Brainstorming, yet another thing AI can help with?

AI (or LLMs) has advanced rapidly in recent years, with programs that can generate text, code software, and even defeat humans at complex games like chess and Go. Now, scientists are exploring how AI can aid in research and discovery in a more conversational way.

At the Center for Functional Nanomaterials (CFN), a U.S. Department of Energy facility at Brookhaven National Laboratory, researchers are interested in developing AI systems to help accelerate discoveries in nanomaterials. While thinking of ways to achieve this CFN leader Kevin Yager imagined an AI chatbot that could serve as a collaborative partner for scientists.

Kevin Yager (Jospeh Rubino/Brookhaven National Laboratory)

“We’ve been looking into new ways to leverage AI to speed up nanomaterial discovery,” said Yager. From automating experiments to analysing data, AI is transforming many aspects of research at CFN. But Yager wanted to create an AI assistant focused on scientific language and reasoning.

This is why Yager and his team built a scientifically knowledgeable and scientifically chatty chatbot. Powered by recent advances in natural language processing, it can understand questions, summarise publications, and point researchers to relevant information. Unlike mainstream chatbots, Yager’s creation leverages a vast library of domain-specific documents, allowing it to ground its responses in field specific factual knowledge.

Behind the Scenes

A key challenge was stopping the chatbot from generating plausible but false information (known as hallucinations), a common issue with AI language models. Yager’s solution was an embedding technique that encodes the semantic meaning of text. By linking users’ questions to relevant snippets from published papers, the chatbot assembles truthful, contextual responses.

“It needs to rely heavily on documents to provide sourced answers,” explained Yager. Although not perfect yet, the chatbot can already tackle difficult questions and stimulate new research ideas as part of a conversational approach.

AI Sidekick

Yager envisions scientists using AI tools like this chatbot to free up their time for higher-level thinking. As computers handle routine tasks, researchers can focus on challenging problems and big-picture questions.

He imagines that potential applications include:

  • Summarizing the key points of publications
  • Organizing and classifying research documents
  • Getting scientists up to speed on new topics
  • Suggesting novel research directions
  • Providing factual answers to literature questions

“I’m excited to see where this will go. Three years ago we couldn’t have imagined this, and who knows what we’ll have in another three years.”

Kevin Yager

The Future of AI Assistance

CFN is committed to developing AI as an empowering collaborator for scientists, not just an efficient tool. As research incorporates more automation, important ethical discussions are arising around its safe and fair use.

But the possibilities seem endless. Yager’s chatbot hints at a future where AI language assistants could turbocharge scientific progress through effective information sharing. And scientists everywhere now have access to the code for building their own specialized AI sidekicks.

So harness this new technology for your next breakthrough discovery! With an AI assistant by your side, accelerated insights and innovations may be just a question away.

You can read all about their work in their paper published in Digital Discovery and you can check out the chatbot code yourself from their github repository.

Staff Writer

Our in-house science writing team has prepared this content specifically for Lab Horizons

Leave a Reply

Your email address will not be published. Required fields are marked *