Taming Toxic Talk: Using AI Chatbots to Intervene on Reddit


Jeremy Foote
Purdue University


Deepak Kumar
UC San Diego


Ryan Funkhouser
Purdue University


Dyuti Jha
Purdue University


Hsuen-Chi Chiu
Purdue University


Hitesh Goel
IIT Hyderabad

People are mean online

  • In just an 18 month period on Reddit, people posted over 14 million highly toxic comments.
  • Toxicity is incredibly common, and it has negative effects
    • Self-censorship
    • Fear
    • Harrassment
    • Etc.

Kumar et al., 2023. Understanding the Behaviors of Toxic Accounts on Reddit. WWW ’23.

Conversations can help

  • In some contexts, interpersonal conversations can change people’s attitudes and behaviors.
  • However, having conversations with those posting toxic behavior requires a huge amount of emotional labor.

E.g., Kalla and Broockman, 2020. Reducing Exclusionary Attitudes through Interpersonal Conversation: Evidence from Three Field Experiments. American Political Science Review.

Can bots help?

  • We ask whether chatbots can help.
  • Will people posting toxic content engage with chatbots?
  • Do these conversations help to change their behavior?
  • Large-scale study of thousands of Reddit users, working with moderators of some large subreddits.

Results so far are mixed

  • Qualitative analyses of conversations shows promise, but…