Company targets new audiences with machine learning announcements
AWS re:Invent 2020, held as a virtual event this year for more than 500,000 attendees, was once again a blitzkrieg of major announcements that collectively have the potential to reshape the course of cloud computing over the next few years.
Among the biggest news in several areas of AWS’ cloud portfolio, artificial intelligence (AI) and machine learning took top billing, with this year revealing a step change in capabilities and strategy.
Below, I take a look at some of the highlights and assess what they mean for the market and Amazon’s strategy, which is starting to move into some important new directions.
AI Goes Mainstream
AWS CEO Andy Jassy kicked off the event, highlighting that the pandemic has accelerated cloud adoption by several years and that AI is shifting from a niche experiment inside technical departments to becoming more mainstream in business processes, a trend CCS Insight has also observed. More than 80% of companies in our Senior Leadership IT Investment Survey, for example, fielded in July 2020, are now trialling AI or have put it into production, up considerably from the 55% reported in 2019.
According to AWS, “tens of thousands” of customers are now standardizing on Amazon SageMaker, its fully managed platform to build, train and deploy machine learning models, with “hundreds of thousands” accessing its AI services, such as Amazon Polly, Rekognition and Lex, twice as many as any other cloud provider. This includes customer Intuit, which claimed in the keynote address that it’s reinventing its culture with AI, as the number of machine learning models it has deployed has shot up by 50% in the past year.
AWS breaks down its capabilities into three domains: frameworks and infrastructure; SageMaker and AI services; and its suite of “off the shelf” models, developer APIs and business solutions. Reinforcing its mission to “put machine learning into the hands of every developer and business”, announcements covered these domains this year, with a particular focus on SageMaker and higher-level, applied AI solutions in areas like business operations, contact centres and industrial and healthcare sectors.
SageMaker Gets a Revamp, Including Bias for Responsible AI
Amazon SageMaker has become one of AWS’ most important products, and re:Invent 2020 saw several new additions to the platform, including SageMaker Data Wrangler, which provides tools to simplify the preparation of data, and SageMaker Feature Store, a fully managed repository within SageMaker Studio for the sharing of machine learning features — attributes in models used to streamline training.
Additionally, SageMaker Pipelines is an MLOps capability that provides a continuous integration and continuous delivery service, purpose-built for machine learning, and SageMaker Edge Manager helps customers operate machine learning models across fleets of edge devices up to 25 times faster.
But the most important improvement by far was Amazon SageMaker Clarify, which helps companies tackle bias in machine learning using tabular and statistical data. The service evaluates both training and inference data for bias by providing several statistical metrics about the data. It also monitors the performance of models in production against bias by checking how they perform against the training data.
Clarify also tackles model explainability using libraries of Shapley additive explanations hardened in the platform. By combining several of its products, AWS is integrating explanations into several areas of the machine learning life cycle from pre-training, to model analysis in training, to production and deployment.
Detecting bias in data and model behaviour, and tackling “black box” AI with greater system transparency through explanations are among the most important requirements we’re now seeing in the market ,so it’s great to see AWS really homing in on this. Although it’s a latecomer to this area, Clarify will be one of the most welcome improvements to SageMaker and a boost for customers who want more in the field of responsible AI.
Moving Up the Stack: Applied AI for Businesses and Industries
Another important area has been AWS’ continued expansion up the stack into higher-level services and solutions for businesses and vertical markets. This year, the company concentrated on three main fields for these solutions: business operations, contact centres, and industrial sectors and healthcare.
In business operations, it announced a preview of Amazon DevOps Guru, a fully managed service for software development that uses machine learning to help developers automatically detect operational problems and recommend fixes as part of their processes. It also launched Amazon Lookout for Metrics, an anomaly detection service offering root cause analysis and recommendations for operational time series data.
In business intelligence, and as an extension of its Amazon QuickSight service, it unveiled Amazon QuickSight Q, which uses natural language processing to enable businesses to ask questions about their structured data in everyday language. It also announced the integration of Autopilot, its automated machine learning solution, into Amazon Redshift and Neptune database products. This helps database engineers lacking machine learning experience to build and deploy models directly in those environments.
Lastly, there was a large set of announcements for Amazon Connect, its contact centre product. They included Amazon Connect Wisdom, a service that taps machine learning to enable contact centre agents to search their various corporate knowledge bases for relevant content when handling calls. The shift to remote operations during the pandemic has helped transform contact centres around the globe, and this has been a major boon for Connect. The platform signed on 5,000 new customers this year alone and is becoming an important showpiece for AWS’ solutions.
Upping the Ante with Vertical Market and Industrial AI Solutions
The areas grabbing the biggest headlines, however, were in solutions for vertical markets, particularly for industries hard-hit by the pandemic, such as healthcare and industrial sectors.
Building on its Amazon Comprehend and Transcribe Medical services launched in 2019, AWS announced Amazon HealthLake, a new, HIPAA-eligible cloud-based service that applies machine learning to large volumes of health and life sciences data.
It also released several products for industrial sectors aimed at improving assembly line production, quality management, worker safety and remote operations in factories and warehouses.
- Amazon Monitron is an end-to-end machine monitoring system that employs machine learning to enable predictive monitoring of machinery in industrial environments such as bearings, motors, pumps and conveyer belts. Comprising sensors, a gateway device and a mobile app, the system can be deployed in as little as an hour, analysing vibration and temperature data to identify potential failures or abnormal activity.
- Amazon Lookout for Equipment allows businesses to harness their existing Internet of things sensors to detect abnormal behaviour through machine learning, which is also used in Amazon Lookout for Vision to quickly analyse large volumes of images and spot defects or irregularities.
- AWS Panorama, an appliance and software development kit, enables businesses to enhance on-premises cameras with computer vision, allowing them to analyse and make AI-based predictions about the content of a video stream locally.
The moves signal several important changes in its strategy. Firstly, the firm is clearly putting its foot on the accelerator in the race to become the best cloud for industrial workloads and the transition to Industry 4.0. It’s also getting better at pulling innovation from its retail business, especially in the areas of fulfilment and factory operations technology. If there’s one tech firm held as the gold standard for industrial operational efficiency with new technology, that company is Amazon, and this could become a formidable asset as it competes in these industries against Microsoft, IBM and Google in the future.
AWS is also starting to reach new audiences for its AI beyond data scientists and developers, such as business intelligence professionals, operations teams, business analysts and database engineers. For machine learning to reach its potential in the enterprise market, it needs to be far more pervasive with business users who have little to no expertise with the technology. It is this gap that many of these solutions are starting to bridge.
What It All Means
What’s striking this year is that AWS is not only assembling a market-leading AI portfolio, especially in higher-level services, but in addressing business and industry problems it’s becoming much more purposeful with new releases. In becoming more adept at applying innovation from its parent retail business AWS is opening a direct line of communication to new audiences for its AI, particularly C-suite and business executives, as opposed to just the developer and IT professionals that it focussed on in the past.
At the same time, by tackling tricky areas with Amazon SageMaker such as data bias and explainability, MLOps, feature reuse and data preparation, it continues to address the immediate pains machine learning practitioners have with the technology.
Having launched an eye-watering 250 new machine learning capabilities in the past 12 months alone, AWS is also innovating and rolling out AI products at a pace matched by few players. Check out the below image I took at the event a couple of years ago and compare with the one from this year to see the progress AWS has made with its portfolio.
New Directions for AWS
But the firm isn’t without challenges as it looks ahead. At times, its pace can be bewildering so the company will need to simplify its portfolio and continue to focus on integration. This is particularly true for SageMaker, where the endless feature names can often blur important distinctions between simple features and highly strategic services. It will also need to integrate the platform with AWS Outposts for the growing number of customers who want to run machine learning on their premises.
Above all, AWS will also need to become more confident in not only responding to customer needs as the primary focus of its product strategy, but also anticipating their demands, offering more forward-looking products, and helping with implementation practices. Several emerging areas that went under the radar this year, such as AI security and privacy, are going to be important in the near future. According to our survey of senior leaders, for example, security and privacy for machine learning are now priorities for business leaders focussing on AI strategy, and both Microsoft and Google have ramped up their efforts in these areas in 2020.
Overall, re:Invent 2020 revealed a step change and new directions in AWS’ strategy, with announcements that have the potential to reshape the course of cloud computing over the next few years. A deeper focus on these areas will be vital for AWS to maintain its dominant position in 2021 and beyond.
A version of this article was first published by CMSWire on 16 December 2020.
Subscribe to our blog
Make sure you don't miss out on our fresh insights on topical news in the connected world
"*" indicates required fields