Dislikes & Arguments
We should be arguing with each other more!
Not the way we currently argue with each other. Definitely not the way we argue with each other online. But we should be having more productive arguments with each other. Arguments where we listen, understand, reflect, and respond rather than react. Productive arguments are important, especially among people working towards similar goals, to open up to different perspectives and work together.
In my experience with discussions escalating into arguments, and having witnessed a single comment ignite hateful exchanges online, one observation stands out: individuals often perceive those arguing with them as the "other side," even when their larger goals are fundamentally similar.
Exploring this area reminded me of a book called 'The Science of Hate by Matthew Williams'. In the book, he talks about in-group and out-group dynamics and also the bonding effect from mutual dislikes.
I wanted to experiment with using trivial dislikes among participants in an argument to create a sense of connection and see if there is a difference in the way the argument unfolds.
I created a Discord server called Dislikes and Arguments. I used Discord since usernames can be customized for each server. When someone joins a server, they would be asked to edit their usernames to add 1-3 icons from a list of icons with dislikes assigned to them. For the first version, I've used trivial dislikes specific to New Yorkers.
The members of the server can then start a new channel on a topic they would like to discuss or join an existing discussion.
I've built 2 bots to complement the discussions for the server
1) Anbot
2) Connectbot
Anbot
Anbot is a bot that monitors the server for unkind messages and responds to unkind messages by asking them to reconsider the way they have expressed their thoughts, ideas, or feelings.
Anbot uses the AFINN-111 for sentiment analysis for the messages sent. The AFINN-111 contains a list of English words, each word having a rating between 5 to -5. The bot breaks down the messages and creates an array of the list of words in a message, and calculates the total based on the rating assigned to each word.
Anbu(அன்பு) means kind in Tamil. The name Anbot was made from combining the words Anbu+Bot
While Anbot worked in most cases, its limitations were:
1) Works only for English
2) Did not understand typos/spelling mistakes
3) Lacks context and replied to messages that were not unkind
Connectbot
I created ConnectBot as a way for people to find others with similar interests within servers with a large number of members. ConnectBot uses Hugging Face Transformers Library to compare the similarity of different messages in a channel and respond with the username of the person who has expressed interest in a similar topic but has some differences in opinion
When 'match' is sent on a channel, the messages are passed through the extractor which assigned a numerical vector(embeddings) to the messages. An array of embeddings is created from this. The embedding for the query message (the message sent right before match) is also calculated.
The cosine similarity is then calculated between the query message and the other messages in the channel. To create a balance between interest in a topic and differences in opinion, I started by matching someone to the 3rd most similar message instead of the most similar message.
The username of the sender of the third most similar message is retrieved and sent on the channel.
You can join the server through the button below. I've created a channel called testing-playing-with-bots if you wish to interact with the bots without participating in a discussion.
If you have any questions or are interested in collaborating on this project, please email me! carolinenivetha@gmail.com
Next steps:
-
Testing the server with people interested in different areas
-
Outcome when common connection is made through likes vs dislikes
-
Experiment with different kinds of likes and dislikes