Microsoft's new emotion-detecting AI can be hilariously bad at detecting emotions

Productivity killer of the week: A free Microsoft tool that can guess what you're feeling, but not very well

Productivity killer of the week: A free Microsoft tool that can guess what you're feeling, but not very well

Microsoft's new AI tool analyzes pictures of people to reveal their emotions on a scale from zero to one – but it seems as though humans are still a lot better at reading humans than machines are. ( Beach Police Department)

First, they taught machines to figure out how old we are. Then, how closely related two people might be.

Now, Microsoft has launched an algorithm designed to recognize how humans feel, simply by reading our faces – and anyone who's not freaked out by it can try the beta version for themselves right now.

The team behind Microsoft's Project Oxford, a set of advanced machine learning technologies for developers (or "intelligent services"), announced the public beta release of its new Emotion API Wednesday morning.

"Humans have traditionally been very good at recognizing emotions on people's faces, but computers? Not so much," reads a blog post published to Microsoft's website today. "That is, until now."

The post's author goes on to explain that this emotion tool can currently be used to help computers identify eight core emotional states in photos of human faces: Anger, contempt, fear, disgust, happiness, neutral, sadness and surprise.

While Microsoft does not reveal which pictures (or how many of them) were used to "train" the system, the emotions it can recognize are said to be based on facial expressions that are "cross-culturally and universally communicated."

Like most artificial intelligence systems, the Emotion API was designed to become smarter as it receives more data.

That's where you come in.

Anyone with an internet connection can currently access the beta version of the emotion tool on Microsoft's website and upload a photo to test it out.

The computer will analyze a picture within seconds to reveal how much of each specific emotion the subject is expressing on a scale from zero to one.

According to Microsoft's Emotion API, Sidney Crosby was rather angry about scoring the goal that won Canada a gold medal at 2010's Olympics in Vancouver. (CBC News/ Kadobnov/Getty Images)

A score of, say, 0.900261164 in the "anger" category – as shown on the above image of NHL star Sidney Crosby – would mean that the person photographed is roughly 90 per cent angry.

Some of the images we tested pulled a pretty accurate reading, but others were way off.

That photo of Crosby, for example, was taken directly after he scored the gold medal-winning goal in overtime at the 2010 Vancouver Olympics. 

Was he angry? Possibly, but it stands to reason that he was probably more joyful than anything when that picture was taken.

Many of those who were playing with the tool today shared similarly wonky results on Twitter.

Conversely, others seemed pleased with the results they got.

One clever sleuth even used the tool to find out what Mona Lisa was hiding behind her famous smile.

Neutral happiness. Thanks, computers.


To encourage thoughtful and respectful conversations, first and last names will appear with each submission to CBC/Radio-Canada's online communities (except in children and youth-oriented communities). Pseudonyms will no longer be permitted.

By submitting a comment, you accept that CBC has the right to reproduce and publish that comment in whole or in part, in any manner CBC chooses. Please note that CBC does not endorse the opinions expressed in comments. Comments on this story are moderated according to our Submission Guidelines. Comments are welcome while open. We reserve the right to close comments at any time.

Become a CBC Member

Join the conversation  Create account

Already have an account?