Image: Flickr

Can computers read? Sure a machine can recite a block of text in a (less and less) robotic voice and maybe even pronounce all the words right, but most can’t pick up on the meaning and feeling behind the letters and punctuation. Your personal computer can’t read through your emails and remark in Scarlett Johansson’s infectious voice that, “Aw, you seem sad today. Is everything OK?” Not yet at least.

Spike Jonze’s Her may be set in a futuristic world where you can chill with the characters in your video games and dish about kissing girls. But artificially intelligent programs that can recognize the sentiment in pieces of text already exist, and now researchers at Stanford University have made one available to anyone online.

The Stanford team, working under the supervision of AI expert Andrew Ng, released a semantic analysis tool to bring these machine-learning insights to any researchers or curious regular folk who don’t know how to code and aren’t able to program a computer. You can visit the website, etcML, and then drag and drop a news article, bit of prose, Twitter account, hashtag, etc and the algorithm will spit out a classification of the text.

At this point, the results don’t go beyond the most rudimentary level of predictive analysis: polarity. It’ll tell you if the sentiment is negative, positive, or neutral.

Classifying the sentiment of my Twitter account. Blue is positive; orange is negative. Image: etcML

But if you really want to get into it, you can create your own “training” program to teach the algorithm to seek out the specific qualities you’re looking for, like media bias or how successful a crowdfunding campaign will be. (Pro tip: Use plural pronouns in pitches—we, us, our versus, I, me, my.)

You’ve probably heard about similar programs lately; sentiment analysis is a hot field, especially with the explosive growth of Big Data. While Her-style hyper intelligent operating systems may still be a ways off, semantic classification technology already has some real-world applications.

Researchers at Georgia Tech used a predictive model to find the magic words to include in your Kickstarter pitch (references to “unveiling” the “future” were the ticket). This week Wired analyzed large datasets from OKCupid and Match.com to find certain words that make your online dating profile more compelling (both sexes want to hook up with someone who likes “surfing” and “yoga”).

Stanford researchers used the technology to help grade short-answer tests for Massive Open Online Courses. Questions with written answers are a much better way to gauge how well a student is learning the subject matter, but a MOOC professor could never read through the thousands of responses. The answer? Teach a robot to do it.

But it’s not hard to imagine the problems robots-readers could cause. What if the algorithm misinterprets a sentence and scores a student wrong, who then fails to get into college? Or politicians rig a predictive model to make a case that the news media has a political bias? Or, in the other extreme, is it possible machines will get so good at understanding human sentiment we start falling in love with our computers?