Senior suicides have tripled in the last decade due to the large baby boomer population retiring alone. Loneliness is the biggest problem seniors face that contributes to their mental health. According to an AARP survey, a third of adults over the age of 65 are lonely and highly prone to suicidal thoughts. With this in mind, we aspired to create a solution to this problem by allowing senior citizens to gain a friend!

What It Does

GAiN a Friend is a speech-driven video chat service powered by Generative Adversarial Networks (GAN). This is done by selecting a human face and voice that resonates with the user most. After the persona is selected, the user can begin interacting with the GAN-generated person 24/7.

How We Built It

GAiN a Friend is built using Generative AI, particularly Generative Adversarial Networks. In order to run these large models, we needed to use Deep Learning VMs from Google Cloud to run these models on high compute engines. For conversational AI and speech conversation, Houndify APIs were used.

Ethical Considerations

Mass media has been increasingly focused on a future with AI, with movies such as Her and Ex Machina to TV shows such as Black Mirror. As our team was brainstorming this idea of a chatbot to improve communication and curb feelings of isolation, we kept considering the ethical implications of our work and focused on three main conditions.

  1. Privacy and Data Ownership: As seen from the first ELIZA chatbot in the 1960s, humans are inclined to trust bots (and not believe that it's a program!) and tell it sensitive information. It is imperative that a user's valuable information stays with the user and does not fall into other hands. With this in mind, our team decided to store the program and information locally and not share the data collected. We are also not collecting any videos of the user to ensure that their autonomy is respected.
  2. Transparency: In order to build trust between the user and company, we must be upfront about our chatbot. To ensure there is no confusion between the bot and a human, we have added a disclaimer to our site letting the user know their conversation is with a bot. Moreover, our code is open source so anyone that is curious to understand the "behind the scenes" of our chatbot is welcome to take a look!
  3. User Safety and Protection: There have been many cases of hackers influencing digital personal assistants to go rogue or turn into wiretaps. To ensure this doesn’t happen, our “friend” uses the Houndify small-talk API to make sure the user can have broad conversations. Furthermore, with minorities and LGBTQ+ elders feeling significantly more lonely than others, our chatbot will not make any assumptions about the user and can be easily accessible by all populations.

Challenges We Ran Into

  • Learned how to use GANs for the first time
  • Using Python 2.7 packages for Python 3.5 was... INSANELY CHALLENGING
  • Dealing with deprecated packages
  • Caching errors on Jupyter Notebook that have cryptic errors
  • Using JavaScript in a Flask app for audio recording
  • Learned how to use ethics into consideration in our design solution
  • Working with team you've never worked with before

Accomplishments That We Are Proud Of

We are proud of being an ALL WOMEN team :)

What's Next for GAiN a Friend

In the future, this concept can easily be expanded to the service sector for service providers, such as psychologist, physicians, and family members. GAiN a Friend wants to incorporate additional API. Focus on a more accessible design for the App development to help elderly folks interact with the app better.

Built With

  • conversational-ai
  • deep-learning
  • generative-adversial-networks
  • google
  • google-cloud
  • houndify
  • python
  • scikit-video
  • speech-to-text
  • torchvision
Share this project: