Ken Goldberg just sent this excellent piece:
Multiplicity has More Potential Than Singularity
Stephen Hawking, Bill Gates, and Elon Musk are extremely intelligent humans. So it’s not surprising that the concerns they’ve recently raised about Artificial Intelligence (AI) surpassing human intelligence have generated widespread anxiety. But fears of the “Singularity” are distracting attention from a far more important development: Multiplicity.
“Multiplicity” describes an emerging category of systems where diverse groups of humans work together with diverse groups of machines to solve difficult problems. Multiplicity combines ideas from machine learning, the wisdom of crowds, and cloud computing. Multiplicity is not science fiction; it’s central to systems we use everday: Twitter, Salesforce, Netflix, Siri, and Uber.
Consider the Internet search problem. Given a word or two, find the relevant documents among billions and list them in order of importance. Google’s solution requires a diverse set of algorithms and computing platforms. It also requires ongoing input from a diverse group of humans, who make subtle decisions about content and links every time we post a tweet or update or create a web page. Humans also help Google’s search improve over time by providing ongoing feedback every time we click on one of the suggested links.
Google’s search engine is a Multiplicity system that requires a diverse group of machines and a diverse group of humans. Multiplicity is also essential for the movie and book recommendations provided by Netflix and Amazon, for Facebook’s News Feed, and for Apple’s Siri voice recognition system. Most of the recent progress in AI, for robot driving and “deep learning” networks for understanding images and video, can be characterized in terms of Multiplicity: rather than eliminating humans, our input and feedback will play a vital ongoing role.
Multiplicity systems are extremely complex and much more research is needed to effectively combine groups of machines, groups of humans, and groups of both. We need new statistical machine learning methods that combine input from an ensemble of algorithms. Recent results in Ensemble Learning  show that a sufficiently diverse group of algorithms, each tuned to different subspaces of inputs, will be better at classification than any single algorithm.
Although Economics, Psychology, Political Science, and Sociology study group behavior at different scales, more research is needed on how humans with complementary skills can be brought together to solve problems. It was recently shown that the diversity of a group is more important than its total IQ for collective problem-solving [2, 3]. New Research is needed on how to effectively integrate the skills of human groups with the power of cloud computing. One example is Cloud Robotics, where demonstrations from diverse groups of humans are shared over the Cloud and combined using statistical machine learning techniques such as Partially Observed Markov Decision Processes (POMDPs) to produce policies that maximize the probability of success and self-monitor to alert humans when confidence decreases . This plays an important role in Google’s approach to self-driving cars.
Multiplicity is neither science fiction nor an existential threat to humanity. It supports many of the most sophisticated and effective systems we use every day and it involves us rather than excludes us. But Multiplicity is not yet well understood. It deserves our attention.
Ken Goldberg is UC Berkeley Professor of Engineering.
Summary (and video) of presentation at the world Economic Forum in January 2015:
Short Essay in Medium.com: