ChatGPT Improves Efficiency in Materials Science | Technology Advancements
OpenAI has launched its new AI-based chatbot called ChatGPT, which is designed to improve the way people work and learn. The University of Wisconsin–Madison is already deploying ChatGPT to search scientific literature for faster and more cost-effective extraction of information for materials engineers.
Professor Dane Morgan at UW–Madison pioneered the use of machine learning, a type of data-based AI, in his lab for evaluating and searching for new materials. Staff scientist Maciej Polak brainstormed ideas for other tasks that AI could help with. One crucial task is reading papers to acquire data.
Materials scientists commonly download and scour long research papers to find specific data to add to their datasets. Polak sought to make this process more efficient by using AI to extract critical information on relevant data. Chatbots like ChatGPT can extract data from papers sentence by sentence and then decide whether the sentence contained the relevant information. The bots then present the data in table form, which can be checked by human researchers.
According to Polak, this technique reduced the researchers' paper-reading workload by 99%, but he is interested in making further improvements. The team is working towards fully automating the process by engaging in 'prompt' engineering, which will allow the bots to extract and then double-check the information in their dataset. The technique does not require much effort or in-depth knowledge. Polak said that previously, people needed to write hundreds of lines of code to do something similar, and the results were often unsatisfactory.
It's important to note that integrating AI into research does not replace graduate students and scientists. Instead, these tools could enable researchers to undertake projects they didn't have the time, money or personnel to handle. Morgan believes that the new technology will transform how research is done, similar to how Google search changed the way people gathered information. Today, people explore a field using search tools and read papers and resources to extract data, but they can now collect data and information from one of these large language models in hours rather than weeks.
The AI community looks forward to the further development of ChatGPT and other AI-based tools that will help accelerate research and discovery in various fields.
