Horror Stories From Inside Amazon's Mechanical Turk

Artificial Intelligence is underpinned by tedious "Human Intelligence" micro-tasks, like meticulously describing elements from a data set. Mechanical Turk is Amazon's platform where gig workers can perform these tasks for tiny sums. Several surveys of MTurk workers allege excessively low rates or no payment as well as traumatic tasks, like describing disgusting content: images of decapitation, videos of botched surgeries, etc.

[WOR-2] See reference details

Your doctor’s computer may have suggested opioids because drugmakers paid to tweak the software, DOJ says

American start-up Practice Fusion, which sells professional software to doctors, was paid by unspecified drug companies to encourage physicians to overprescribe opioid painkillers at the height of the opioid crisis in North America. Practice Fusion paid a federal settlement but continues to sell its software.

[WT-4] See reference details

Amazon Choice label is being 'gamed to promote poor products'

The "Amazon's Choice" label is not attributed by employees. It is automatically awarded according to undiscolosed criteria including user reviews. Unscrupulous companies can generate fake reviews to "earn" this recommendation.

[WT-5] See reference details

How India’s data labellers are powering the global AI race

This article explores data labeling companies in India, where employees label image after image to train autonomous vehicle AIs. This manual task which underpins all Artificial Intelligence could itself be automated in time.

[AI-3] See reference details

Une intelligence artificielle a conçu un antibiotique surpuissant : c’est une première

An AI used in an MIT lab developed powerful antibiotic halicin (an homage to HAL, the AI in 2001: A Space Odyssey). When tested on mice and ex vivo human cells it effectively killed many antibiotic resistant pathogens including the one responsible for tuberculosis.

[AI-4] See reference details

To fight the coronavirus China is undertaking its biggest experiment yet in rule by data: an app that automates quarantines. Alibaba software, using unexplained data, now decides whether millions can leave their home. It also shares data with the police.

As the Covid-19 pandemic was just getting started outside of China, the New York Times explored how China was assigning color-coded QR codes (green, yellow, red) to its citizens depending on their health status: quarantined or not. After analyzing the application's code, NYT reports that it does more than indicate whether a person poses a contamination risk: it also transfers data to the police. The app used is the popular AliPay wallet application. Many countries later adopted (black-and-white) QR codes on dedicated smartphone apps as the preferred means to signal vaccination or medical status during the pandemic.

Note from the archivist: "I think that after this project my anxiety levels will be so high that I'll go raise goats in the mountains," says a generally upbeat developer on the Monkey Moon forum.

[SUR-4] See reference details

Fooling Facial Detection with Fashion

To fool facial recognition software integrated into videosurveillance systems, this blog article presents histogram designs which can be printed on masks.

Note from the archivist: This article was written in 2019, before Covid-19 made public masking so widespread.

[SUR-5] See reference details

British Cops Want to Use AI to Spot Porn—But It Keeps Mistaking Desert Pics for Nudes

The London police hoped that machine learning would help them to detect child pornography images on the Internet. The system has been a disappointment, sometimes mistaking images of the desert for naked bodies.

[SUR-7] See reference details

How I hacked modern Vending Machines

Vending machines are sometimes equipped with software to manage user accounts. A hacker explains how he hacked his university's vending machines by taking advantage of vulnerabilities in the code to access unlimited funds. He then alerted the supplier.

[WT-7] See reference details

Comment Facebook utilise son datacenter pour chauffer la ville danoise d'Odense

Meta manages two data centers covering 50,000 square meters in Odense in the Netherlands. The heat they generate is supplied to a local urban heating network, which distributes it to heat homes in town.

[SOL-6] See reference details