We use cookies to make interactions with our websites and services easy and meaningful, to better understand how they are used and to tailor advertising. You can read more and make your cookie choices here. By continuing to use this site you are giving us your consent to do this.

Remember toasters? Of course you do. Back before there were popular toastsharing apps like “Crusti” and hook-up sites for bakers like “Hotbred,” people used to have machines in their kitchens that made bread into toast. It fulfilled the function of an appliance, which was to make some aspect of life easier. No more people standing over slices of bread with a blowtorch; this contraption did it for us. Presumably, millions of person-hours were saved by letting the toaster do this crucial work.

Eventually computers came along and acted as larger-scale appliances for all the various toasts of our lives. Computers took the place of typewriters and post offices and theaters that showed adult movies. Now, with artificial intelligence, we have one appliance to rule them all. With machine learning, AI will integrate into all our devices and fulfill that great duty of an appliance: making life easier.

Machine learning is what we should’ve been doing this whole time with computers, really. Instead of programming the computer with every rule and command and response, we give it a goal and let it figure it out based on all the data we push into it. What we’ve been doing as humans, more or less, bumbling our way to some conclusion or other.

What’s exciting and a little bit magic-seeming about it is the way the system will exhibit behaviors not explicitly programmed into it. This is not in itself remarkable; children have been saying the damnedest things to parents for millennia. What’s unexpected is the way in which the hidden is revealed, whether that be predictions about what you might be interested in based on your web searches or the psychedelic ways data can be represented.

“Deep learning” is a branch of machine learning that illustrates just how far removed the human can be from the computation process. “The beauty of deep learning, even for some complex patterns, is that it actually learns not just the final output, but also a lot of the intermediate steps that once required humans and experts to design,” says Richard Socher, Chief Scientist at Salesforce. Socher is the founder of AI company MetaMind and a man who is giving data something to do beyond just sitting there and accruing endlessly.

Socher is saying that machine learning is the AI equivalent of the old cliché about giving a man a fish versus teaching him how to fish. The man gets to eat a lot more fish, and the deep learning system gets to serve more functions than just the one it was programmed for.

It does this by building up its own rules from a granular level of data. “Deep learning tries to get the data in as raw a format as possible: just the pixels in the case of image recognition, or a sequence of words in the case of language,” he says. “When you ask a machine learning model to come up with everything from scratch, you don’t teach it about the domain; you feed it data for it to figure out what that domain is all about.”

That “domain” could be the vocabulary of the Spanish language or images of toast. The machine would learn to abstract from many examples — to create, in its programming, firm but shifting concepts. A blueprint of the world, in other words. That’s what language does as well, which is why Socher, like a lot of others, is working on teaching machines to teach themselves to recognize and use it. From the ground up, of course.

Next Time: You Sound So Natural!