Next Week Tonight: Simulating Counterfactual Narratives of the future using Agentic Knowledge Graphs

Understanding the ripple effects of events—both real and speculative—is essential for navigating complex futures. Large Language Models (LLMs) have emerged as powerful tools that offer a user-friendly and narrative experience for question answering and reasoning across large corpuses of unstructured data [15, 96]. While LLMs can respond to complex ‘what-if’ questions, they typically provide single, unverifiable answers. Even with retrievalaugmented generation (RAG) that grounds LLM responses on external sources, the opacity of reasoning pathways undermines trust in model outputs [97]. Next Week Tonight builds on the narrative and reasoning capability of LLMs further by enhancing the exploration of what-if futures and making it more transparent and evidencebased. NWT exposes the underlying knowledge graph, allowing users to inspect inference pathways directly. This also enables the generation of multiple, diverse scenarios from a single condition—each following different but explainable causal chains. In testing 15 counterfactual prompts that span diverse news topics, NWT produced scenario narratives that were rated as significantly more causally coherent, transparent, and easier to audit than standard chat completions. Beyond technical performance, NWT reinvents scenario planning as an interactive narrative experience - encouraging curiosity, critical thinking, and deeper engagement with the complexities of future events. By surfacing not only what could happen but why and how, NWT aims to empower analysts, policymakers, and the public to navigate uncertainty with greater clarity and confidence. Github: https://github.com/viral-medialab/next-week-tonight

Name thesis:degree_name

Master

Department dc:contributor.department

Program in Media Arts and Sciences (Massachusetts Institute of Technology)

Grantor dc:publisher

Massachusetts Institute of Technology

Year dc:date.issued

2025

Author dc:creator

• Agarwal, Gauri

Advisor dc:contributor.advisor

• Lippman, Andrew B.

Statement dc:rights

• In Copyright - Educational Use Permitted
• Copyright retained by author(s)

Licence dc:rights.uri

• https://rightsstatements.org/page/InC-EDU/1.0/

Handle dc:identifier.uri

https://hdl.handle.net/1721.1/164132

OAI identifier oai:identifier

oai:dspace.mit.edu:1721.1/164132