Co-Intelligence

Talking about AI can be confusing, in part because AI has meant so many different things and they all tend to get muddled together. Siri telling you a joke on command, the Terminator crushing a skull, algorithms predicting credit scores. We've long had a fascination with machines that can think. In 1770, the invention of the first mechanical chess computer stunned those who saw it. A chessboard set upon an elaborate cabinet with chess pieces manipulated by a robot dressed like an Ottoman wizard. It toured the world from 1770 to 1838. The machine, also known as the Mechanical Turk, beat Ben Franklin and Napoleon in chess matches and led Edgar Allan Poe to speculate on the possibility of artificial intelligence upon seeing it in the 1830s. It was all a lie, of course. The machine cleverly hid a real chess master inside its fake gears. But our ability to believe that machines might be able to think fooled many of the best minds in the world for three quarters of a century. Fast forward to 1950, when a toy and a thought experiment, each developed by a different genius of the still-developing field of computer science, led to a new conception of artificial intelligence. The toy was a jury-ranked mechanical mouse called Theseus, developed by Claude Shannon, an inventor, prankster, and the greatest information theorist of the 20th century. In a 1950 film, he revealed that Theseus, powered by repurposed telephone switches, could navigate through a complex maze, the first real example of machine learning. The thought experiment was the imitation game, where computer pioneer Alan Turing first laid out the theories about how a machine could develop a level of functionality sufficient to mimic a person. While computers were a very new invention, Turing's influential paper helped kick off the nascent field of artificial intelligence. Theories alone were not enough, and a handful of early computer scientists started working on programs that pushed the boundaries of what was soon called artificial intelligence, a term invented in 1956 by John McCarthy of MIT. Progress was initially rapid as computers were programmed to solve logic problems and play checkers. Leading researchers expected an AI to beat grandmasters in chess within a decade. But hype cycles have always plagued AI, and as these promises went unfulfilled, disillusionment set in, one of many AI winters, in which AI progress stalls and funding dries up. Other boom-and-bust cycles followed, each boom accompanied by major technological advances such as artificial neural networks that mimic the human brain, followed by collapses AI could not deliver on its expected goals. The latest AI boom started in the 2010s with the promise of using machine learning technologies for data analysis and prediction. Many of these applications use a technique called supervised learning, which means these forms of AI needed labeled data to learn from. Labeled data is data that has been annotated with the correct answers or outputs for a given task. For example, if you want to train an AI system to recognize faces, you need to provide it with images of the faces that have been labeled with the names and identities of people in them. This phase of AI was the domain of large organizations that had vast amounts of data. They used these tools as powerful prediction systems, whether optimizing shipping logistics or guessing what kind of content to show you based on your browsing history. You may have heard the buzzwords big data or algorithmic decision-making describing these kinds of uses. Consumers mostly saw the benefits of machine learning when these techniques were integrated into tools such as voice recognition systems or translation apps. AI was a poor, albeit marketing-friendly, label for what this sort of software did, since there was very little about these systems that actually seemed intelligent or clever, at least in the ways humans are intelligent and clever. To see one example of how this sort of AI works, picture a hotel attempting to forecast its demand for the upcoming year, armed with nothing but existing data and a simple Excel spreadsheet. Before predictive AI, hotel owners would often be left playing a guessing game, trying to predict demand while grappling with inefficiencies and wasted resources. With this form of AI, they could instead input a wealth of data, weather patterns, local events, and competitor pricing, and generate far more accurate predictions. The results were a more efficient operation and, ultimately, a more profitable business. Before machine learning and natural language processing became mainstream, organizations focused on being correct on average, a rather rudimentary approach by today's standards. With the introduction of AI algorithms, the focus shifted to statistical analysis and minimizing variance. Instead of being right on average, they could be right for each specific instance, leading to more accurate predictions that revolutionized many back-office functions, from managing customer service to helping run supply chains.