The Innovation Ultimatum. Steve Brown
of the most important things humanity is working on. It is more profound than, I dunno, electricity or fire.” While hyperbolic, this statement from the head of one of the world's most powerful tech companies should make us all sit up and listen.
AI's hype is largely justified. Just as digital technology was a vital component of any successful business strategy in the 1990s and 2000s, so AI must be central to strategic plans of the 2020s.
The Next Era of Computing: Traditional Digital versus Artificial Intelligence
Artificial intelligence and traditional digital computers are complementary. Importantly, AI solves problems that are costly or impossible for traditional computers to solve. The two technologies will coexist and work side-by-side, each solving a different set of problems. For a language translation app on a smartphone, traditional computing presents an attractive interface, while AI handles voice recognition and language translation functions.
AI solves problems using a radically different approach from traditional computers. Traditional computers are programmed to solve problems. Programs apply a set of rules to data and compute results. Said another way, they take rules and data as input, and output results. AI solves problems without using preprogrammed rules. Machine learning, a popular form of AI, takes data and results as its input and infers rules as its output. Through a complex training process, AI finds patterns and associations between data and results and divines its own rules for how they connect. This trait lets us solve entirely new problems with AI. It's why AI can seem so magical: It solves problems we don't know how to solve ourselves.
To find complex associations in data and build rules, AIs must be trained with thousands or sometimes millions of examples. Today's AIs are not like human brains. While some of the organizing principles are the same, human children learning about the world don't have to see a million automobiles before they can recognize one and adorably yell the word “car.”
A 1950s Concept and 1980s Algorithms Meet Modern Computing Horsepower
The term “artificial intelligence” was first coined in the 1950s. The core algorithms behind today's AIs were first proposed in the 1970s and popularized in the mid-1980s. But it was 2012 before the recent crop of AI breakthroughs began to appear. Why the quarter-century delay? Older computers lacked the performance to run AI applications. High-end graphics processors, GPUs from companies like Nvidia, eventually provided the computing horsepower needed. Their parallel number-crunching architectures, designed to create realistic video games, turn out to be pretty good for training an AI. As well as fast computers, AIs need training data to learn from. As digital storage costs fell and broadband speeds increased, data flooded in from many sources: billions of industrial sensors, millions of smart cameras, billions of people sharing trillions of photos and billions of videos, and trillions of clicks on social media. Users upload 500 hours of video to YouTube every minute and more than 1.2 billion photos to Google Photos every day (Source: Wikipedia).
With cheap, powerful computing, an avalanche of training data, and a small army of AI-savvy researchers and developers, artificial intelligence is now poised to solve myriad problems and create many exciting new capabilities.
What Can You Do with It?
Artificial intelligence can solve a wide variety of problems. Considering all the possible applications of AI can be overwhelming. I've found it helpful to cluster AI applications into eight broad categories:
1 Machine vision
2 Natural language processing (NLP) and voice platforms
3 Exploration and discovery
4 Better-informed decision-making
5 Predicting the future
6 Seeing the world through a new lens with super sensors
7 Solving complex problems by learning from experience
8 Creating and co-creating content
In all of these eight application categories, AI is used to find patterns and associations in data and to make statistical predictions. Each application uses this fundamental feature in different ways. Apply the associative capabilities of machine learning to images and you get machine vision; apply it to historical data and you get predictions; apply it to cursive text and you get handwriting recognition. In voice platforms, AIs trained on human speech determine what words you are saying. AIs trained on historical weather data are used to make predictions that inform the weather forecast.
Artificial intelligence finds important associations that we may not previously have discovered: the complex association between the molecular structure of a chemical compound and its physical properties, or the complex set of circumstances that lead to an outbreak of disease. This characteristic enables AI to solve problems that we don't yet know how to solve ourselves.
As we review each of the eight main applications of AI, think about how each one might impact your business, your life, and society at large.
Machine Vision: Computers Open Their Eyes
With the advent of artificial intelligence, machines have evolved eyes and ears. Computers can now see, hear, and “understand” something about the world that they inhabit. This understanding is still rudimentary. Computers can recognize an image of an apple and correctly categorize it with the five letters a-p-p-l-e, but they don't understand what an apple is, that it grew on a tree, or what it tastes like.
Machine vision has many interesting applications across a wide range of industries. Stocktaking robots audit shelf contents in grocery stores. Facial recognition algorithms turn faces into passwords. Agricultural robots spot-spray herbicide on weeds. Quality assurance AIs perform visual inspection on the manufacturing line. Autonomous machines—robots, drones, and self-driving vehicles—all rely on machine vision, too.
EarthNow, an ambitious startup funded by Bill Gates, Airbus, Softbank, and others, is a splendid example of machine vision's future potential. EarthNow will operate a constellation of satellites, each containing four powerful, high-definition cameras. The company's goal is essentially to create a real-time version of Google Earth, though with an important twist: Artificial intelligence, built into the satellites, runs applications to interpret camera images and add intelligent insight. These applications will reveal important details about activity on our planet.
A lightning strike in a remote location can start a devastating forest fire that rages out of control. Detection, rapid response, and early containment may save millions of acres of land from incineration, prevent damage to structures, and save lives. With EarthNow, global fire detection is just another application that runs on the satellites. The network becomes an eye in the sky that watches for fire starts and automatically alerts local authorities, 24/7/365.
EarthNow has proposed a range of other exciting applications for their satellite constellation. The system could provide real-time traffic information to city managers, real-time crop health information to farmers, and alert law enforcement or government agencies to illegal fishing, mining, and logging activity. Marine biologists will be able to track whale migration and volcanologists will monitor volcanic activity. Global asset tracking applications include tracking ships at sea, trucks on the road, planes in the air, and shipping containers in transit. Other applications include improved weather forecasting, law enforcement, and news coverage. EarthNow is a powerful platform that raises serious privacy concerns. This is why EarthNow chose to build machine learning capabilities into the satellites themselves. Users have very limited access to data from applications—real-time traffic data, the GPS locations of whales—but not the image data itself. The machine vision capabilities of the EarthNow platform present an exciting new set of opportunities for research scientists, public safety professionals, local governments, and a wide range of businesses.
Natural Language Processing and Voice Platforms
Human