What’s hot in AI this year? Here’s what the analysts say.
Unsupervised learning, e.g., when the machine “learns” what is a spam email without first looking at a lot of emails labeled “spam” or “not spam,” is the holy grail of the AI field according to its leading practitioners.
An interim step on the journey to unsupervised learning is a hybrid approach, with some of the data labeled, but letting the machine guess the labels for the rest of the data, using associations. Google has developed one such technique, called Graph-based learning, which uses semi-supervised learning. Using its Knowledge Graph technology, which makes relation associations between words, Google is able to leverage the associations to replace the cumbersome task of labeling all of the data. Google is already using this technology for many of its products like question answering, reminders, visual object recognition, dialogue understanding, and smart email replies. Semi-supervised learning is expected to see increasing usage for very large data sets, where data labeling is an issue, especially around vision and language.
Voice assistants for the home proliferate
VoiceLabs estimates that 33 million voice-activated devices will be installed in the U.S. by the end of 2017. Amazon (Alexa), Microsoft (Cortana), Google (Google Assistant), and Apple (Siri) are investing heavily in bringing consumers into their own devices’ ecosystem by inventing ingenious ways for lock-in. One way to win customers will be offering exclusive features or specific discounts (e.g., inclusive subscriptions to content channels for a certain time period.
Social media-based messaging services in China such as WeChat have established and popular chatbots to aid in daily tasks. Facebook is just beginning to drive integration through the use of adverts which link to chatbots, as well as sponsored adverts in Facebook Messenger. These virtual agents will grow in presence and popularity, streamlining eCommerce activities such as enabling users to book flights and hotels, or to order items directly by speaking with a bot through an app.
But they are moving rapidly from consumer applications to offering assistance business users. A survey of corporate executives found that 32% said voice recognition chatbots are the most used type of AI tech in their workplace. Gartner predicts that chatbots will power 85% of all customer service interactions by the year 2020. Slack, Skype, Oracle, Salesforce, other enterprise messaging and collaboration platforms and numerous startups offer in-house or software-as-a-service functionality, helping employees do their jobs faster and better. Like the smartphone, business users of virtual assistants will eventually want these artificial intelligences to follow them throughout the day—possibly giving rise to Bring Your Own Robot (BYOR) movement.
AI as extension of enterprise IT
The enterprise use cases that are attracting the most investment today are automated customer service agents, quality management investigation and recommendation systems, diagnosis and treatment systems, and fraud analysis and investigation. The use cases that will experience the fastest revenue growth over the next five years are public safety and emergency response, pharmaceutical research and discovery, diagnosis and treatment systems, supply and logistics, quality management investigation and recommendation systems, and fleet management. The ability to recognize and respond to data flows using algorithms and rule-based logic enables AI applications to automate a broad range of functions across many industries and augment the work employees, making them more productive.
Self-driving grows up
According to McKinsey, self-driving cars will save an estimated 300,000 lives per decade by reducing fatal traffic accidents. This is expected to save $190 billion in annual critical care and triage costs. With Google alone racking up over 1 million miles testing autonomous vehicles, focus will shift from potential benefits to the necessary regulation. Legislators and policymaker will start the long process of designing and implementing the new legislation. 2020 could be the first year to see a marketable autonomous vehicle and society must begin to prepare for that day. We will see more lobbying groups in Washington DC and more vendor and user coalitions forming, to prepare the ground for widespread use.
The many faces and uses of hardware
Alternative hardware platforms such as field-programmable gate arrays (FPGAs), application specific integrated circuits (ASICs), and specialized processor architectures will increasingly compete for attention and investment dollars with Graphics processing units (GPUs), which have been the dominant hardware platform for AI applications, specifically deep learning systems. As AI algorithms change to account for applications like autonomous driving or personalized medicine with dynamic inputs, there is a case for having memory storage on the processor itself. The evolving nature of algorithms and workloads will determine what architecture is best suited for which application.
The emergence of the AI services market
As happened recently with big data and data science, there is emerging opportunity for services related to AI, including vendor selection, implementation, training, application and algorithm development and integration, and consulting. As the skills and experience related to machine learning and AI are in short supply, we will see expansion of the on-demand services provided by cloud vendors.