WHOCAN

WHOCAN with a query

Inspiration

When prompted to address issues regarding financial inclusivity and affordability, it is impossible to ignore that access to suitable education, healthcare, humane living conditions and more is not a given for everyone. Here at UM Dearborn I am a part of the Inside Out Prison Exchange where university students go to Macomb Correctional Facility and have conversations with those inside about topics concerning social justice, race, and philosophy. Among class conversations, we also talked about the reality of prison life as well. Prisons are underfunded, with short staff, limited access to rehabilitation programs, and poor living conditions affecting the correctional system’s ability to humanely care for those in our prison systems. And the people who find themselves are victims of a society that neglects the less fortunate, forcing many into a life of crime. While it doesn’t excuse the acts that landed them there, it makes it glaringly obvious that something needs to be done. A common misconception is that supporting, participating in, and/or donating to nonprofit organizations means we are doing everything we can to support programs such as prison reform, after-school-care, homeless shelters, etc. The reality is that the most effective way to elicit change in government-funded institutions like schools and prisons, is by changing the laws that allocate the valuable funds that can change the lives of our state’s most vulnerable inhabitants. When asking my incarcerated classmates what we can do to help those on the inside, to make a genuine impact on those INSIDE prison walls, the overwhelming response was: VOTE. The best way to solve this issue is through reformation both by elected officials as well as proposals pushed through during local elections. Another example is schools. Especially for schools in impoverished areas, where low property taxes lead to dangerous underfunding, the responsibility to pick up the slack falls to the community. Again, donations and after school programs only go so far, when the reach solution is intentionally increasing funding from the government in the places that need it most. When more people vote, the law becomes a better reflection of our consensus as a community and precious government funding finds its rightful place in the hands of those who will make the most of it. Thus improving the lives of our state’s most vulnerable residents, such as children, prisoners, the homeless, and more.

What it does

We want to convince this valuable population of voters to make a difference by making it quick and easy to search up such issues, by providing a one step solution to find the candidates names they will check on the ballot. WHOCAN uses generative AI to make it easy to identify the politician who aligns with your views on issues that matter the most, providing an easy-to-scan list with the person’s name and a summary of their stance on the topic searched. It only takes two steps: first, type in a few key words regarding the topic and your polling district number, next fill in the names on the ballot that align with your views.

How we built it

We utilized the Google Cloud Platform (GCP) and Vertex AI for WHOCAN to process the candidates website and generate a summary given the users custom prompt. To do this, we tested several iterations of prompts in the Vertex AI Prompt to try and get the AI to give us the output we desired. This involved trying various methods such as zero-shot and one-shot prompts, and in the end we went with a zero-shot prompt that utilizes the text-bison-32K in order to ensure we don’t max out our input tokens with the imputed website and prompt. For the backend, we first created a website scraping tool where when given a URL, the program would download the HTML file and remove all HTML tags so we are left with just the content tags. Following this, we would submit our custom prompt to the Vertex AI API and once the AI had finished processing the response, we display the AI’s response to the user.

Challenges we ran into

Not all politicians have websites with the same format, making it difficult to scrape them using AI. Also, the AI platforms we had access to largely did not have access to the internet. This meant we had to carefully scope the project so that we could finish in the time required while still demonstrating the functionality with different politicians, views, and districts. Many politicians either had no campaign website with their stance on issues, or their campaign websites had been shutdown since the election had passed.

Accomplishments that we're proud of

We feel that this application could actually be used for social change. Facilitating the gathering of information is a task that would help voting become more accessible to people, hopefully improving the voter turnout which could have a butterfly effect lasting decades. It helps solve the problem that all three of us face when skipping the polls, especially in off years. We don’t know enough about what is being proposed or the people running to make a difference anyways. This website allows for more effective crossing or party lines, because it focuses on the issues and doesn't even tell the user which party the candidate represents.

What we learned

We learned how to interact with the Vertex AI API to query the LLM for the summaries we were looking for. We had to learn how to engineer our prompts to get the AI to provide the output we wanted as well as how to embed the AI prompt into our program to allow the website to query the AI.

What's next for WHOCAN

Many candidates don't have information on their stance on various issues, which could be solved by making candidates aware of this issue and asking for more coverage in order to make it easy for voters to educate themselves, thus improving voter turnout. Also, depending on the state and/or level of government, it is difficult to get a usable list of candidates from which to conduct research. In order to take WHOCAN to the next level, we would have to ask legislators to standardize such information and urge politicians to better publicize their views. However, this idea is very scalable if these issues were solved and we feel it could also be applied to proposals and acts that are being pushed through as well.