Can AI Identify Our Mood Swings?

When it comes to the human mood, AI is a unique way to keep track of things. But just how accurate is a machine at learning the bizarre way human emotions work? They’re not exactly stable, and most people have mood swings at some point, which may be difficult for a machine to catch. Recent studies have been looking into just that.

Social Media’s Use in AI Learning

Social media has always been a way to collect information about people, but Facebook, in particular, is very good at gathering information that can lead to identifying psychological traits. Michael Kosinski did a study on it several years ago. His study looked at what people liked on Facebook. Using just 10 things that people liked on Facebook, the study showed that it was possible to determine some basic personality traits. These included things like openness, agreeableness, neuroticism, and conscientiousness. If they had a list of 70 likes from the person, they could determine their psychological profile better than a friend could.

You’ll likely recognize this as something that is used in advertising on social media these days. Kosinski and his team decided to run ads aimed at those particular people and their psychological type. Their ads reached over 3 million people, and the sales were much higher with targeted ads, of course. People who saw an ad tailored to their personality were 50% more likely to purchase. Likewise, game ads were 30% more likely to be downloaded when the wording was tailored.

While this proved useful for advertisers, what does it all have to do with mood? AI predictions could make ads even more targeted if they determine what your current mood is. That means it’s big business.

AI and Mood Swing Prediction

Just how accurate is AI in predicting mood swings so advertisers can take advantage of your current state of mind? It turns out, pretty accurate!

This study, run by Johannes Eichstaedt and Aaron Weidman from the University of Michigan, indicates that machine learning is really quite good at picking up on how human moods shift. To determine this, the machine analyzes Facebook posts, using natural language processing. By monitoring how people are talking on their posts, the AI system can detect if they are happy, sad, angry, or depressed.

While the Michigan study used words to determine emotions, others have tried to use AI to determine mood from facial expressions. This is expected to be more accurate because people’s faces reflect their actual emotions, rather than what they say or write, which could be a lie. Eye movement, facial expressions, twitches, and more can help the AI figure out just what someone feels. This was done by humans previously, but AI is now being tested.

In some cases, AI was rather biased, where it tended to assign negative emotions to certain races more than others. It was also difficult for the AI to determine the difference between cultural expressions, such as smiles. As time goes on, though, AI is able to learn, and that could change everything.

Applications of Mood AI Predictions

Of course, there’s the obvious use of ad targeting to reach people who are in just the right mood. After all, an ad for a comfort food or calming experience will be more likely to appeal to someone who is feeling sad or blah. Likewise, someone who is happy and extroverted will be more likely to purchase something tailored to their mood.

There are some pretty serious ethical issues with using moods to target people. It is also a potential privacy violation, though most people have agreed to let this information be collected when they sign up with any given social media site. However, it does spark fears that advertisers may be going too far.

There is another option, though. The information that is collected on how mood swings occur could be used for good. An example of this would be during a major disaster or event. What is the psychological impact on the people affected? Do they need additional help? The AI system could determine who is being psychologically impacted by lockdowns or natural disasters and ensure that they receive the opportunity for additional support.

Another good option for using this type of information is monitoring how well a patient is doing with new medications or lifestyle changes. By monitoring their mood via social media, it would become evident for a doctor to see whether the patient is doing well or needs a change in medication.

AI can also be used to determine if someone who is driving is able to effectively drive while drowsy or if they are too angry to manage distractions. This could help prevent accidents and reduce potential risks.

The Future of Mood AI

The original study was somewhat skewed since it only included a portion of the population on Facebook. It also used the user’s own determination of their mood to check against the AI’s prognosis. It’s essential to train the AI system, and that can take quite a bit of time.

Over 3,000 Facebook users were used to give the machine a head start on learning to determine human emotions. The posts were checked by humans and rated for various emotions. Then the AI was taught those things. From there, it could be used to determine what new posts were focused on and determine the current mood. Over time, it was possible to map mood changes.

Will this be a valid method of research in the future? It will likely be used at some point simply because the information is right there for anyone to use. However, the technology is still being developed, and the data is still not as accurate as it could be. With future learning, however, the AI system should be able to track humans very well and decide if they are upset or not. Just how that information is used will be up to the people.