Developing new materials is a time-consuming and labor-intensive process for chemists. However, there is hope that artificial intelligence (AI) could help alleviate this burden. A recent study published in the Journal of the American Chemical Society explored the use of AI to search scientific literature, with the goal of extracting relevant information for predicting experimental results.
In the past, researchers have attempted to use AI for this task, but they faced challenges in terms of technical expertise and adaptability to new topics. The team led by Omar Yaghi wanted to see if the latest language models, such as ChatGPT, could provide a more accessible and flexible solution.
The researchers used ChatGPT to analyze text from scientific papers by providing it with prompts to guide its search for experimental information. They carefully designed these prompts to minimize the model’s tendency to generate incorrect responses and ensure the best possible results.
Testing the system on 228 papers that described the synthesis of metal-organic frameworks (MOFs), the AI extracted over 26,000 relevant factors for approximately 800 compounds. Using this data, the team trained another AI model to predict the crystalline state of MOFs based on these experimental conditions. They also created a user-friendly chatbot to answer questions about the data. Importantly, this system does not require coding expertise and can be easily adjusted to focus on different topics by modifying the prompts.
Dubbed the “ChatGPT Chemistry Assistant,” this new AI tool has the potential to benefit other areas of chemistry as well, according to the researchers.
Key Words: artificial intelligence (AI), chemists, materials, scientific literature, ChatGPT, language models, experimental results, metal-organic frameworks (MOFs), synthetic materials, clean energy, predictions, accessible, coding, user-friendly, ChatGPT Chemistry Assistant, chemistry.