You are here

March 14, 2018

Artificial intelligence (AI) remains one of the hottest technologies being assessed and implemented across enterprises and government agencies today. This is due in large part to AI’s ability to automate processes, increase efficiency, and empower each individual within an organization.

The transformative power of AI has piqued the interest of many government agencies and military organizations. To help answer their questions about what AI is, where it came from, and how it can benefit the government, we’ve focused an entire series of articles on the technology.

In our first article, Dave Vennergrund provided some background on the origins of AI and the technologies that have helped it become a revolutionary force in the workplace today. We then featured two interviews with emerging technology companies that are using AI to help improve employee service and better protect networks.

Although we’ve established that AI is as old as computing itself, it wouldn’t be the force that it is today—and will be into the future—without new innovations making it more prolific and powerful. And one of the largest enablers of new AI solutions is the cloud. That’s why in this next article in our AI-focused series, we’re talking with Kris Skrinak, a Solutions Architect at AWS, the world’s largest public cloud provider.

During our conversation with Kris, we analyze the role that the cloud has played in making AI what it is today, the tools AWS provides to innovators looking to build the next big AI solution, and what the future holds for AI and the cloud. Here is what he had to say:

Thinking Next (TN): Many experts attribute the cloud to the rise of AI tools and applications entering the marketplace today. How is the cloud a driver for advanced AI solutions? What role is the cloud playing in their development?

Kris Skrinak, a Solutions Architect at AWS
Kris Skrinak, Solutions Architect
Amazon Web Services

Kris Skrinak (KS): The cloud has democratized access to compute resources that were available to only the largest institutions just a few years ago. A state-of-the-art machine learning server with full storage and networking resources could cost over $60,000 for the hardware, and that doesn’t include the DevOps and hosting facilities required to turn it on and maintain its use. Now those resources are available for less than 40 cents a minute on-demand, and even less when taking advantage of pricing during off hours. This opens up access to students, researchers, inventors, and builders of all kinds in an unprecedented way. These economics enable enterprises to experiment, test, and try new methods and configurations of their existing data, which enables agility and deep customization. 

Managed services also play a vital role. A few years ago, you needed a team of data scientists to enable features, such as computer vision, natural language processing, and pattern matching. Today, any developer can add these features to their apps as easily as any code in their toolbox. Services, such as Amazon Rekognition Image and Video, enable object identification via the camera on your phone, in manufacturing, or in home Internet of Things (IoT) devices for security, customer services, and entertainment. We speak with Alexa now almost as naturally as with any person in the room. Amazon Lex, the powerful engine behind Alexa, is available as a managed service to any builder so that patients may ask for assistance or entertainment from their bedside without complication, teachers can enable a wealth of literature and instruction to their students, and many other use cases. 

A significant breakthrough in AI today is the ability to put an expert by your side at any time and to interact with them the way you would if you were physically together. This is a break from our historically push-based relationship with computers. We push keys on the keyboard, tap our screens, and push forms out of the way to get to what we want. Now, we interact naturally through language and other forms of perceptions. Managed services and platforms that enable extending deep neural networks, such as Amazon SageMaker, are opening up new forms of interacting with computers and smart devices at work, home, and where we play. 

TN: What tools does AWS make available to organizations and enterprises looking to develop AI solutions?

KS: Amazon provides a comprehensive range of tools for builders of all kinds to take advantage of the newly enabled AI opportunities. We describe the offerings in the form of a stack: a layered cake where our affordable hardware innovations on the base layer support platform tools in the middle and easy-to-use managed services on top. I’ve described the icing on the cake, the managed services. Let’s look at the platform layer. 

Amazon SageMaker is a product that provides end-to-end services for your AI and machine learning project. We took a look at the entire AI development process, from data exploration and discovery, to model development and to deployment, not only in the cloud but to phones, edge computing, and smart devices, and created a simple platform to host each of these workflows. 

SageMaker has a place for everything. Prior to SageMaker, when builders took on a machine learning project they had to marshal several resources: build an environment and create an ad-hoc digital conveyor belt so the final product could go into production. Now they can focus on business goals and objectives, discover opportunities in data, and map them to where SageMaker provides an accessible entry point. 

It starts with hosted Jupiter notebooks, a popular tool for data scientists. Data writes code and data scientists’ key role is to craft the models that make that happen. Our notebooks come with a complete set of popular tools, such as MXnet, TensorFlow, CMTK, PyTorch, and others prepackaged and ready for use. 

Because data writes code, we provided simple connectors to your data lake: S3, Spark, relational databases, such as Amazon Aurora and no-SQL data stores, which are now available serverless where you pay by the second. 

Next is the training system. Models use the data to train systems of near infinite flexibility. SageMaker reduces the cost of training large models by separating the notebook from the training systems. Training is performed by submitting a job to a queue. We provide highly efficient pre-trained algorithms to perform common modeling tasks that run up to 10 times faster than open-source options. 

None of this matters if you can’t delight your customers. To get models integrated into your mobile apps, enterprise systems, or IoT devices, SageMaker provides a set of one-click tools that remove all the heavy lifting, such as allocating compute resources and auto-scaling. If you have a significant prior investment in on-premises hardware, you may use SageMaker to deploy those models.  

So that’s SageMaker. Depending on how you count, it’s either three or four complete Machine Learning developer services. We also provide Amazon Machine Learning, a simple analysis service that takes spreadsheet-formatted data and provides regression and classification. 

We still need people, sometimes a lot of them, to help us understand our data. Mechanical Turk is a human-in-the-loop service that augments your development and data discovery process where an eye, ear, or thoughtful mind are indispensable. Our Network Partners, such as CrowdFlower, take that expertise to another level by providing deep domain expertise in almost any area including law, finance, design, and e-commerce. 

On the bottom layer, the hardware stack, we provide a powerful machine learning processor: the P3, which can provide up to 8 v100 NVIDIA Volta GPUs per instance. Amazon Greengrass now provides an inference engine that can run models in smart devices and on the edge, and the DeepLens Camera makes it easy for developers to quickly build apps that can see and respond at the edge. Using Greengrass you can keep the cloud close and the edge closer. 

TN: Do you have any examples of interesting AI tools and applications that have been built with and powered by Amazon Web Services’ cloud offerings and tools?

KS: Narrative Science, Intuit, Digital Globe, and ZipRecruiter come to mind.

TN: How can cloud-driven AI tools and solutions such as these help the federal government today? What challenges do you see the government facing that AI can help it overcome?

Public services are benefiting from these advances. The ability to see, hear, and quickly detect anomalies has advantages for security, personalization, and cost optimization for defense, social services, and forecasting. Along with the benefits they provide, we’re entering a new era of data governance, that is, who can use what data and where. There are opportunities and threats where currently the security offenders have the advantage. 

At Amazon, security is always our top priority. In addition to the multiple layers and means of security available in the cloud including ubiquitous encryption both in transit and at rest, just-in-time authorization, and authentication and clear delineation between public and private compute spaces we also provide the AWS GovCloud (US), Amazon's cloud region designed to host sensitive data, regulated workloads, and address the most stringent U.S. government security and compliance requirements. 

TN: In your opinion, what does the future of AI look like? What advanced tools and capabilities are coming down the pike and how will they disrupt the status quo?

KS: The ability to get the expert in the room, on demand, and interact naturally is a fundamental game changer. 

That expert may be a doctor, lawyer, financial advisor, or a family friend. They may know exactly where you are and give you location-specific guidance. They may spot unhealthy patterns in your finances, diet, or exercise, and provide guidance to help you sleep better at night, have more energy, and live a longer, happier life. 

High-power, low-cost facilities are now available to anybody. High-school teenagers can use SageMaker to load up their soccer team roster and game history to predict who should best play the team in this weekend’s game. We take for granted that we talk to Alexa not only with our phones but now with refrigerators and smart devices at home and work. 

At Amazon, we ask an interesting question: what won’t change? In the coming years, we’re still going to read books and listen to music, our kids will go to school and we’ll eat dinner at the end of the day. We’ll need to be safe in our communities and help others when possible. Each of these activities will leverage the advances in machine learning and the tools that enable Amazon builders to create that better world. 

Learn more about how CSRA and AWS partner to support federal customers.

Please note: The content on this page was originally posted on prior to its acquisition by General Dynamics. This content was migrated to on July 9, 2018.​