Computers already can do some pretty crazy (and increasingly human) things: they can understand language, recognize images, even write poetry. It's no wonder some people worry about science fiction becoming reality, with computers replacing the role of humans in the economy--or worse, just plain wiping us out.
But what's really going on with computers and their cognitive capabilities? How are they becoming smarter? And who's leading the charge?
Even for the layperson with a passing interest in technology, these are important questions to ask. After all, if HAL (or some 21st century version of him) exists somewhere, you'd want to know how he works, right? Computers are only going to get smarter, so now's as good a time as any to get clear on where technology is headed.
In two words, the future is "machine learning." Here's a primer on what it is and why you should care about it.
How does machine learning work?
Currently, computers are better at solving some problems than others as the video below created by a group of Google (now Alphabet) employees explains. For example, mapping the trajectory of an asteroid is a fairly straight-forward task. Identifying that a stop sign partially obscured by snow is in fact a stop sign? That's not so simple. Machine learning helps solve that kind of problem by teaching the computer through examples how to recognize such fine distinctions in a world that doesn't always operate according to strict rules. It's what allows Facebook to recognize your face in photos or what lets an ATM recognize read your messy handwriting on a check.
The video focuses on a category of machine learning called "artificial neural networks." Basically, mathematical functions in an algorithm act a little bit like neurons in the human brain, sending signals to each other. Those functions work together to ultimately identify an image, word, sound, and so on.
What role does machine learning play in the tools that use it?
If you use a motor vehicle as a metaphor for an app, machine learning occupies the role of the chemistry of combustion, says IBM Watson CTO Rob High. A car with a combustion engine won't run if there's no combustion (duh), but combustion alone is insufficient for the car to fulfill its purpose as a mode of transportation. You need the rest of the engine and doors and a steering wheel and all of the other car parts to make the car a car. It's the same idea with machine learning. "Machine learning is not the only science that we leverage in these cognitive systems, but it's important in the engineering," says High.
A machine learning framework is embedded deep within an application, communicating with other features of the app as well as with the interface that you use. In other words, you come in contact with this technology far more often than you probably realize.
Who's leading the charge behind machine learning?
You've probably seen two companies associated with machine learning: Google and IBM. The latter has devoted an entire division (Watson) and at least $1 billion to this area. What's interesting about the development of this technology, however, is that much of it is open-source. IBM Watson's suite of tools for functions such as natural language processing are made freely available to developers, and IBM itself makes use of open-source machine learning frameworks. Just last week, Google made its own framework, called TensorFlow, open-source.
Why would these companies open up their individual advancements to developers everywhere? High says doing so will strengthen and improve machine learning technology more generally. There's another reason, though, that isn't purely altruistic. The release of TensorFlow is expected to allow outsiders--primarily machine learning researchers and engineers, to improve on products such as search and Google photos, and share those improvements with Google, explains spokesman Jason Freidenfelds.