Video presentation of our paper (More Victories, Less Cooperation: Assessing Cicero’s Diplomacy Play) presented at ACL 2024:
https://arxiv.org/abs/2406.04643
Thanks
=======================
We thank Meta for granting access to over
40,000 games played on the online platform
webdiplomacy.net and for open sourcing Cicero.
This commitment to open science allowed this independent reproduction of Cicero’s juggernaut abilities but also let us have some fun. We especially
thank to Mike Lewis for offering valuable insights
into Cicero’s communication.
Our thanks also go to Tess Wood for training
AMR annotators, Sarah Mosher for their English AMR annotations, and Isabella Feng for her exploration of LLM-based AMR parsing. We thank Kartik
Shenoy, Alex Hedges, Sander Schulhoff, Richard
Zhu, Konstantine Kahadze, and Niruth Savin Bogahawatta for setting up DAIDE baselines.
We also thank the small community of researchers looking at communication and deception
in Diplomacy for their feedback, commentary, and
inspiration: Michael Czajkowski for discussing the
nuances of detecting persuasion; Stephen Downes-Martin for teaching us that deception is far more
than lies; Karthik Narasimhan and Runzhe Yang
for their insights into lie detection and stance; and
Larry Birnbaum and Matt Speck for discussions on
mapping DAIDE and English. And thanks to Justin
Drake, Niall Gaffney, and the members of TACC
for setting up environments for computer–computer
games and making sure that we had GPUs ready
when players were ready to play.
Finally, sincere thanks to the member of the
Diplomacy community who took the time to play
against Cicero in this unconventional setting.
This material is based upon work supported by
the Defense Advanced Research Projects Agency
(DARPA) under Agreement Nos. HR00112290056
and HR00112490374. Any opinions, findings, conclusions, or recommendations expressed here are
those of the authors and do not necessarily reflect
the view of the sponsors.