This content is part of the Conference Coverage: Your guide to AWS re:Invent 2017 news and analysis

AWS AI services toss machine learning keys to dev teams

AWS re:Invent featured updates to database and internet of things technologies, but new cloud AI services such as SageMaker broaden the appeal of the cloud provider's AI tool set.

AWS hopes a sprinkling of artificial intelligence tools and services will whet the appetites of AWS users who aren't...

AI experts, while also keep pace with aggressive cloud competitors.

AWS unveiled five artificial intelligence (AI) services for its public cloud platform at AWS re:Invent 2017. These AWS AI services include a trio of deep learning services to perform language tasks, video capabilities for existing services and even a video recorded with onboard compute.

Arguably the AI highlight was Amazon SageMaker, a managed service with 10 popular machine learning algorithms to guide developers and data scientists inexperienced with AI technology. The service uses open source Jupyter notebooks to visualize Simple Storage Service (S3) data, and it can run frameworks such as TensorFlow and Apache MXNet.

SageMaker evolves Amazon's Machine Learning service for IT professionals to build, manipulate and train machine learning models, said Karl Freund, senior analyst at Moor Insights & Strategy.

"Before they had point product solutions; now they have a platform," he said.

Intuit, a business and financial software company, has more than 40 AI and machine learning-based products or features in production, including in its popular TurboTax and QuickBooks software. The company plans to incorporate SageMaker into its workloads and, later, its software, in combination with in-house expertise in AI and machine learning, said Ashok Srivastava, senior VP and chief data officer at Intuit.

Intuit's AI strategy is to roll AI into its product lines to beef up security and fraud protection, as well as enhance customer care with AI and computation linguistics to answer tax questions, product personalization and input error detection. Intuit will roll SageMaker into those initiatives with an in-house staff that has developed about 150 pending patent applications.

"We've been pursuing AI and machine learning for the last 10 years," he said. "[We see the] cloud platform as another stepping stone in the direction that we're headed."

Butting heads in the AI market

Amazon has its share of enterprise customers using AWS AI services, such as Liberty Mutual and Capital One, but its bread is buttered by the startup community, which have fewer barriers to AI adoption and experimentation.

"The enterprises are kind of slow to adapt, because they're big and have to think things through," Freund said. Amazon Lex made some inroads with larger businesses, he said, and enterprise developers build skills and functionality for consumers on the Alexa platform. AWS' breadth of services can keep costs down, and that's been the cloud provider's biggest selling point to businesses of all sizes. "Amazon's strategy has been, and I believe still is, to exploit their incredible scale and compute to offer very low-cost production platforms in AWS," he said.

But other cloud providers have entered the AI race, notably Microsoft, which has infused AI technology into products. That provides a challenge for Amazon to not only win over customers inclined to remain on the Azure platform but to influence enterprises that are locked into licensing contracts.

"Amazon is focused on providing the service-level [tools], but I don't see them in the product game and I don't see them as doing as much with AI themselves," said Adrian Bowles, analyst at Aragon Research. "Amazon is, in so many ways, like a big box store. They've got some of everything, but if you have specialty needs, maybe not."

Srivastava said he'd like to see more enhancements to AWS AI for numerical and text predictions, anomaly detection and distributed optimization, but Intuit doesn't plan to move to another cloud provider. "The scale with which [AWS is] operating and the vision that they have really match the roadmaps that we are interested in pursuing," he said.

Let's go to the video

Other AWS AI services focus on video and image technology seeks to attract a significant slice of customers. Rekognition Video, a service that detects and categorizes thousands of objects and faces in a video, is an offshoot of Rekognition Image, which detects and evaluates images.

I can't sell a $30,000 application if we're only going to help five cases every six months, but I can sell [using an AWS service for] $6 a month.
Chris Adzimasenior information systems analyst, Washington County Sheriff's Office

The Washington County Sheriff's Office, just west of Portland, uses Amazon Rekognition Image to cross-reference surveillance or eye-witness images of unknown suspects in crimes with roughly 300,000 mug shots to identify persons of interest. The effort provides leads for detectives to pursue, and it has aided in five cases since starting the project less than a year ago.

"That may sound like a small number, but when you're talking about five cases where we didn't have a deputy or detective searching through pictures manually, it's a great time savings," said Chris Adzima, senior information systems analyst for the sheriff's office.

While the county hosts the majority of its workloads on premises, Adzima's team uses Rekognition Image along with S3 to store images and Elastic Compute Cloud to power the project, which he estimates saves taxpayers thousands of dollars compared to pricier image recognition software and hardware.

"Some of the services on AWS, we're talking about $6 a month," he said. "I can't sell a $30,000 application if we're only going to help five cases every six months, but I can sell $6 a month."

Adzima plans to use Rekognition Video to analyze all faces in each frame of a video, rather than rely on a single blurry image from a surveillance camera, to determine the sharpest image and more precisely identify facial features. He's also considering Kinesis Video Streams, another service introduced at re:Invent which streams video content from devices to the cloud for further processing.

And he sees potential in DeepLens, a video camera with built-in compute for deep learning, which could identify suspects by their scars, marks and tattoos. Adzima also plans to experiment with a SageMaker model deployed on DeepLens that can determine the intent of a suspect in real time, to help deputies respond faster and appropriately, such as de-escalate a situation if a person might become hostile.

But he sees limitations in Rekognition Video, such as the ability to train models himself and the speed of the video service -- seconds can be an eternity for vulnerable deputies. "It's good and speedy, but it's not going to be something that I utilize as I'd like to," he said.

Speaking my language

AWS also added three language processing services with machine learning capabilities. Transcribe provides long-form speech recognition to analyze files and return their content as text, log files or subtitles. Translate uses machine learning and deep learning models to automate human language translation. And Comprehend provides natural language processing to extract entities, key phrases, language and sentiment from a document.

These AWS AI services edge Amazon closer to feature parity with Microsoft, which has an API for each of those technologies, Freund said.

Amazon also added a machine learning capability, ML Inference, to its Greengrass service to enable machine learning capabilities on internet-connected devices.

But as with any high-level service, buyers must beware of lock-in. Amazon and Microsoft are working in good faith on an open source deep learning interface called Gluon; neither side is likely to slide the door open for customers to easily depart. You'll still be able to use a framework like Google's TensorFlow, which has developed a wide following, but the more you build on AWS, the deeper the roots go. That could require you to write an AI application from scratch if you move from one provider to another.

"Amazon and Microsoft both recognize that the frameworks themselves can't be a lock-in," Freund said. "However, the APIs, whether they're offered by Microsoft or Amazon, are clearly sticky."

David Carty is the site editor for SearchAWS. Contact him at [email protected].

Dig Deeper on AWS machine learning