Amazon’s Customer Obsession Takes a New Form

Earlier this week, AIM covered the algorithm governing the bad shopping experience on Amazon’s e-commerce platform. Today, its cloud services platform AWS announced AI Service Cards, a resource material for its business customers to understand its AI services better and use the tools responsibly. 

Service Cards are a form of responsible AI documentation where customers can access information on the intended use cases and limitations, responsible AI design decisions, and best deployment and operating practices. 

The company said it would issue three types of Cards: Amazon Rekognition for Face Matching, Amazon Textract for AnalyzeID, and Amazon Transcribe—Batch. 

The company also said that the Cards add to their effort towards building Responsible AI. Along with Amazon SageMaker Clarify and Amazon SageMaker Model Monitor, the company will provide capabilities to businesses to detect any bias in datasets and models, as well as lead to better monitoring and reviewing of model predictions via automation and human intervention. 

Jessica Newman, director of the AI Security Initiative at the University of California at Berkeley, comments on the same, sharing that platforms are increasingly publishing such disclosures to emphasise Responsible AI practices. Newman further added that tech companies shouldn’t merely be depending on the goodwill of companies for safe use of company’s services. 

The launch of these Cards comes against the backdrop of concerns regarding Amazon’s AI services on its cloud computing platform that exhibited discrimination towards different groups. For instance, Amazon had to scrap its AI recruiting tool because it showed bias against women. However, several leaked documents suggest that perhaps the company has been working, albeit in secret, on upgrading its hiring algorithm. Although, one wonders if this recruiting tool will simply be served to its customers with a warning card, ultimately hoping that its business customers will figure it out in the end.

Amazon’s algo 

But it's not just AWS AI services that have raised concerns recently. Amazon’s e-commerce site is populated with sponsored product posts or with Amazon’s products, which has raised several eyebrows about whether Amazon is true to its primary leadership principle of being customer-centric. 

While it generates good advertising revenue for the company, Amazon’s push for sponsored products has been lethal to organic sellers who fight for a short window to rank higher in the search engine. Therefore, building trust in the algorithm makes it critical for Amazon to have its customers see what they are looking for besides what it wants them to see. 

Currently, Amazon is using the A10 algorithm for its product listing. The algorithm considers four key components: title and feature description, price, in-stock, and sales velocity. In addition, the sales velocity factor includes promotion campaigns, advertising strategy, product image, and seller authority. Although the company deems it as “organic search”, it has been leveraging upon the sales velocity factor to provide more visibility to ads—as well as products of its brands—which, for some reason, doesn’t count as “sponsored” and is often simply regarded as “highly-rated”.