Updating algorithms to cut AI bias, creating iterations of ML: Amazon Web Services’ Klein

Jeff Bezos-led Amazon’s cloud services division, Amazon Web Services (AWS), has been continuously working on updating algorithms to remove bias from its artificial intelligence (AI) services and solutions, a top executive told TechCircle.

“Looking into stopping bias in AI is a continuous effort. For example, we have updated our Rekognition engine nearly seven times after its release to make sure that there are no biases,” Olivier Klein, head of emerging technologies, Asia-Pacific (APAC), at AWS, said.

Explaining further, Klein said that AWS follows a unique approach of a small-scale rollout and rollback in order to train its machine learning (ML) models.

“We advise our customers to first test the ML model on one per cent of the entire dataset and see if it is yielding satisfactory results that can be monitored via metrics. If it yields good results, then it can be scaled up gradually in terms of deployment; if not, it should be automatically scaled back, iterated and tried again,” he explained, adding that AWS’ ML platform SageMaker allows an automated rollout and rollback.

The APAC’s head for emerging technologies also said that his company is partnering third-party service providers and system integrators to ensure that its AI services run without biases at the customer end.

Asked specifically about ML, Klein said that the company was seeing a rising demand in terms of personalised services from its customers. 

“Since we are driven by consumer demand and feedback, we have launched services such as Amazon Personalise. It is an ML service that makes it easy for developers to create individualised recommendations for customers using their applications. Developers can also choose to add demographic information from users such as age or geographic location to the service,” Klein said.

He added that Amazon SageMaker also has capabilities to add iterations of ML services, especially after the addition of third-party ML models to the company’s marketplace. 

Speaking about augmented reality (AR) and virtual reality (VR), Klein said that the company’s Amazon Sumerian service is now generally available in India and is seeing demand from not only retail or automotive customers but also from firms that are looking to “drastically” change customer experience. 

“AR and VR can be a great asset to retail, not only from the customer experience point of view, but it can also help in looking and understanding customer data at a group or an individual level,” Klein said, adding that the demand of a personalised experience from endpoint customers is also helping in the proliferation of such technologies.

Interestingly, he also pointed out that technologies like AR and VR are also being driven by the demand for mobile experience, which in the APAC region and especially in India, holds significant importance.

“A lot of people in the APAC region will experience the internet for the first time via the mobile phone,” he said.

Klein also said that the demand for mobile experiences is also leading to the uptake of services such as Amazon Textract, showcased last year. 

“Amazon Textract is a service that automatically extracts text and data from scanned documents. Amazon Textract goes beyond simple optical character recognition (OCR) to also identify the contents of fields in forms and information stored in tables,” Klein explained.

In an earlier interview, Craig Stires, AWS’ APAC head of AI, analytics and big data, said that the company was recruiting rapidly in the three other markets in the region owing to the demand for services related to emerging technologies.

“We are absolutely growing the teams,” Stires said. “We have been expanding them to provide the right kind of technical and customer support in all the markets in the region.”