Igor Osipov, Ph.D., is VP Academic & Institutions at scite.ai. He leads scite institutional engagements globally in academic markets and consolidates the work of its channel partners and resellers networks worldwide.
Igor brings significant experience in open science research, relationship management, and business development in the academic and publishing industry. He previously worked as VP Academic & Government EMEA for Digital Science UK and CEO of Digital Science Eastern Europe for six years, helping to launch Dimensions, and was Regional Managing Director for Elsevier Science & Technology for five years. Igor completed his Ph.D. in social sciences & decision-making at the University of Alberta, Canada, M.A. at the University of Sussex, UK, and BA at RGGU (Moscow) / University of Alaska, Fairbanks.
Most recently, he served as VP & Advisor of Global Academic & Channel Partnerships for Writefull, successfully launching and growing Writefull’s university engagements and channel partnerships network, and has been advising several projects in IT, AI, Healthcare, and Philanthropy sectors.
Q: What is the extent of scite’s contribution in progressing from a literature review based on individual knowledge to a semi-automated bibliometric approach and ultimately to a fully automated method? Also, what is the researchers’ role in interpreting the output generated by these automated processes?
IO: Our approach has been rooted in human researchers’ perspective rather than relying solely on generative AI. Since scite’s launch in 2018, our focus has been direct collaboration with publishers and data custodians to establish a robust foundation of research. We believe validated scientific data should serve as the core of our system before adding features involving LLMs or other generative AI. With over 185 million science-related documents, including peer-reviewed articles, books, commentary, and others, our platform forms a massive foundation to build various search and discovery systems on top of, including potentially a place for automated literature reviews. We believe our approach and data can address three significant challenges:
- While large language models excel in creating well-structured texts, sometimes those texts simply do not make sense in terms of factual and scientific coherence.
- Hallucinating citations generated by LLMs are a major concern for their use in research.
- LLMs are limited to the dataset they were trained on, and therefore their “knowledge” stops at a specific point in time. Inevitably, it leads to factually incorrect information in otherwise well-written texts.
Transitioning from a structured bibliometric approach to a fully automated literature review seems imminent. However, even in its current stage, full automation may still exhibit biases based on the references chosen. We firmly believe that AI, assistants, and generative systems serve as valuable aids to researchers rather than replacements. Human oversight is essential as these tools, despite their intelligence, require direction. They augment human efforts, enhancing efficiency while necessitating human guidance. At scite, we believe that human creativity and originality are something that LLMs will probably always struggle with to a certain extent.
At scite, we strive to create tools that empower human researchers, emphasizing human control in directing the course of AI advancements. Ultimately, we envision AI as a vessel to optimize human efforts, acknowledging that humans remain at the helm of innovation.
Q: What’s your take on the current and future policies of scientific journals for acknowledging the use of AI tools in research papers (authorship, reliability, training process based on AI-generated texts, etc.)?
IO: Publishers increasingly employ AI tools to streamline processes, aiding both themselves and authors, particularly those not fluent in English. While a paper might hold scientific merit, concerns arise regarding its structure, language quality, and flow. Major publishers are integrating free author-centric tools to refine output quality without altering scientific methodology or findings. This shift, evident in several leading publishers, denotes the industry’s rapid advancement.
However, ethical concerns surface as AI-generated texts with misleading references emerge, prompting a need for regulation and ethical use. This is something that publishers are rightfully concerned with.
Q: Considering the evolving landscape of technologies such as artificial intelligence, digital platforms, and advanced citation management software, how do you foresee the formatting of research papers changing in both the near and long-term future?
IO: Referring to Josh Nicholson, Co-Founder of scite, in our article “The Future of Citations: Displaying Citation Statements Natively on the Version of Record,” we have reached a stage where all types of citation statements can be categorized. This advancement enables immediate recognition of supporting or contrasting evidence when referencing something. While this process is significantly faster than manual efforts, accessing the full context of citation text still requires a separate visit to the scite report.
Our vision for the near or not-so-distant future involves integrating this information directly into the article or book being viewed. We aim to enable users to hover over a reference within the record and instantly see an excerpt of the citation statement. This would provide immediate insight into why and how a paper or any key phrase or concept is referenced in the broader literature. We are actively collaborating with publishers to implement this feature, which we believe will revolutionize researchers’ experiences by providing quick access to crucial information without the need for separate reports. This streamlined approach is poised to significantly benefit researchers, offering prompt insights without additional steps.