For context let me take you back to the early 70's, a time when Linotypes still cast hot metal during newspaper production. I had a relatively recent degree in computer science from the University of Delaware. CS then had three legs: automata theory and formal languages with compiler writing, artificial intelligence (AI) and numerical methods. I brought this with me to the ANPA/RI, the newspaper trade association now called the NAA. The ANPA/RI wanted to paginate newspapers digitally. I designed a laser-based device and wrote software that showed how digital pagination should work. (I still have sample pages printed in 1974.) The ANPA management thought computers were fast, but they had no idea (until I told them) of the computational load entailed in producing pages digitally. The solution was good in theory, but no good in practice. A PDP 11 couldn't put out the bits to dot pages quickly enough. The patent was sold and I was asked to think about what's next.
The answer didn't require a lot of thought. To make pages electronically one needed electronic designs, i.e., digital dummies or ad layout geometries. Without these, manual dummying would necessitate a very inefficient workflow. Thus were the origins of the Layout series of programs established. The project fit nicely into my skill set. AI was my favorite academic area. Display ad dummying was another rule-based search problem, just like the chess programs I wrote as a post-graduate.
Back to last week's world changing competition: Lee Sedol vs. AlphaGo for the world title in Go. This was not just another man vs. computer competition where superior calculating ability wins. Go isn't like chess. Anyone who knows Go will tell you that it is a game of supreme intuition and creativity. In Go the possibilities seem unfathomable and far beyond those of other well-known perfect-information board games. Immense complexity arises from Go's quite simple rules.
Google's AlphaGo from its DeepMind subsidiary beat Go maste and world title holder Lee Sedol decisively 4 to 1 in the five game match. Throughout Asia tens of millions watched the match on TV and the Internet. Read more here about how AlphaGo’s play reached a level “close to the territory of divinity” according to the Korea Baduk Association.
So what does this have to do with newspaper productivity growth? AlphaGo's programming was unlike others. AlphaGo used neural networks and thousands of master games to learn the subtle patterns of master level play. Soon its best opponent was a prior version of itself. The new versions producing improved results were selected over their predecessors.
If you think about it, you should realize that many analytic tasks, even those requiring creativity and intuition, are now within the grasp of machine learning. Will machines diagnose disease better than doctors? Will human financial advisers be bested by algorithms? Will the best composers be software? Which jobs will future robots replace?
Mark my words: last week we witnessed a modern John Henry moment.
And, by the way, ever since version 14, Layout-8000™ has gathered data on how its operators use it for both automated and manual dummying. Hiding behind the name LayoutHistoryAdBoss is a pattern database and learning algorithm designed to enable further efficiencies. It encapsulates human dummying expertise. It is being deployed at the design centers of large newspaper groups. With LHAB one gets hints as to where advertisers wouldn't want their ads. What's left is good enough and, through learning, Layout-8000 becomes even better, smarter and more automated.
Seems like there will soon be a machine intelligence revolution that will rival the industrial revolution.