Machine Learning and the Cylons

At WWDC Apple announced an upgraded Photos app with object and facial recognition powered by something called deep learning. This is one of the hottest topics in computer science and it has major implications for developers. Deep learning, or machine learning, is the process by which computers learn to solve tasks, answer queries, identify patterns, sort and filter, without constant input from coders.

One method of machine learning is called supervised learning. In supervised learning software is fed huge quantities of labeled training data and uses that data to identify new examples. So, for example, an app like Photos might take millions of examples of boats, sort commonalities, and then correctly identify a picture of a boat you’ve taken.

While useful, this approach still requires large quantities of data and significant human input. Another approach has been adopted recently: predictive learning. This attempts to model the way humans, especially children, actually learn. The idea is, humans learn by trying things within an environment, observing the results, and drawing inferences. For instance, a machine might learn to recognize boats by observing humans referring to boats. Researchers hope this approach, as well as more sophisticated supervised learning, will yield more responsive and nuanced AI.

The applications are theoretically endless and developers are already engaged in a wide variety of fields. Chatbots are one popular application of machine learning. While they’ve existed for well over a decade- popularly since AOL’s SmarterChild- they have just recently hit their stride. No longer do these bots exist solely for basic queries. Increasingly, they are integrated into a broader matrix.

WeChat, with a huge (mainly Chinese) user base of 700 million, allows businesses to interact with customers without creating their own apps or websites, through bundled services. Facebook Messenger gives users the ability to do things like hail Ubers and pay bills in-app. With Amazon’s [Alexa](
– Machine learning) customers can easily re-order common items, or even purchase recommended items- all without any real searching. Slackbot offers developers particularly rich opportunities- it allows for the creation of sophisticated, subject-specific, bots. Taco Bell is developing a tacobot which can not only take orders but answer complicated queries, and do so in a way that seems natural. Even Microsoft is getting in the game with its recent purchase of Wand Labs. This field is called conversational commerce and it is an increasingly major focus of machine learning.

Predictive analytics has also benefited immensely from these methods. Machine learning companies like Dato are leading the way in integrating new analytics applications. So, for instance, features like sentiment analysis, which would attempt to analyze feedback to products and sort it by degree of positivity. This could presumably be turned towards identifying responses that might indicate profitable leads. Or it could predict when customers are considering competitors. This would help developers and businesses identify trends relatively inexpensively and allow them to shift focus to high potential products.

Nor are the applications wholly practical. A few weeks ago the Google Brain team debuted the first offering of Project Magenta – a project aimed at wholly machine created music. From a base of just 4-notes Google’s computer created a 90 second piano tune with a recognizable verse and bridge. And while it may have lacked an arc or resolution, it was clearly music. It’s based on the open source TensorFlow platform, which means developers can peak under the hood and engage with the process. Indeed, advancing the cause of machine learning for art and music is one of Google’s expressed purposes.

All of these advances in AI are great news for developers looking for new markets (many of these companies have API’s), but a little weird for the rest of humanity. Luckily, for those of us who worry about Skynet, or the rise of the Cylons, there’s still hope. You see, music is not the only art Google is trying to tame. In another boundary pushing project, a Google team fed a program 3000 Romance and Fantasy books in an attempt to answer one question: could a machine take a beginning and an ending, and fill in a coherent middle? The answer: not so much. The results were more insane poetry than natural dialogue. An example:

he said.
“no,” he said.
“no,” i said.
“i know,” she said.
“thank you,” she said.
“come with me,” she said.
“talk to me,” she said.
“don’t worry about it,” she said.

Ok. “i won”t worry about it,” i said.