For a couple of years, I worked in an AI lab to build an application that uses AI to help us become happier. So people often ask me: how does it work?
Awareness
We assume people are happier if they lead a values-based life. The problem is, most people are not aware of their values. So we built a mobile-based journaling app, where people write down their happy moments daily. Over time, patterns of values emerge from their entries.
“That’s old school!” You might think. But we use AI to read your entries, recognize your values, and surface them to you to make you more self-aware.
How does the AI work
Now you might be interested: how does AI do it? We used a technology called natural language processing (NLP). NLP deals with how computers understand the human language.
Here is the basic idea. If you give the computer enough examples of what you’re looking for, over time it learns the rules. Then the computer can make decisions for that specific problem on their own.
For example, we give the computer 100 short sentences of the value “Family” and 100 short sentences of the value “Achievement”. An algorithm processes these sentence-value pairs and builds a model, which is how the machine keeps its knowledge. Next time, when we give the computer a sentence: “Hanging out with my mom beats going to the Warrior’s game”, and asks the computer what value could it be. The computer names “Family” as the most likely value.
The problem is, where do we get these sentence-value pairs? This is the hard part. We built a dataset. We asked 34,000 people to each write down three moments that made them happy in the last 24 hours. We collected 100,000 happy moments from the crowd. We went through the manual labor to look at each moment while defining a value for each.
Then, we fed the computer these 100,000 moment-value pairs. The computer trained the algorithm. Then we have a smart computer brain that reads a moment and suggests a value based on the pattern it detects.
Image: A wordgram generated from the Happy Moments dataset (source)
Biggest NLP challenge
We soon faced a big challenge. Multiple values can contribute to the same happy moment. Without additional input from the writer, the computer does not know the definitive answer.
For example, a user writes “I had a great hike with my friends in the beautiful mountain today.” Three values potentially contribute to happiness here. It could be Social - hanging out with your friends, Fitness - hiking makes a person happy, or Nature - seeing the beautiful mountain.
We don’t have a perfect solution for this. Neither does the latest research. We settled for a satisfactory but imperfect solution that meets our needs. When ambiguity happens, we suggest multiple values and let the user choose which one is the most important.
Nudgingfor habit building
Can AI make us happier? Not by itself. In conjunction with other ideas like persuasive design, it can help nudge people to take more actions aligned with their values. Persuasive design believes that ability and motivations lead to behavior change. Being aware of your values increases your ability to be happier. Motivation is the next challenge.
Experiments are the most effective to motivate habit-building if executed consistently. You track your daily mood. Over time, you see a good mood on days you have done value-aligned activities, and vice-versa. This helps you internalize the importance of your action.
Where is it now
So, does it work? We did a controlled experiment with local college students (IRB compliant). Students who used our app for five weeks showed an improvement in their mood. But as soon as the study is finished, 75% of the users dropped off the app. Theoretically, AI can make us happier, but we have to commit to the opportunity. We need a better feedback loop and more storytelling.
This product is a lab prototype. The app is no longer in the App Store. You can still use the 100,000 happy moments dataset or learn more in this MIT Tech Review.
This essay clearly explains what the app does.
One comment: The title is "can AI make us happier?" It would be good if you provide an answer to this question, given your research and app-building. The answer can even be, "I don't know" or "given the project's status, it's too early to tell"