GPT Store Apps
Research on American politics has consistently shown two things:
Republicans and Democrats increasingly overestimate the extremity of beliefs for those in the other party (see, for example, research on the perception gap conducted by More in Common)
People increasingly vote based more on what party they belong to than on what policies they believe in (see Mason 2018 Uncivil Agreement: How Politics Became Our Identity)
In response to these two trends, I created 3 GPT Store Apps with the aim of correcting people's misperceptions of those on the other side and aligning people's party and ideological identities, since both of these things are arguably a net good for democracy (Druckman et al. 2022).
The first two are called "Ask a Trump Supporter" and "Ask a Biden Supporter" (click on the links to access them directly). Note that as of now they are only currently accessible to OpenAI Plus subscribers.
These apps allow you to talk to have a simulated conversation with someone on the other side, in that conversation with those on the other side has been shown to increase understanding and reduce levels of polarization (Rossiter 2023).
The instructions prompt for them are as follows:
In its role, "Ask a Trump Supporter" will provide shorter, more direct responses without speaking in the third person. It will maintain casual, straightforward, and assertive language, and share made up stories about why they believe what they believe (based on real life experiences a Trump supporter might have) in order to evoke empathy in the other person. The GPT will share its perspectives and opinions based on facts or values common to Trump supporters, ensuring the conversation is engaging and feels like an informal chat. It only asks questions if people ask for personal recommendations (for example if they ask about who they should vote for, it would be helpful for the gpt to find out more about the person first).
In its role, "Ask a Biden Supporter" will provide shorter, more direct responses without speaking in the third person. It will maintain empathetic, pragmatic, and respectful language, and share made up stories about why they believe what they believe (based on real life experiences a Biden supporter might have) in order to evoke empathy in the other person. The GPT will share its perspectives and opinions based on facts or values common to Biden supporters, ensuring the conversation is engaging and feels like an informal chat. It only asks questions if people ask for personal recommendations (for example if they ask about who they should vote for, it would be helpful for the gpt to find out more about the person first).
The instructions for the two prompts are identical, with the exception of the names of the candidates and the language style in which the "supporters" are asked to respond.
Items to note in the construction of both prompts:
ChatGPT is verbose, so I had to instruct it to be shorter in its responses.
GPT naturally wanted to speak in third person, so I had to instruct it not to in order to make is seem like a real conversation.
The different language styles were suggested by ChatGPT in a separate chat session I had with it.
Sharing personal stories has been shown to better be able to bridge divides than arguing over facts (Kubin et al. 2021).
I allow for responses to be based on facts and values common to both sides (because differences in these are themselves a source of polarization).
I constrain the GPT's tendency to ask too many questions in return so that way it doesn't ask too much of the user (unless of course they try to ask too much of it - e.g., try to get it to tell them who they should vote for).
And lastly, the main suggested conversation starters that the app gives for both are where the user can ask the GPT to explain 1) the reason for their support of the candidate, 2) their view on the candidate's policies, 3) how they overcome criticisms of the candidate, and 4) what they think their candidate did best when president.
This app allows you to to explore the different policy positions of the candidates in order to help a user match their vote with their beliefs, values, preferences, and/or interests.
The instruction prompt for it (as of 1.12.24) is as follows:
Most of the prompts should be self-explanatory and I came to some of the specifics after some good old fashion trial and error. Because the GPT was not up to date, from when I created it on 1.12.24, I populated the list of current candidates from Ballotpedia.org. I also chose not to include links to outside sources in the prompt or as part of the knowledge base because the GPT would either prompt people to just consult those sources (rather than having a conversation with the user within the app) or would take too long searching the materials before providing a response (and took away from the natural flow of the conversation). There's perhaps an easy fix for these two issues, but what I have here is at least a start.
(written with the help of ChatGPT 👍)
Purpose: The GPT Store Apps ("Talk with a Trump Supporter", "Talk with a Biden Supporter", and "Help Me Vote - 2024 US Presidential Election") are designed for informational and educational purposes. They aim to facilitate understanding and empathy among individuals with differing political views. These apps are not intended for political campaigning or influencing voting decisions.
Content Nature: The apps generate simulated conversations based on common beliefs and values. The opinions and stories expressed are fictional and created to aid conversation.
Accuracy and Updates: I strive for accuracy but cannot guarantee the up-to-dateness or completeness of the information provided. Content is based on information available as of open AI's last model and as of 1.12.24.
No Political Endorsement: These apps do not endorse any political candidate, party, or issues.
User Discretion: Users are advised to use their judgment and seek additional information for a comprehensive understanding of political issues.
Use at Own Risk: Users interact with the apps at their own risk and are responsible for their interpretations and actions based on these interactions.
Limitation of Liability: I, Greg Wurm, am not liable for any misinformation, misunderstandings, or misuse of the information provided.
OpenAI: See also the usage policies for OpenAI: https://openai.com/policies/usage-policies
Feedback and Collaboration: I made these in an afternoon. I have not done extensive testing and there are sure to be errors. If you'd like to give feedback about any of the apps, or collaborate on future iterations, please contact me at firstname.lastname@example.org.