ADVANCED MATERIALS & PROCESSES | MAY/JUNE 2023 10 MACHINE LEARNING | AI ChatGPT SPEEDS MATERIALS RESEARCH Researchers at the University of Wisconsin-Madison are using OpenAI’s ChatGPT AI language model to quickly extract information from scientific literature. MS&E professor Dane Morgan has used machine learning for years in his lab to search for new types of materials with great success. Now, staff scientist Maciej Polak is brainstorming with Morgan about other tasks AI might help with. Polak knows that materials scientists often comb through long research papers to search for one small group of numbers to add to their datasets. “We thought we could just offload all of these time-consuming tasks onto an AI that could read those papers for us and give us that information,” says Polak. Asking chatbots, even ChatGPT, to look for and extract data from the full text of a paper remains beyond their capabilities. So Polak refined the technique, asking the bots to review sentence by sentence and decide whether each contained relevant data or not, a task that boiled papers down to one or two key sentences. He then asked the bots to put the information in a table, at which point a human could review it. The technique yielded an accuracy of 90%, allowing researchers to extract data from a set of papers to create a database on the critical cooling rates for metallic glasses. While the technique reduced the team’s paper-reading workload by about 99%, Polak wanted to refine it even more. So the scientists engaged in “prompt” engineering—figuring out questions that would cause the bot to double-check the information it pulled. They applied this approach to the extracted data table, and then asked the bot a series of follow-up questions to introduce the possibility that the dataset was wrong. That forced the AI to recheck the data and flag mistakes. In the vast majority of cases, the AI was able to identify faulty information. “Asking the program to extract data and then asking it to check if it is sure with normal sentences feels closer to how I train my children to get correct answers than how I usually train computers,” says Morgan. “It’s such a different way to ask a computer to do things.” He emphasizes that integrating AI into research does not replace graduate students and scientists. Instead, AI could allow researchers to pursue projects they haven’t had the time, money, or people to undertake. wisc.edu. USING AI FOR IMAGE ANALYSIS Researchers at the DOE’s Oak Ridge National Laboratory, Tenn., Graphic courtesy of text-to-image generator Stable Diffusion, using the prompt “researchers working with huge piles of data.” AI-generated image representing atoms and artificial neural networks. Courtesy of ORNL. developed a new software package inspired by machine learning to provide end-to-end image analysis of electron and scanning probe microscopy images. Called AtomAI, the software applies deep learning to microscopy data at atomic resolutions, thus providing quantifiable physical information including the precise position and type of each atom in a sample. Researchers can then quickly derive statistically meaningful information from immensely complex datasets. These datasets routinely include hundreds of images that each contain thousands of atoms and abnormalities in molecular structure. This improvement to data analysis allows researchers to engineer quantum atomically precise abnormalities in materials, and can be used to gain deeper insights into the materials’ physical and chemical qualities. AtomAI is also built to reduce errors in image processing by accounting for unintended changes in the image data, such as images of non-target materials, and by incorporating certain unchanging physical characteristics into the model. ornl.gov.
RkJQdWJsaXNoZXIy MTYyMzk3NQ==