Nanda Kishore Sreenivas

Liam Hebert, Gaurav Sahu, Yuxuan Guo et al.

We present the Multi-Modal Discussion Transformer (mDT), a novel methodfor detecting hate speech in online social networks such as Reddit discussions. In contrast to traditional comment-only methods, our approach to labelling a comment as hate speech involves a holistic analysis of text and images grounded in the discussion context. This is done by leveraging graph transformers to capture the contextual relationships in the discussion surrounding a comment and grounding the interwoven fusion layers that combine text and image embeddings instead of processing modalities separately. To evaluate our work, we present a new dataset, HatefulDiscussions, comprising complete multi-modal discussions from multiple online communities on Reddit. We compare the performance of our model to baselines that only process individual comments and conduct extensive ablation studies.

Nanda Kishore Sreenivas, Kate Larson

We study the consequences of information asymmetries and misaligned incentives in settings with multiple independent agents. We model an interaction between a Sender, who holds vital private information but cannot act, and a Receiver, who must make decisions but is dependent on the Sender's information. We find that the Sender learns an optimal communication strategy that the Receiver reliably acts on. Importantly, this strategy is highly sensitive to the degree of conflict in the agents' rewards and the amount of environmental information the Receiver can already observe. We introduce a mechanism allowing the agents to form linear contracts, where a price is established for the information. We demonstrate that the Sender learns to use these payment structures to improve its rewards, though this comes at a cost of "fairness" between agents as the Sender is able to extract much of the Receiver's surplus. This raises questions about fairness, contract design, and learning in the context of multi-agent systems.

Nanda Kishore Sreenivas, Shrisha Rao

Modeling social interactions based on individual behavior has always been an area of interest, but prior literature generally presumes rational behavior. Thus, such models may miss out on capturing the effects of biases humans are susceptible to. This work presents a method to model egocentric bias, the real-life tendency to emphasize one's own opinion heavily when presented with multiple opinions. We use a symmetric distribution centered at an agent's own opinion, as opposed to the Bounded Confidence (BC) model used in prior work. We consider a game of iterated interactions where an agent cooperates based on its opinion about an opponent. Our model also includes the concept of domain-based self-doubt, which varies as the interaction succeeds or not. An increase in doubt makes an agent reduce its egocentricity in subsequent interactions, thus enabling the agent to learn reactively. The agent system is modeled with factions not having a single leader, to overcome some of the issues associated with leader-follower factions. We find that agents belonging to factions perform better than individual agents. We observe that an intermediate level of egocentricity helps the agent perform at its best, which concurs with conventional wisdom that neither overconfidence nor low self-esteem brings benefits.