Post Status: Active
Initiative recipient country/countries: Nigeria, Africa (not specified)
Initiative dates: 2023 –
Last updated: 3 September 2024

Initiative Description

Online information pollution undermines elections and democratic processes, by polarising societies and inciting violent insurrection. In Nigeria, the 2023 general election was marred with various fake news, propaganda, hate speech, and incited violence on disenfranchisement. Mis/disinformation alike erode public trust and has derailed democratic systems in sub-Saharan Africa from Burkina Faso, Mali to Niger.

In today’s world, more than traditional news platforms like newspapers, radio and televisions, people are depending on social media platforms to get the latest trending news. These platforms especially the micro blogging platform, X (formerly Twitter) serves as one of the best way to disseminate important information, reach wider target audience and track engagement/feedback from the general public in very short amounts of time but at the same time it also allows the spread of fake news to a large section of the population. The effects and scale of mis/disinformation are ubiquitous. Social media platforms are designed to disseminate content using algorithms that can modify patterns of individual exposure in opaque ways, often prioritising content that provokes extreme reactions from users who spread the information over social media with little regard for its veracity. This problem is exacerbated because newsrooms lack the capacity and tools required to verify the integrity of news articles, videos and images posted on social media platforms like X, Facebook, Whatsapp, Tiktok etc and counter the spread of fake news especially in this era of AI advancement where it becomes more difficult to separate deep fakes from synthetic media.

Fact-checking aims to identify claims expressed online and label them as factually true or false and it has significant real-world impacts as it enables debunking false claims and supporting the true ones to prevent disinformation that may polarise societies and inciting violent insurrection. However, it is a challenging, technical, and time-consuming task for humans. Fact checkers, journalists, and editors have to fact-check information and identify reliable evidence from various sources manually combining several methods together. It is a mundane task which by the time completed, the damage would have been done.

ChatVE is a hybrid chatbot integrated on X (Twitter) which combines the power of Generative AI and fact checkers to provide quick conversational information through text or text-to-speech to fact check in real time thus tackling online mis/disinformation and propaganda. Using our own trained Large Language Model (LLM), ChatVE fact checking algorithm can assess the credibility of claims and provide interpretable valid evidence that explains why a certain claim is considered as factually true or fake by tagging on X (@ChatVE_) + specify the task using the default "!factcheck" command + context of the claim to be fact checked.

ChatVE's solution fast-tracks fact-checking, the chatbot that accepts queries in URL, Image, Video and Document format, fact check the claim and return the verdict at the speed of social media in 3 stages:

• Claim detection - This is where ChatVE identifies claims that require verification from input value.
• Evidence Retrieval - At this stage, ChatVE finds sources supporting or refuting the claim; knowledge base, text corpus, tables, metadata, timestamp, wayback machine. ChatVE can browse the internet on its own, a feature that was missing in the early version of ChatGPT.
• Justification - The last stage is where ChatVE assesses the veracity of the claim based on the retrieved evidence. This stage is decomposed into two:
  - Verdict Prediction - Assign truthfulness or false labels to claims; marking piece of media as manipulated, missing context, edited, transformed, staged, Satire, misleading.
  - Justification - Provide interpretative explanations for the verdict by citing and referencing reputable sources.

Related imagery: