Abstract

Since 2014, Catalyst Balkans has been amassing philanthropy data across seven Western Balkan countries in the Giving Balkans database. Traditionally reliant on press clippings, the manual categorization of news became increasingly cumbersome. Catalyst Balkans implemented AI solutions to remedy this, adopting a hybrid model: automating specific tasks while reserving critical classifications for human judgment. The AI approach entailed semantic clustering of media articles and employing large language models for historical data labeling and relevance predictions. A significant hurdle was the linguistic variance across the Balkans. Although initial trials with multilingual models faltered, language-specific models performed well. The vision remains to harness AI for further autonomous data processing. Building on this success, future aspirations point toward minimizing the Philanthropy Data Analyst (i.e., human) role. Two avenues emerge: One suggests overhauling the existing method to align more with AI capabilities, from reshaping classifications to rethinking donation sum calculations. The second leans on replacing analysts with AI agents anchored by large language models, like ChatGPT-4, which preliminary tests with ChatGPT-3.5 turbo show could handle tasks once seen as exclusive to humans. A blend of both strategies is the most practical for further automation. Conclusively, Catalyst Balkans’ experience embodies the fusion of technology and conventional data practices in philanthropy. While AI streamlines data collection and boosts quality, challenges persist, such as false positives and data veracity concerns. Moreover, the fluctuating nature of AI technology and costs necessitates agile adaptation.

Full Text DOI

BibTeX

@incollection{ugazio_case_2024,
  address = {London},
  edition = {1},
  title = {A {Case} {Study} on {AI} {Usage} for {Collecting} {Philanthropy} {Data} in the {Western} {Balkans}},
  isbn = {978-1-00-346861-5},
  url = {https://www.taylorfrancis.com/chapters/oa-edit/10.4324/9781003468615-18/case-study-ai-usage-collecting-philanthropy-data-western-balkans-nikola-milinkovi%C4%87-marko-galjak},
  doi = {https://doi.org/10.4324/9781003468615-18},
  booktitle = {The {Routledge} {Handbook} of {Artificial} {Intelligence} and {Philanthropy}},
  publisher = {Routledge},
  author = {Milinković, Nikola and Galjak, Marko},
  editor = {Ugazio, Giuseppe and Maricic, Milos},
  month = nov,
  year = {2024},
  pages = {240--256}
}