Reviewing' Its Website After It Suggested Bomb-Making Items (nytimes.com)
An anonymous reader shares a report: Amazon said on Wednesday that it was reviewing its website after a British television report said the online retail giant's algorithms were automatically suggesting bomb-making ingredients that were "Frequently bought together." The news is particularly timely in Britain, where the authorities are investigating a terrorist attack last week on London's Underground subway system. The attack involved a crude explosive in a bucket inside a plastic bag, and detonated on a train during the morning rush. The news report is the latest example of a technology company drawing criticism for an apparently faulty algorithm. Google and Facebook have come under fire for allowing advertisers to direct ads to users who searched for, or expressed interest in, racist sentiments and hate speech. Growing awareness of these automated systems has been accompanied by calls for tech firms to take more responsibility for the contents on their sites. Amazon customers buying products that were innocent enough on their own, like cooking ingredients, received "Frequently bought together" prompts for other items that would help them produce explosives, according to the Channel 4 News.
Looks like you're building a pipe bomb. Would you also like some acid to throw in white women's faces in case it doesn't go off? How about a fidget spinner in case both of your attacks fail?
The algorithm isn't faulty. It works exactly as designed. But it is also completely blind to the deeper meaning of the result. Take this as a cautionary tale against all software-augmented decision making. Software is not inherently fair and impartial. It just blindly follows a rigid set of rules that don't include any moral values. And sometimes, the developer may have even made the rules intentionally malicious.
In recent days amazon has been found to be deleting reviews of Hillary Clinton book that are negative - they justify this by saying that "no one could have read the book that fast" yet they don't block the great reviews from people who have had the same amount of access to the book as the negative reviewers.
That's not accurate - it seems they removed non "Verified Purchaser" reviews, which were predominantly one star; it does seem likely that most of those people might not have read the book, and certainly they didn't get it from Amazon. From Slate:
Amazon has since removed hundreds of reviews—both positive and negative—from unverified reviewers, but since these were overwhelmingly in the one-star camp, the book’s rating has now jumped from a 3.2 rating to a 4.9 rating overall.
Amazon has been deleting negative reviews from people it can't verify have bought the book from it. That's entirely reasonable - the book is being review bombed heavily, and the goal is to provide reliable reviews (it's not your political free speech soapbox).
There are plenty of people who have bought the book from Amazon, had time to actually read it and decided to leave a review. Not all of them are positive, but the barrier to entry (the cost of the book + time) does mean that people who review it are at least interested in the content.
They are doing the same thing with Zoe Quinn's new book. Steam is doing something similar with game reviews, because people review bomb them in response to some random thing the developer posted on Twitter.
const int one = 65536; (Silvermoon, Texture.cs)
SJW, n: "Someone I don't like, and by the way I'm a fuckwit" - AC
You would think they would be more concerned with why so many people are buying these items together that the algorithm is showing them as frequently bought together.
I browse on +1 so AC's need not respond, I won't see it.
Well, I suppose this is an occasion for Bayesian reasoning. The probability that you buy bomb ingredients given that you're a terrorist being high doesn't mean that the probability that you are terrorist given that you buy bomb ingredients is necessarily high. In fact even if someone is *definitely* building a bomb, it doesn't automatically mean they are a terrorist.
On YouTube there's a whole genre of videos devoted to large Tannerite explosions. Tannerite is an impact-initiated binary explosive that is popular for making shooting targets. Normally it's used in small quantities but of course there are many videos of people setting off very large Tannerite explosions, involving hundreds of pounds of the stuff.
Some people are just fascinated with explosives. So what you have to ask is whether it is more common to be into blowing things up for DIY amusement or blowing things up to hurt people. Where people go wrong with this kind of question is they rely on their intuition in guessing prior probabilities. If blowing things up holds no fascination for them, they assume that that kind of thing must be rare, or even non-existent.
Post may contain irony: discontinue use if experiencing mood swings, nausea or elevated blood pressure.