Blog

Experimentation: AI Assistant to find relevant scientific documents and extract accurate answers directly from sources.

Opscidia, in partnership with NGI Search and funded by the European Commission, has been experimenting with an innovative question answering system for scientific documentation.

The system developed by Opscidia offers an innovative approach to accessing scientific information. It allows users to ask questions in natural language, and obtains precise answers extracted directly from relevant scientific documents selected by the AI. Key elements of the answer are highlighted in the sources, facilitating verification of the information proposed by the artificial intelligence.

This technology is based on the latest advances in artificial intelligence, particularly in natural language processing and semantic analysis.

This experimental project is perfectly aligned with the objectives of illustrating how AI can be used to improve access to scientific information, while respecting the principles of openness and transparency advocated by the NGI initiative.

    1. How does it work?
    2. Science Checker’s technical architecture and functionalities

1 .How does it work?

The Science Checker is an innovative question answering tool for scientific documentation. The system works in four simple but powerful steps:

 

  • The user asks a specific question (What are the effects of global warming on coral reefs?).

 

  • Next, the tool suggests a selection of relevant scientific articles.

 

  • The user can then choose which articles to query.

 

  • Finally, artificial intelligence highlights the answers directly in the selected articles. (The AI highlights key passages, such as the increase in the frequency of coral bleaching episodes due to rising ocean temperatures).

 

This approach considerably reduces the time spent searching for relevant scientific documents and information. At the same time, it offers greater accuracy and relevance thanks to the use of advanced AI and machine learning technologies.

3. Science Checker technical architecture and features

The core of the project is based on a unique and sophisticated technical architecture, developed by Opscidia’s team of experts, including Fréjus (CTO) and Loic (ML Engineer) for R&D and Oumayma and Loic ( Full Stack Developers ).

 

Thanks to the integration of LLMs and KGs, the system offers precise and contextually relevant answers, while reducing search time and improving the accessibility of scientific information.

How does it work?

The two-way paradigm for transparency and logical reasoning lies at the heart of Science Checker’s technical prowess. Sounds complicated, but it’s simpler than you think.
This paradigm combines large-scale language models (LLMs) with knowledge graphs (KGs).

LLMs generate natural language responses, while KGs structure and verify these responses in documents to ensure accuracy and consistency, and provide evidence to support the natural language responses.

This bi-directional approach contextualizes information and reduces errors, providing accurate and reliable answers to scientific questions posed by users.

The impact of this tool extends far beyond mere efficiency: it promises to accelerate discovery, improve research accuracy, and open up new avenues for more ethical and accessible science. The future of research looks brighter, faster and more inclusive than ever, thanks to tools like Science Checker that push the boundaries of what’s possible in the exploration and exploitation of scientific knowledge.