February 23, 2016
From: Reddit AMA
What do you think of the current “deep learning” methods? Will that fit into Wolfram software?
Yes, we’ve done a lot with these things, and will be doing a lot more. See e.g. https://www.imageidentify.com that we released a year ago. We’ve also got a lot of machine learning built directly into the Wolfram Language (and we use machine learning to automate it, so you don’t need machine-learning experts to use it).
For me it’s been really interesting to see the trajectory of neural networks. I worked on them back around 1980, and I really couldn’t get them to do anything interesting. I wasn’t sure they would ever do interesting things. But it turns out that—using essentially just the kinds of things I was doing back in 1980—we finally can make them do interesting things. It’s all a very interesting episode in the history of science and technology. (See e.g. https://writings.stephenwolfram.com/2015/05/wolfram-language-artificial-intelligence-the-image-identification-project/)
And being able to have this kind of functionality alongside all our precise symbolic computational capabilities in the Wolfram Language is extremely powerful—and I expect lots more from it in the future.