A recent report by ZDNet highlights the increasing focus on generative artificial intelligence (Gen AI) as a significant driver for IT investments in the coming years. Tech analyst Gartner forecasts that global IT spending will reach $5.74 trillion by 2025, reflecting a 9.3% rise from 2024. This growth is attributed, in part, to organisations exploring the diverse applications of Gen AI in various sectors.
Generative AI encompasses a range of technologies that allow for the creation of text, images, and even code, with many businesses and individuals already experimenting with these capabilities. However, as James Fleming, Chief Information Officer at the Francis Crick Institute, discusses in an interview with ZDNet, leveraging these technologies in a scientific context presents unique challenges that diverge from typical business applications.
Fleming remarks that "doing scientific AI, as opposed to creating publicly available large language models, is quite a different discipline," emphasizing the need for precision in scientific work. Unlike general AI models, which may yield near-accurate results, scientific applications require a demonstrable accuracy under specific conditions. He states, "It's not good enough... Most of the time, it's got to be exactly right under certain conditions." The stakes are particularly high in research areas with significant real-world implications, such as healthcare.
The primary concern, as communicated by Fleming, is the importance of "provability" in the scientific realm, particularly when innovations are presented to clinicians. He explains, "If you're putting an innovation in front of a clinician and saying, 'I think this tool can predict the evolution of cancer,' for example, they're going to say, 'Can it? Show me why.'" This points to a fundamental necessity for both accuracy and transparency when employing AI technologies in research settings.
To counteract the pitfalls associated with the black-box nature of many popular AI models, the Crick Institute has adopted an iterative approach to integrate AI into its research processes. Fleming highlights the importance of building "provenance and trust from day one," stressing that progress must be gradual and systematic. This methodology has already yielded advantages in enhancing existing scientific practices, notably in the field of microscopy, where AI aids in turning complex images into actionable data.
By undertaking a methodical range of tests and analyses, researchers are able to derive meaningful insights from AI-driven processes, such as determining cellular differences between cancerous and non-cancerous cells. For instance, Fleming notes a research project focused on Parkinson’s disease that involved a classifier designed to identify afflicted patients in stem cell populations. After discovering that the model couldn't initially explain its findings, the team employed a reverse interrogative strategy to elucidate the results: “the dominant thing is the ellipticity of the cell... they’re more oval,” which subsequently directed further investigations.
Currently, the Crick Institute is advancing its AI efforts through detailed, multi-layered models to boost the reliability of research outcomes. One prominent endeavour led by Samra Turaljic at the Cancer Dynamics Laboratory involves predicting the genomic evolution of kidney cancer. By leveraging historical genomic databases alongside advanced pathology imaging, the team has created predictive models that integrate data across a decade of research. Fleming underscores that this extensive framework legitimizes their results and enables real-world applications: "The result is you create something that can clinically predict the evolution of the kidney."
Fleming's insights suggest that the efficacy of AI technologies in research and business applications hinges upon a measured, strategic approach. By embracing this iterative process, organisations can work towards unlocking the potential of AI while ensuring accuracy and reliability in their findings. In a landscape where rapid AI advancements are touted by vendors, the value of understanding data in its early stages and refining models incrementally remains a crucial tenet for driving innovation.
Source: Noah Wire Services