AWS this month suffered a setback to its agenda of expansion, when the city of Orlando, Fla. ended its facial recognition in law enforcement partnership with the Amazon Rekognition technology.
After 15 months of back and forth with this AWS facial recognition program, Orlando and Amazon have officially gone their separate ways. On July 18, Orlando declined to renew its second pilot program with Amazon Rekognition, AWS’ image analysis service. Bandwidth, video resolution and positioning issues plagued the use of the technology, and the city was unable to set up a reliable camera stream, Rosa Akhtarkavari, Orlando’s chief information officer, told the Orlando Weekly. The city simply lacked the IT infrastructure to support AI software, she said.
The first pilot began in December 2017 and ended in June 2018, before the second started back up again in October 2018. In theory, the city planned to use Rekognition’s facial recognition algorithms to identify and track suspects in real-time. If configured and supported properly, law enforcement officers would upload an image of a suspect and get notified if Rekognition found a match.
With Orlando out of the picture, Oregon’s Washington County is the lone police department that still uses AWS facial recognition technology. Both partnerships faced legal and media pressure from the American Civil Liberties Union which argued unchecked surveillance technology threaten privacy and civil liberties, and that Rekognition, in particular, misidentifies African-Americans as criminals at a higher rate than other races. Other cities, such as Oakland and Somerville, MA have banned government use of the software. And Amazon’s own employees and shareholders even wrote a resolution that called to halt AWS facial recognition sales to government agencies, though Amazon’s board of directors struck the missive down in its annual shareholder meeting in May 2019.
Amazon claims that the apparent racial bias occurred due to misuse and misunderstanding of the service. It has also argued that it’s up to the federal government, not Amazon, to legislate the use of this technology.
AWS expands CloudWatch, adds event-driven offering
While Amazon bore a blow to its AI ambitions this month, it still improved some bread-and-butter capabilities of the AWS platform. The recently announced Amazon CloudWatch Container Insights and Anomaly Detection capabilities, along with the expansion of its EC2 Spot Instance service, should expand AWS’ compute and monitoring flexibility.
AWS also added Amazon EventBridge, a serverless event bus that integrates users’ AWS applications with SaaS applications. As more of its customers turn to event-driven applications and architecture, AWS needs a better way to integrate and route real-time data from third-party event sources including DataDog or PagerDuty to service targets, such as AWS Lambda. EventBridge eliminates the need to write custom code that connects application events and should enable more efficient AWS event-driven architectures.
Amazon CloudWatch Container Insights and Anomaly Detection give users more ways to analyze their metrics and improve performance and security. CloudWatch Container Insights collects and organizes metrics from AWS’ container services and files them in CloudWatch’s automatic dashboard. It also handles diagnostics, which can help users identify issues such as container restart failures. Users can set alarms for certain container metrics, including use of CPU, memory or network resources. Container Insights is in open preview.
CloudWatch Anomaly Detection uses machine learning algorithms to analyze data regarding the performance of your system or application. This anomaly detection capability then analyzes the metric’s past data to generate a model of expected values and establishes a high and low value of accepted metric behavior. Users can then decide how and when they are notified. CloudWatch Anomaly Detection is in open preview and priced per alarm.
Spot Instances for Red Hat Enterprise Linux and Amazon SageMaker
Amazon EC2 Spot Instances let users obtain unused EC2 capacity at a discounted rate, and AWS recently extended those capabilities to users with a basic Red Hat Enterprise Linux (RHEL) subscription. Before, only premium RHEL subscribers could access Spot Instances. At its NYC Summit this July, AWS also revealed Spot Instances support for SageMaker users to train machine learning models, which AWS claims could cut training costs by up 70%.