LANA FENG, HUMA.AI
According to a recent survey conducted by Huma.AI, the majority of medical affairs leaders and professionals see potential for Chat Generative Pretrained Transformer (ChatGPT)-like technology to be applied at life sciences companies to improve efficiency and accuracy in a variety of functions. In addition, they believe a generative AI platform could improve patient outcomes in health care settings by gathering unstructured data from disparate sources — such as details from an amalgam of medical images — to ultimately assist in assembling treatment strategies. As such, while AI should never be seen as a replacement for human judgment, it possesses the power to enhance the use of scientific information along with medical care decision-making when it is applicable to patient health and treatment.
This survey is beneficial because for the first time in history, consumer and enterprise needs have converged significantly. Many scientists already use ChatGPT at home for personal reasons such as assisting children with homework. Because of this, they have firsthand knowledge of the functionality of ChatGPT, and it is not a surprise that they want a similar experience while working with their professional resources. Considering its role in biomedicine, GPT can analyze images that alpha testers of OpenAI products have used. Through such a process, amazing new products that can read and interpret a variety of graphics, with high accuracy and at scale, have been discovered.
Anyone who has used ChatGPT would agree that the specificity of the question affects the comprehen-siveness of the AI interface’s response.
This is not to say that ChatGPT-like technology doesn’t have its limitations. It is vital to have an expert engaged when the data is gathered and reported. Anyone who has used ChatGPT would agree that the specificity of the question affects the comprehensiveness of the AI interface’s response. Secondly, getting to the best answer takes multiple iterations of modifying the way a user asks the question. This is referred to as prompt engineering, and experts typically ask more accurate questions because they have the domain knowledge and an understanding of the data.
The context is also important for receiving more accurate answers. The quality of the data directly affects the generated intelligence. However, the accuracy that can be obtained from this technology has improved in recent years. In short, GPT4 is superior in accuracy to GPT3.5, while GPT3.5 is more accurate than GPT3.
The recent models are more precise because they have passed professional exams, such as USMLE, with increasing proficiency. Huma.AI conducted among its clients an independent side-by-side validation against the current gold standard, which is a manual curation of the same data by domain experts. The results showed that an “expert-in-the-loop” approach has achieved 97% to 98% accuracy.
Furthermore, approximately half of those in the Huma.AI survey said they thought their life sciences companies would use generative AI within the next two years, while several have already incorporated it into their business approach. In fact, almost every large life sciences organization is interested in bringing in a form of generative AI system.
The most pertinent discussion within these companies is whether the system should be built or bought. While it is reasonable to ponder this choice, it is a significant undertaking to build AI solutions in-house, particularly when speed is considered alongside relevant cost factors. With that said, generative AI is evolving at breakneck speed; therefore, it can be cost-effective to purchase such a system from those with expertise in such platforms.
Time is of the essence for this technology. The potential for guidance in diagnostics is not only powerful but also critical to expediting drug development, given that generative AI platforms can work in real time. Knowing that the breakneck pace of generative AI has quickly become the norm, it is incumbent upon life sciences professionals to embrace its immediate effect and partner on a strategic vision.
Meet the author
Lana Feng, Ph.D., is cofounder and CEO of Huma.AI, a leading generative AI platform. She has over 20 years of experience in biotech and pharmaceuticals, focusing on precision medicine. Feng worked with Novartis’ Oncology Business Unit and built the BioPharma division at Genoptix. She has a doctorate in molecular, cellular, and developmental biology from Indiana University Bloomington.
The views expressed in ‘Biopinion’ are solely those of the authors and do not necessarily represent those of Photonics Media. To submit a Biopinion, send a few sentences outlining the proposed topic to [email protected]. Accepted submissions will be reviewed and edited for clarity, accuracy, length, and conformity to Photonics Media style.