This article is based on a presentation at IBA’s Analytics Conference on Cybersecurity on November 11, 2022 given by Ergin Soysal, Senior Advisor – NLP, Eli Lilly & Company and Adjunct Assistant Professor at the University of Texas Health Science Center at Houston.
Natural Language Processing (NLP) is disrupting everything, according to Ergin Soysal: how businesses are made, how technologies are processed, how entities are extracted, and even what humans value and how they become successful.
Soysal, a senior NLP Research Scientist at Eli Lilly and Company, has devoted his career to unlocking meaningful data from biomedical texts after experiencing obstacles in consuming existing biomedical information during his clinical practice as a Medical Doctor. In his presentation “Harnessing the Power of NLP” during the Spring 2023 Analytics Conference on NLP/NLG, Soysal explored real-world applications of NLP and National Language Generation (NLG) for the pharmaceutical industry, including how NLP can help analyze complex medical data — and some of the wider impacts of this technology on human nature and values.
Natural Language Processing and Generation
NLP encompasses a number of different parts, including text analytics, text mining, information retrieval, and text classification. The text mining involves generating numeric values from a text by examining word count, word distribution, or patterns in the text, Soysal said, and tries to extract some insightful data from the existing text. For the text classification, a number of applications have been developed for NLP in the last 10 years, including sentiment analysis and stance analysis, which allows organizations to trace time to market, customer attitudes, and product success through social media and other reports.
NLG has been the new “hype” process for the last two or three years, according to Soysal. In NLG, users input precursor information like structured data or architectural form text, and the information is analyzed and returned to coherent text that can be understood by humans. NLG gives users quick processing capability in a huge balance of text, making it easier to determine the relationship between certain entities like diseases, medications, customers, and organization names that appear in a text.
Applications for Biomedicine and Pharmacy and the Megamontiliasis Example
In the pharmaceutical industry, companies like Eli Lilly can use NLP to perform text processing of documents from clinical trials, legal documents, and scientific literature for business management and product support. NLG can be used to perform many tasks involving generation of human language such as creating data driven reports to communications with customers, the government, and other branches within a company. On the other hand, natural language processing can also help extracting information from existing documents, and these information can be used to understand everyday business events.
“Information extraction is actually, by itself, just the beginning of everything else,” Soysal said.
Soysal gave an example of machine learning for a made-up disease called “megamontiliasis.” An imaginary patient was suffering from megamontiliasis, so a physician ordered a mogamin test, and when it received a positive result, the physician prescribed a low dose mogpinex treatment.
Readers can tell just by looking at the context that “megamontiliasis” is a disease, Soysal said. Similarly, machine learning looks at the context around certain words, makes “annotations,” or markings, to categorize each word, then learns patterns. In this example, the words “patient,” “was,” and “suffering,” indicate that “megamontiliasis” can be marked as a disease.
For NLP, the “name of the game” in machine learning is to guess the next word, Soysal explained. If a sentence is entered into a machine learning algorithm, the algorithm should be able to guess the next word, and then learns the correct pattern and is able to produce a coherent text. For pharmaceutical companies, machine learning offers powerful tools to process information without error.
“The value for this is that if we can digitally process, match and mark things, we can design more perfectly and make decisions without missing anything,” Soysal said.
NLG also has potential beyond text generation to link information from different sources for knowledge discovery, Soysal said. He cited a Gartner study that predicts that by 2025, more than 30% of new drugs and materials will be systematically discovered using generative AI techniques and ChatGPT-like technologies. This number is illogically high, Soysal said, but it proves that the system can take a number of complex sources and offer a concise digested knowledge source.
Looking Forward
As NLP models and technologies like ChatGPT advance, Soysal noted that they will change the way business is done and may make it harder for small and medium-sized companies to get the hardware they need to process data.
“These language models are growing faster and faster,” Soysal said. “Initially, it was possible for me to use a small model on my local computer, or even a larger model, and I was able to do experiments and prepare a draft algorithm. Now, it’s almost impossible since an average model needs a relatively larger server to fulfill model requirements.”
The growth of these technologies also means they can interact in a more precise, human-like way, which is something that Soysal said users should be wary of.
“That gives me the intuition that we are approaching the ability to build the human brain more and more precisely,” Soysal said. “Even Elon Musk, who was responsible for OpenAI, said that advances in ChatGPT are something we should be worried about. As we use this kind of aid, we might start losing our capability to generate writing, and we will continue to lose some values that make us human and make us successful people.”
He emphasized that it is critical to consider the ethics around NLP and NLG.
“In NLP, we have a solution for almost every problem,” Soysal said. “That gives us insight, a kind of tool to make things easier for us, but we must not just do this easily, but also wisely.”
Leave a Reply