2017 laid the foundation for faster, smarter AI in 2018

2017 laid the foundation for faster, smarter AI in 2018
From Engadget - December 22, 2017

This might be a helpful time to clarify that AI is often a catch-all term for an assortment of different technologies. There's artificial intelligence in our digital assistants like Siri, Alexa, Cortana and the Google Assistant. You will find artificial intelligence in software like Facebook's Messenger chatbots and Gmail's auto-replies. It's defined as "intelligence displayed by machines" but also refers to situations when computers do things without human instructions. Then there's machine-learning, which is when computers teach themselves how to perform tasks that humans do. For example, recently, an MIT face-recognition system learned how to identify people the same way humans do without any help from its creators.

It's important not to confuse these ideas -- machine-learning is a subset of artificial intelligence. Let's use the term machine learning when we are talking specifically about concepts like neural networks and models like Google's TensorFlow library, and AI to refer to the bots, devices and software that perform tasks they have learned.

Still with me? Good. This year, AI got so smart that computers beat humans at Poker and Go, earned a perfect Ms. Pac Man score and even kept up with veteran Super Smash Bros. players. People started using AI in medicine to predict diseases and other medical conditions, as well as spot suicidal users on social networks. AI also began to compose music and write movie scripts.

Everywhere you look, there's someone trying to add AI to something. And it's all facilitated by neural networks that Google, Microsoft and their peers continued to invest in this year, acquiring AI startups and launching or expanding AI divisions. Machine-learning has progressed quickly, and it's going to continue improving next year.

One of the biggest developments as we head into 2018 is the shift from running machine-learning models in the cloud to your phone. This year, Google, Facebook and Apple launched mobile versions of their machine-learning frameworks, letting developers speed up AI-based tasks in their apps. Chip makers also rushed to design mobile processors for machine learning. Huawei, Apple and Qualcomm all tuned their latest chipsets this year to better manage AI-related workloads by offering dedicated "neural" cores. But barring a few examples like Face ID on the iPhone X and Microsoft Translator on the Huawei Mate 10 Pro, we have not yet seen concrete examples of the benefits of chips tuned for AI.

Basically, AI has been improving for years, but it's mostly been cloud-based. Take an image-recognition system, for example. At first, it might be able to distinguish between men and women who look drastically different. But as the program continues training on more pictures in the cloud, it can get better at telling individuals apart, and those improvements get sent to your phone. In 2018, we are poised to put true AI processing in our pockets. Being able to execute models on mobile devices not only makes AI faster, it also stores the data on your phone instead of sending it to the cloud, which is better for your privacy.

It's clear the industry is laying the groundwork to make our smartphones and other devices capable of learning on their own to improve things like translations, image-recognition and provide even greater personalization. But as the available hardware gets better at handling machine-learning computations, developers are still trying to find the best ways to add AI to their apps. No one in the industry really knows yet what the killer use case will be.

Eventually, every industry and every aspect of our lives -- from shopping in a mall to riding a self-driving car -- will be transformed through AI. Stores will know our tastes, sizes and habits and use that information to serve us deals or show us where to find what we might be looking for. When you walk in, the retailer will know (either by recognizing your face or your phone) who you are, what you have bought in the past, what your allergies are, whether you have recently been to a doctor and what your favorite color is. The system's AI will learn what you tend to buy at specific times of the year and recommend similar or competing products to you, showing the information on store displays or tablets on shelves.


Continue reading at Engadget »