IMFD researcher Andrés Abeliuk wins Fondecyt Initiation Project 2026

Andrés Abeliuk, researcher at Millennium Institute Foundational Research on Data academic at the Department of Computer Science at the University of Chile, was recently awarded a Fondecyt Research Initiation Project 2026, as part of the call for proposals made by the National Research and Development Agency (ANID).

The initiative, entitled "Towards Fair and Contextualized Simulations of Public Opinion in Chile using Large Language Models," seeks to explore new ways of analyzing and representing Chilean public opinion through large-scale language models currently used in generative artificial intelligence systems.

The Fondecyt Initiation competition aims to encourage and strengthen the development of scientific and technological research of excellence, promoting the inclusion and consolidation of new researchers by funding two- or three-year projects in all areas of knowledge.

According to Abeliuk, his research analyzes how language models can become an innovative tool for simulating public opinion in Chile. "These models have the capacity to process large volumes of information and generate responses that reflect social patterns more quickly and at a lower cost than traditional surveys," he says.

The main objective of the project is to evaluate and improve the ability of these models to fairly and accurately represent the opinions of different groups in Chilean society, incorporating both survey data and the way in which People and justify their opinions.

Currently, many of these models are trained primarily with data from US contexts, which can introduce significant biases when applied in countries such as Chile. "This research is relevant because it proposes methods to reduce these biases, improve the representativeness of historically underrepresented groups, and develop more reliable tools for social analysis and public policy formulation. In addition, it offers cost-effective and complementary alternatives to traditional surveys," says the DCC UChile academic.

The initiative also takes an interdisciplinary approach, integrating computing, social sciences, and sociology, which allows for the development of artificial intelligence to be addressed from both a technical and social perspective. In particular, the project incorporates criteria of equity and epistemic justice, emphasizing that diverse voices, experiences, and social realities are adequately represented in AI systems. In this way, the research seeks to contribute to the development of technologies that are more responsible, transparent, and sensitive to the Chilean context, preventing these systems from reproducing or amplifying existing inequalities.

This line of work is based on the researcher's previous investigations into bias, representation, and fairness in algorithmic systems, such as the recent study "Fairness in LLM-Generated Surveys." This work shows that language models do not achieve the same level of accuracy for all social groups when predicting survey responses: they tend to perform better on People with higher educational levels and defined political positions, and worse on older adults, People , indigenous peoples, and groups with lower educational levels. These results reinforce the project's motivation to move toward more fair, representative, and contextualized artificial intelligence models.

Source: DDC UChile