Preface: In the rapid expansion of AI tools over the past year, many scientists have seen the benefit of asking AI tools like Iris.ai to help with research, though questions abound on how and when to use them. We asked Anita from the EU-based Iris.ai about how their machine is trained specifically for helping researchers, and learned how working with an AI that emphasizes accuracy is crucial for science, and also about how Iris works to remove bias when exploring the scientific literature. This interview is part of our ongoing series on innovative companies that support researchers!
Can you tell us a bit about Iris.ai and how the company came to be?
Iris.ai was founded in 2015 due to its founders’ understanding of the importance of science and a belief that if people had access to all of the world’s scientific knowledge and its implications at their fingertips at any time, it would radically speed up scientific innovation.
To that end, Iris.ai is a comprehensive platform for academics and researchers’ processing needs. Based on a collection of smart language and image processing algorithms together with clever engineering, Iris.ai’s platform can provide users with context-based and exploratory search, a wide range of smart filters, reading list analysis, auto-generated summaries, autonomous extraction, and systematising of data. These features work cross-disciplinary across all research and more complex areas of science. All of these features amount to saving researchers vast amounts of time when looking to find the right research, insights and data – up to 90% in some cases.
How does Iris.ai help to increase access in science?
Iris.ai has a chance to help researchers process the abundance of knowledge available today. With more than 7,000 research papers churned out daily as of 2018, there’s so much undiscovered knowledge out there which could be mission-critical for researchers of all levels.
Using smart language and image processing algorithms combined with engineering, Iris.ai’s platform helps to make this knowledge overflow more manageable rather than spending hours trying to find the right article. Even then, researchers must spend more time analysing the papers and extracting the necessary data. Saving time in identifying relevant research papers is key and ensures that researchers can spend their time on what matters, drawing the information from papers. Additionally, the platform ensures that researchers don’t miss a paper that may be crucial due to human error. All of this amounts to greater access to science.
Is there an ideal researcher that Iris can assist – individual students writing theses, group leaders positioning their 5-year research trajectory, scientists in big interdisciplinary consortia?
Iris.ai’s platform straddles a number of different use cases, but the biggest focus is on supporting industry research and development teams. This is in a range of industries, such as chemistry, material science and food safety.
As the tools of the Researcher Workspace can be adapted to the workflow of each researcher, it can be used for a variety of research processes. Fundamentally, they are all based on the same principles – the tool has to understand the context of what the researcher needs, but the examples vary. Examples include:
· A new researcher, or a researcher new to a given field, can use the Explore tool to broadly explore the topics they are to research – to gain vocabulary, good example articles and be able to draw conclusions about the direction of their research before diving in.
· A researcher who is looking for research within a specific context, which is easy enough to explain to a colleague but very hard to put into a boolean keyword query, can load their content collection and apply one or several self-written context filters to narrow down their reading list to only articles describing their context, not necessarily in the same words.
· A researcher with a document collection of hundreds or even thousands of documents filled with experiment data can use the Extract tool to unlock the data from text and tables and systematise it into a database or spreadsheet so they can take it to the lab or further analysis – fully automated.
So far, researchers have been humans interpreting data the best they can, but we know there is bias within the scientific literature—how does Iris deal with this when helping me learn?
Partiality exists in scientific research in any field, and Iris.ai’s Researcher Workspace helps researchers tackle these biases to find all relevant papers. There are two ways that the Researcher Workspace overcomes these biases within the scientific literature. Firstly, Iris.ai’s platform never uses a citation system in its algorithms. This means that the results and recommendations generated by Iris.ai’s tool do not depend on how many times others have cited a research paper. Instead, Iris.ai’s algorithm looks at papers’ content to deliver research papers that directly apply to a user’s field.
Secondly, Iris.ai’s Researcher Workspace treats all papers equally, regardless of source, whether from a leading or lesser-known university. As a result, Iris.ai is identifying research papers that may not have been on users’ radars. These two features of Iris.ai’s Researcher Workspace are helping researchers remove bias in the handling of scientific literature, supporting them with more informed, accurate and innovative research projects.
Can Iris tell me if there’s disagreement within my new research field or the evolution in the thinking in a field over time?
Currently, the Researcher Workspace provides users with all the relevant research papers in a given scientific topic based on the users’ query, and with the saved time the user can spend more time evaluating disagreements or evolution. However, with time the Researcher Workspace will be able to support users with spotting disagreements within new research fields and developments in scientific arguments and research as well. Our roadmap maps out the steps to get to this point. These steps include developing features for hypothesis extraction and verifying facts to further build on the Researcher Workspace. Iris.ai was made for the scientific community and researchers, and with that in mind, we are taking action to create a tool that meets this group’s specific needs.
It seems like working with AI assistants is becoming more common—has Iris been brought into the classroom to assist new generations of researchers?
Iris.ai is at the forefront of helping the next generations of researchers. Academic libraries can help post-doctorate and university students understand and wade through vast academic literature. Yet, academic libraries often receive little funding, so integrating AI assistants can be a challenge.
To that end, Iris.ai wants to nurture the next generation of researchers by giving them a top-down view to understand what research areas in their field need more attention, helping them to focus on reading relevant papers that matter to them. That’s why we provide subscriptions for individual researchers and students with up to 75% discount. Much like academic libraries, these next-generation researchers’ purse strings are tight, so we offer our tool, the soon-to-be-launched Researcher Workspace 1.0, at a much lower cost for an individual subscription.
Science is so international, is it possible for Iris to assist researchers in languages other than English?
In its current state, Iris.ai’s machine is set up only for English, the dominant language of scientific research, with 70% of papers published in English. However, we are looking to expand the language offering in line with the needs of our users.
Can you tell us about your vision for the future of Iris.ai?
Headline after headline is filled with words such as ‘ChatGPT’, ‘LLMs’ and ‘Bard’. We are in the midst of an AI revolution where organisations are trying to make AI-powered search engines work for everyone. Yet, what’s missing is factuality, as the current approaches are better at making LLMs’ results convincing and plausible-sounding rather than fully accurate. We see Iris.ai as playing a fundamental role in tackling this issue, enabling generative AI responses to be both plausible and factual.
As part of this, we aim to create a Science Question-and-Answering machine, a generative AI chat interface for scientific research. It is a tool that incorporates the impressive fluency of LLMs and ChatGPT but doesn’t compromise on accuracy when it comes to scientific facts, and there are no hallucinations. A tool where anyone with a research question can ask it of the machine, and in a conversational way figure out the currently known answer to their question – where the facts are based on the global collection of scientific output.
We would like to thank Anita for sharing her insight, and congratulate the Iris.ai team on winning the EIC!