How AI and Machine Learning Can Improve Scientific Data Handling

Listicle How AI and Machine Learning Can Improve Scientific Data Handling Neil Versel Artificial intelligence and machine learning (AI/ML) have become widespread in scientific research. A subset known as generative AI (GenAI) that can produce detailed text and images with just a few human prompts was thrust into the general public’s consciousness with the November 2022 release of ChatGPT. A year later, ChatGPT became the first nonhuman entity to make the journal Nature’s annual list of the 10 most important contributors to science. A 2023 survey by the same journal found that 30% of postdoctoral researchers were employing AI chatbots to generate and edit code, manage literature and refine the text of their scientific papers. More recently, an Elsevier survey of corporate scientific R&D professionals indicated that 96% believe AI will accelerate knowledge discovery and 93% expect AI to lower business costs. More than 85% said that these technologies would improve work quality and free up time to pursue higher-value projects. How ever, similarly high numbers expressed concerns about misinformation, errors and reduced critical thinking. GenAI is merely one type of AI/ML, though the lines between GenAI, predictive AI and machine learning — that must be trained on specialized datasets — have been blurring. While forecasts can vary widely, the AI/ML field as a whole has been exploding. A 2023 analysis suggested that the global AI market share would grow by about 19% annually to more than $2.575 trillion by 2032. Another analysis pre dicted a 14% annual growth in the market size of AI just for clinical trials, to $4.4 billion, in 2023. Nvidia, which makes high-performance computing chips that fuel AI development, briefly surpassed Microsoft and Apple in June 2024 as the world’s most valuable company based on market capitaliza tion. The technology is not without growing pains, however, as evidenced by an Nvidia sell-off that shaved $500 billion in value less than two weeks later. Another Nature article suggested that ChatGPT may be “corrupting” scientific peer review. In a Harvard Data Science Review study, researchers found that ChatGPT creator OpenAI’s GPT-4 large language model became less accurate between March and June 2023, while its supposedly less advanced pre decessor, GPT-3.5, improved. As developers work through these issues, scientific researchers are finding all kinds of applications for AI/ML. This listicle will explore how AI/ML is being used in different areas of scientific research to accelerate discoveries and improve efficienc 1