Amazon Web Services Announced Effort to Reduce Bias in Machine Learning

By Jemima McEvoy Sunday, December 20, 2020

Earlier this month, Amazon Web Services (AWS) announced a new capability entitled SageMaker Clarify, the commerce giant’s attempt to combat biases that have emerged in machine learning. Though still extremely new to the tech space, this announcement has stirred excitement in the industry. Machine learning (ML), defined as an application of artificial intelligence (AI) that gives machines the ability to automatically learn and improve from their experiences without being programmed to do so, has long been riddled with largely ignored systemic issues. Here’s what Amazon is trying to do to address them.

Bias in Machine Learning

In 2017, it was still considered difficult to deploy machine learning models. As a result, these machine learning models were predominantly limited to major companies with large amounts of resources. Amazon has played a large part in changing this standard. In November 2017, the company launched its own machine learning platform called Amazon SageMaker, allowing developers to build, train, and deploy machine-learning models in the cloud.

“Customers saw it was much easier to do machine learning once they were using tools like SageMaker,” explained Bratin Saha, Amazon’s vice president and general manager of machine learning services and Amazon AI. “Machine learning was no longer niche; machine learning was no longer a fictional thing. It was something that was giving real business value.”

But as Amazon customers came to rely on the technology (SageMaker was actually one of the fastest-growing services in AWS history), its flaws became clear. One key flaw is unintentional biases in the modeling process. A machine learning model is heavily impacted by the person who created it, meaning that data scientists’ backgrounds can influence the way machines are programmed to learn. The consequence of this is unintentional preferences and biases that can lead to problematic outcomes like the reinforcement of certain stereotypes.

An example of this is facial recognition systems, which have proven extremely accurate at identifying white faces but not people of color due to the racial makeup of the data engineers behind the algorithms. “AI software is only as smart as the data used to train it,” explained a 2018 New York Times article about biases in facial recognition software. “If there are many more white men than Black women in the system, it will be worse at identifying the Black women.”

These biases aren’t just racial. A study of machine learning algorithms found that discriminatory behavior was also prevalent in an algorithm that recommended jobs in Science, Technology, Engineering, and Math (STEM) fields. “This algorithm was designed to deliver advertisements in a gender neutral way. However, less women compared to men saw the advertisement due to gender-imbalance which would result in younger women being considered as a valuable subgroup and more expensive to show advertisements to,” explained the report.

Amazon’s Solution — Clarify

This is where Amazon Web Service’s recent announcement comes into play. The tech giant is constantly updating its SageMaker tool. In fact, it has added 50 new capabilities to it in the past year — nine of which were rolled out during its re:Invent conference. However, this appears to be AWS’s biggest effort to date to address these bias problems. As explained in Amazon’s official announcement, published on December 8, SageMaker Clarify will attempt to reduce bias in machine learning through increasing the transparency and explanation around ML model behavior.

“Unfortunately, even with the best of intentions, bias issues may exist in datasets and be introduced into models with business, ethical and regulator consequences,” read the announcement, noting that SageMaker Clarify capabilities will allow data scientists to:

  • Detect bias in datasets prior to training and in models after training.
  • Measure bias using statistical metrics.
  • Explain how feature values contribute to the predicted outcome, both for the model overall and for individual predictions.
  • Detect bias drift and feature importance drift over time.

The price of SageMaker for companies’ use depends on the usage, which can be calculated using AWS’s pricing calculator. The Clarify tool comes at no additional cost, according to Amazon’s announcement.

Impacts of the Announcement

It’s unclear how much money SageMaker brings in for Amazon. However, considering that it’s one of Amazon’s fastest-growing services and has been pioneering in the machine learning space, it can be assumed that it’s a very valuable service for the company, whose stock is currently worth over $3,000 and whose founder, Jeff Bezos, is now the richest person in the world with a net worth of over $180 billion. Amazon itself is worth over $1.5 trillion, making it among the world’s richest, just like its founder.

As it’s still in its early days, time will tell which companies pick up the service, which only became available for use last week. However, it’s already being lauded by critics. For example, A Cloud Guru author Kesha Williams called Clarity “the most important announcement out of AWS re:Invent this year,” writing that she “literally cheered” when it was announced. “There are several capabilities that will stand out for customers,” said specialist news site IT Pro.

Amazon will have to compete with other companies that are already digging into solutions to this problem. However, with its name recognition and unique production of accessible tools, SageMaker Clarify could become a dominant tool.

About the Author


Headshot for author Jemima McEvoy

Jemima is a journalist who enjoys reporting on business, particularly small business and entrepreneurship.

Related Articles