EMBEDDING OPPORTUNITY IN ORGANIZATIONS

EMBEDDING OPPORTUNITY IN ORGANIZATIONS

As companies have reached further than ever into people’s lives, they’ve shaped society around their products and services. This transformed society now provides the new foundation for each company’s future growth.

Through new partnerships with customers, employees, partners, and even governments, companies are empowered to build ever-stronger access and trust. This trust will give companies the inroads to further embed themselves into society, becoming ever more indispensable—and empowering their own revolutionary growth. (See Figure 1.)

Driving Learning Through Data Iterations

AI systems are finding ever-wider application across the enterprise as they grow in sophistication. No matter the particular type of AI being used, however, every application begins with large amounts of training data. Imagine a machine learning system designed to find a dog in a picture and decipher the breed. Tens of thousands of “labeled” images are needed: one set will teach the system to pick out dogs in a picture, while other sets of images will distinguish individual breeds. In supervised learning, images are hand-tagged to tell the system not only where the pet is in the image, but also the breed of the pet.

Modern advancements in parallel processing (see Internet of Thinking, Trend 5) and AI algorithms have unlocked the potential of deep neural networks. Inspired by the myriad neural connections of the brain, deep neural networks can learn enormous stores of data, even if it’s “noisy.” As part of their learning process, these algorithms teach themselves new ways of connecting data—meaning deep neural network AIs can continually scale and improve their capabilities. Yet another advance is in reinforcement learning, where the AI becomes its own teacher, with no need for human supervision. DeepMind’s AlphaGo Zero AI taught itself the game of Go without knowing any of the rules beforehand. In a matter of days, AlphaGo Zero had become the world’s best Go player, beating one of its own AI predecessors 100 games to none—the same predecessor that had previously beat the world’s most formidable human Go player.7 The more data an AI is given, the better its predictions become. Learning-based AIs use the data to build a model, which is then checked against test data for success across a variety of factors. In the pet and breed identification example above, a test data set could include an image of multiple pets against a complex or “noisy” background. When a model achieves a desired level of accuracy, it can be used in a production environment. (See Figure 3.)