AI Tools from CAS
SciFinder: SearchSense
"SearchSense" is an AI search assistant incorporated into the primary search box of CAS SciFinder that interprets a search query and returns appropriate references, substances, and reactions, as well as a high-level overview of the topic at hand.
"SearchSense" is available in: SciFinder
- Simply enter a natural language query (or a traditional keyword query) and go. There are no longer separate search tabs for "references," "reactions," or "substances"—everything goes in one search box.
- Example: how do fireflies produce light
- Tip: The AI search assistant is enabled by default. To search without the AI assistant, choose the "Advanced Search" button below the search box.
The results screen shows the AI summary of the topic, as well as top results in categories including references, substances, and reactions.
- Click on the "View All" button to see the complete AI summary.
Example of a full AI summary.
The "References" results screen also includes the AI summary, as well as the "Query Interpretation" link near the top left-hand part of the screen.
- Click the link to see how your natural language search was converted into keywords.
In the right-hand sidebar, you can view the search constructed by the AI tool.
- Click on the "Search Original Query" button if you wish to search your original query without the AI enhancement.
More information about AI enhancements from CAS
Under the Hood
What powers CAS SciFinder AI tools?
CAS SciFinder’s “SearchSense” and related AI features employ proprietary, science-specialized AI models trained and verified by domain experts using the high-quality, structured CAS Content Collection—not general-purpose large language models. The models are closed-domain, optimized specifically for chemistry and scientific information retrieval, and operate in a secure, closed environment to ensure privacy and data accuracy. SearchSense uses natural language processing to accept complex scientific queries, returning AI-generated summaries traceable directly to the original literature. The approach incorporates elements of retrieval-augmented generation (RAG), but details on third-party model vendors or specific architectures are not disclosed; the emphasis remains on scientific reliability, transparency, and human expert oversight.
AI-assisted, human-verified text