AI and Fact-checking

Decentralized Fact-checking & Provenance Community Group,

Hello. I recently found an interesting video: The Ultimate Fact Checking AI (John Hopkins, Stanford) (https://www.youtube.com/watch?v=ry3R7k6x1Pg, 19:10).
The video discusses pertinent topics including decomposition, decontextualization, and, to a lesser extent, deduplication.
Decomposition involves breaking a complex claim down into simpler subclaims. So, given a claim: “Charles Babbage was a French mathematician, philosopher and food critic”, the claim can be decomposed into subclaims: “Charles Babbage was a mathematician”, “Charles Babbage was a philosopher”, “Charles Babbage was a food critic”, and “Charles Babbage was French” [1].
Decontextualization involves taking a sentence together with its context and rewriting it to be interpretable out of its original context while preserving its meaning [2].
Combinations of these two topics are discussed both in the video and in one of the papers that it references [3].
Towards the very end of the video, it mentions deduplication which involves providing one claim per cluster, handling paraphrases.

The techniques discussed in the video could be useful for enabling end-users to fact-check arbitrary selections of content from documents in their Web browsers. Examples of existing systems include Citation Needed [4], and a related project, Add a Fact [5].
It is interesting to think about end-users being able to, similarly, make use of decentralized fact-checking resources from their Web browsers. It is also interesting to think about AI agents and multi-agent systems making use of such resources.

I wanted to share that video (and some of the publications that it discusses [1][2][3]) with the group. Thank you.


Best regards,
Adam Sobieski

[1] Wanner, Miriam, Seth Ebner, Zhengping Jiang, Mark Dredze, and Benjamin Van Durme. "A closer look at claim decomposition." arXiv preprint arXiv:2403.11903 (2024).
[2] Choi, Eunsol, Jennimaria Palomaki, Matthew Lamm, Tom Kwiatkowski, Dipanjan Das, and Michael Collins. "Decontextualization: Making sentences stand-alone." Transactions of the Association for Computational Linguistics 9 (2021): 447-461.
[3] Wanner, Miriam, Benjamin Van Durme, and Mark Dredze. "DnDScore: Decontextualization and decomposition for factuality verification in long-form text generation." arXiv preprint arXiv:2412.13175 (2024).
[4] https://meta.wikimedia.org/wiki/Future_Audiences/Experiment:Citation_Needed
[5] https://meta.wikimedia.org/wiki/Future_Audiences/Experiment:Add_a_Fact