OpenAI has recently released ChatGPT, a new chatbot that has the potential to completely reshape the way we work and learn. The University of Wisconsin-Madison is already putting it to use, helping materials engineers rapidly and affordably glean information from scientific literature. To enable this, Dane Morgan, Professor of Materials Science and Engineering at UW-Madison and his team are applying machine learning tools to their research.
Maciej Polak, a staff scientist working in Morgan’s lab decided to see what else AI could do for them. After some thought, he realised that material scientists spend a ton of time reading papers for data. Normally, this data is buried in long pages of text, but machine learning offers a much faster solution. It was decided that ChatGPT would be tasked with combing through papers sentence by sentence and decide whether each sentence contained the required data – this technique was also used to extract cooling rates for metallic glass.
However, Polak found that while the method they were deploying was working, it still wasn’t enough. He then refined the technique by prompting ChatGPT to double-check the information. Asking follow-up questions helped make sure the AI was providing accurate data and by the end, their accuracy rate climbed to 90%.
The success of this project culminated in the release of two papers on this topic, the first detailing their original technique, and the second detailing their refinement of the method. As mentioned before, AI isn’t a replacement for people, instead it allows them to work on projects that would normally have taken too much time or resources to pursue.
OpenAI is an artificial intelligence lab that is dedicated to the development of advanced and ethical AI. In the past, they have made huge strides in deep learning, natural language processing and reinforcement learning. ChatGPT was developed with the intention of helping researchers take on complex tasks without having to write hundreds of lines of code – now, these tasks can be performed by simply talking to an AI.
Maciej Polak is an experienced materials scientist and staff scientist at the University of Wisconsin-Madison. His groundbreaking work in machine learning has since been put to use by Morgan and his team, revolutionizing the way materials engineers work by increasing the speed and efficiency with which they can extract and analyse data. His most recent project revolved around helping ChatGPT assess the accuracy and relevancy of data tables presented by the AI. This refinement to the team’s technique made it possible for 90% accurate data to be collected from papers.
