Familiar is an AI chatbot built as a supportive companion. This piece was developed as part of the InterfaceCognitive A.I. Collab working with design leaders at IBM: Han-Shen Chen and Bell Labs: Wei Wei. Our app helps users relieve stress and negative emotions built up throughout their day by being there to listen and talk to you. Your Familiar is a self-reflection of your own personal health, changing and growing with your own mood which it picks up through talking with you.
In investigating conversational support our team researched cases, symptoms, and treatments of depression and self-isolation from various papers and conducted interviews. During this process, we interviewed a mix of caretakers and those who have suffered from depression via self-isolation. From each set of interviews looking at the symptoms, effects, and the various treatments used gauging their success. From this data, we developed the persona of Jessica, a stressed college student, who works very hard but causing her to have few friends and no one to talk to about her day to day stresses.
From our initial research, we found that conversations, in particular, listening provided the greatest benefit to suffers from self isolative behavior. However, we found it was hard for people like Jessica to talk to people so instead we looked to animals. Emotional support animals as a proven benefit to providing a great outlet for stresses, connection, and empathy.
Familiar reaches out to you to talk when in need or when your most free. To feed and care for your familiar all you need to do is chat.
Familiar is highly animated using voice, expressionsm physicallity and emotes shows emotions/interest to your day.
Your familiar is unique with specific needs: Food, Hygiene, and Mood. They are satisfied your interactions and changes in their familiar’s status.
Familiar reacts as a personal reflection. As your familiar improves in health our user’s social life improves being open to conversations and others.
Our group focused on the user to pet interaction using multiple methods of engagement including machine learning, character design, and therapy techniques. Our familiar grows with you and in turn, reflects your emotions. The diagram below showing the conversational loop that we built.
To make sure that people would want to engage initially with their familiar our visual style is very bright, colorful, and playful. We created an initial set of familiars from Pandas, Bears, Dogs, and Cat with various colors adding a sense of personalized companions. The semi-flat form also provides us the ability to animate multiple features easily with pre-built emotional states and emotes to provide context to the conversation adding another level of depth to the interaction mixing.
To create a realistic and engaging conversation with your familiar our machine learning system is a 4 step process. First engaging the users by calling out to them at a set time or when pet needs are serious. Second, revealing the user’s task via the pet’s condition and emotions. Third, get user input via Watson’s speech to text analysis and respond accordingly based on text to speech API and condition matching in Waston Conversational mapping. Lastly recording data and feedback setting updated conditions for the next conversation.
This flow has multiple expansions detailed below explaining how the analysis works based on a database of set concerns and emotions related to them. Each of these concerns has related keywords of which there correlating emotional statements are run through a sentimental analysis logging concerns and reflecting back comfort, encouragement, or curiosity via the visual style based on the conditions meet. Thesis concerns can then later be referred back to on another day showing remembrance and interest to help improve localized concern categories.
As part of the final project of the IBM Cognitive AI collaboration, we presented our research and prototype to various colleagues in the field of AI interfaces and UX design. Familiar was considered extremely successful and well regarded by the group, I hope to continue this project after review with my other group members Shu and Yi expanding familiar into possibility as a physical toy-like interface and further detailing out the system’s capabilities using IBM Watson’s conversational interface and node.