AWS has launched three new managed services to bring customers complex machine learning and AI capabilities without having to manage infrastructure or build and maintain algorithms.
During his keynote at AWS Re:Invent in Las Vegas yesterday CEO Andy Jassy announced three new AWS services for artificial intelligence (AI) and machine learning (ML), allowing customers access to image recognition (AWS Rekognition), text to speech deep learning models (Polly) and the engine that powers Alexa (Lex).
Jassy spoke about Amazon’s heritage of developing machine learning internally and releasing it to customers, mentioning its recommendation engine and Alexa assistant as examples. AWS has offered a machine learning service since April 2015, which has already been used by customers for things like fraud detection.
Customers had asked for more machine learning capabilities as a managed service according to Jassy. During his keynote he said customers asked him to “expose those services to us so that more of our developers can use those capabilities without trying to figure out how to build them ourselves". Jassy also told the audience to expect more AI services to be released next year.
Rekognition is the AWS image recognition software which uses AI to add image analysis and facial recognition to apps. This could be used for biometric approval using facial recognition, something banks have been looking at.
Polly is a text-to-voice service which uses AI to parse meaning from written text, allowing apps to speak in 47 voices in 24 languages.
The Washington Post is already eyeing Polly to deliver audio versions of stories to customers as “existing text-to-speech solutions are not cost-effective for the speech quality they offer,” according to Joseph Price, senior product manager at the paper.
Finally, Lex is an open sourcing of the core engine behind Amazon’s AI assistant Alexa which will allow developers to build conversational user experiences or chatbots into web and mobile applications, including third party apps like Facebook messenger, Twilio and Slack.
As Dr Matt Wood, general manager of product strategy at AWS put it, Lex allows developers to "build conversational, intuitive, efficient interfaces to your applications and business data".
Lex is essentially a managed service for natural language understanding (NLU) and automatic speech recognition (ASR) models. Behind the scenes it takes a written or spoken prompt, transcribes it into text and runs a natural language processing algorithm to derive the meaning. Rules can then be set up by developers to derive customer intent.
Customers can also integrate Lex with business logic triggers from AWS Lambda and enterprise data sources - including connectors to Salesforce and Microsoft Dynamics data - to drive personalised responses. For example, a pizza delivery company could use Lex so that a customer ordering a pizza through an app or Alexa could simply ask for their regular order. It will even recognise that you are at your home location to speed up fulfilment.
Pricing and availability
These AI services are generally available in the US and EU regions immediately, with Lex only in preview for the US east region from today.
He said: “With a lot of the things you can use for free there is a lot of heavy lifting, managing and building and you have to keep making the models more robust and improve them over time.
“I think if you look at services like Rekognition the amount of features available and the fact that it’s a managed service that takes away a lot of heavy lifting is going to be compelling to a lot of customers.”
Polly is free up to five million characters per month for the first 12 months and then $4 (£3.20) per million characters.
Rekognition is free to process 5,000 images per month and store up to 1,000 pieces of face metadata each month for the first 12 months. It is then priced on a sliding scale starting at $1 (£0.80) for the first million per month and coming down to $0.40 for over 100 million per month.
Lex is free for 10,000 text requests and 5,000 speech requests each month for the first year and then $4 (£3.20) for each 1,000 speech requests and $0.75 for every 1,000 text requests.