Page 417 - AI for Good Innovate for Impact

AI for Good Innovate for Impact

ICRC: Preliminary discussions have highlighted the relevance of our work in humanitarian
contexts and we have a memorandum of understanding with ICRC. We aim to build on these
initial exchanges. Their role is to provide iterative feedback during the development phase
and serve as early adopters once we expand. 4.4-Productivity

2�2 Benefits of the Use Case

Radiolytica directly tackles the destabilizing effects of misinformation, disinformation, and
hate speech that undermine societal trust. By providing an automated, scalable solution
for monitoring harmful content on community radio, the project enhances the integrity of
information flows, supports accountable institutions, and fosters peace in conflict-affected
regions.

Leveraging cutting-edge AI, advanced speech-to-text, and NLP technologies, Radiolytica
modernizes media monitoring in low-resource settings. This innovative digital infrastructure not
only addresses current communication challenges but also creates a resilient system capable
of evolving with future technological advancements, contributing to sustainable industrial
development.

Radiolytica is underpinned by a collaborative framework that unites global tech experts,
academic institutions, and humanitarian organizations with local Congolese partners. By
involving Congolese organizations, the project ensures that local context and expertise drive
the development and application of AI, promoting inclusivity and capacity building within
the region. This approach fosters mutual learning, strengthens local innovation ecosystems,
and aligns with the vision of building strong, inclusive partnerships that drive sustainable
development.

2�3 Future Work

Our future work will build on our successful pilot to transform Radiolytica into a fully scalable
system for monitoring harmful content across Eastern DRC’s community radio networks.

Scaling and Refinement of AI Capabilities:

We will expand our prototype to monitor up to 150 radio stations, enhancing automated
transcription and NLP analysis. Our focus is on refining our pre-process pipeline for French
and Swahili, reducing errors from high noise levels and overlapping audio. With additional
resources, we also aim to explore incorporating Lingala, broadening our language coverage.

Enhanced Human-AI Hybrid Validation:

A core aspect is our hybrid model that pairs automated processing with continuous human
validation. Local experts, journalists, and media partners will verify AI outputs, ensuring
contextually accurate, reliable data and mitigating risks like AI hallucinations.

Local Partnerships and Capacity Building:

Leveraging collaborations with the Development Economics Group at ETH Zurich and ICRC,
we will strengthen partnerships with Congolese organizations. This will enhance local expertise
in AI model training and promote capacity building in digital transformation, ensuring our
solution is both inclusive and sustainable�

381