{"id":1122,"date":"2024-07-30T15:42:55","date_gmt":"2024-07-30T14:42:55","guid":{"rendered":"https:\/\/blogs.sussex.ac.uk\/librarycollections\/?p=1122"},"modified":"2024-08-09T09:31:39","modified_gmt":"2024-08-09T08:31:39","slug":"discus-library-joint-project","status":"publish","type":"post","link":"https:\/\/blogs.sussex.ac.uk\/librarycollections\/2024\/07\/30\/discus-library-joint-project\/","title":{"rendered":"DISCUS-Library Joint Project"},"content":{"rendered":"\n

The aim of this post is just to give a very quick overview of a joint project between the Library and the Data-Intensive Science Centre at the University of Sussex (DISCUS).<\/p>\n\n\n\n

This began with a successful proposal to the 2024 Development Studies Association (DSA) conference by Danny Millum, Paul Gilbert and Alice Corble, to run a panel entitled \u2018Decentring development thinking by engaging with archives\u2019.<\/p>\n\n\n\n

Danny and Alice, along with Tim Graves from the Systems Team, then decided to submit our own paper to the panel (which surprisingly enough was accepted!), entitled \u2018Camels and chatbots: an experiment in applying AI technology to the BLDS West African Economic Journals\u2019.<\/p>\n\n\n\n

This paper would \u2018draw on previous collaborative analysis of the British Library for Development Studies Legacy (BLDS) Collection, which involved using metadata from the collection to create a mapping tool to contrast its provenance with that of the main library collections at Sussex and use this to explore the potential for applying decolonial approaches to library discovery and research\u2019.<\/p>\n\n\n\n

This time though, the aim was to move from metadata to the data itself, inspired by yet another project (undertaken in collaboration with the University of Manchester) which was digitising another part of the BLDS collections, the rare West African Economic Journals.<\/p>\n\n\n\n

This provided a unique corpus of Global South-originating materials on which to explore the potential of a variety of AI tools, including chatbots, text and image analysis, and visualisation. Out of these journals we focused on the Camel Forum Working Papers from the Somali Academy of Sciences and Arts, hoping these would generate lenses on technological development discourse that offer a radical departure from traditional Global North analytical norms.<\/p>\n\n\n\n

So we had an overall idea and some materials to work with, but were still a bit vague about how we might deploy the myriad new AI tools becoming available. We basically took two main approaches:<\/p>\n\n\n\n

1. CamelGPT<\/strong><\/h3>\n\n\n\n

The first approach was to create a LLM limited solely to the Camel Forum Papers. This has yielded various decolonising-adjacent possibilities, some very straightforward in that the papers are now available to Somali researchers to interrogate via the CamelGPT.<\/p>\n\n\n\n

Others are less straightforward or less proven – we need researchers to try and break the model to see how accurate its superficially plausible responses are, and we\u2019d also like to find some way of comparing and contrasting the responses we are getting here to a comparable subset originating from the Global North.<\/p>\n\n\n\n

A further note – and many thanks to our digital humanities colleagues Jonathan Blaney and Marty Steer here – relates to issues of language. We\u2019d initially claimed that CamelGPT (and by extension ChatGPT) was \u2018language-agnostic\u2019 – that it would treat its contents equally no matter what language they were in, that we could ask it questions in any language, and we could  get replies in any language.<\/p>\n\n\n\n

However, this doesn\u2019t stand up to scrutiny. The Arabic corpus is smaller than the English and French corpora. So this increases the chances that if you ask CamelGPT a question in Arabic it could:<\/p>\n\n\n\n