How we proved that AI enables scientists to speed up literature search and data extraction
– AI tools enable researchers to find literature 78% faster and extract key information from hundreds of scientific documents in a few minutes. Read on!
Who doesn’t love getting a new phone, and the sensation of removing the protective film? It’s priceless…
Did you know that you’re probably experiencing ASMR, “a tingling sensation that typically begins on the scalp and moves down the back of the neck and upper spine?” In other words, you’re enjoying a low-grade euphoria 🤗
In fact, this sensation isn’t priceless. People crave novelty, which is why R&D is one of the most important functions in any company.
But with the booming availability of research and information, R&D teams need better tools to distinguish the noise from relevant knowledge.
R&D is changing
Let’s first get the facts straight: R&D is changing.
The number of published papers and patents grows exponentially. About 2 million papers are published per year, and the growth of published work is growing exponentially, about 4% annually for papers and 5-6% for patents.
R&D spending has reached a record high level globally, almost US$ 1.7 trillion. The biggest R&D spenders are the pharmaceutical and biotechnology industry, which spends on average 15% of its revenue on R&D, followed by software and computer services (10.6%).
Open access and decentralised knowledge networks make it easier for everyone to publish and access research. For example, arXiv.org, an open access website, enables researchers in AI and physics to publish papers before they’re peer-reviewed (a process of 9–18 months, after which point the paper’s knowledge is already out-dated).
R&D is changing, and whether the growth of published papers means actual growth in knowledge is disputed, but regardless it presents a challenge for all researchers who need to navigate all that noise.
That begs the questions, how can AI help R&D leaders drive innovation and free up resources?
Automate the manual work
Researchers and scientists spend on average four hours a week searching for literature and five hours reading papers and patents, time they could’ve spent developing new research.
If you manage a small group of 10 researchers, they spend on average 200 hours a week to simply find literature and read it. On top of that, they do their own research and identify competitive opportunities, the most important work.
Using AI, researchers can find more relevant literature, screen papers and patents 78% faster and extract data from hundreds of scientific documents in minutes. How does that actually work?
Find more relevant literature
To test how AI speeds up literature search, we worked with RISE (formerly Swerea), a leading Swedish research institute focused on composite materials. Two teams competed against each other to address the research question: “Can we build a reusable rocket made of composite materials?”. One of the teams used Iris.ai, whilst the other team used Google Scholar.
The results showed how the AI engine boosted one research team’s efficiency remarkably. The independent jury of RISE professors gave the team using Iris.ai a score of 95%, whilst the team using Google Scholar received a score of 45%.
How did the machine enable one of the teams to find more relevant papers? AI allows researchers to search for relevant literature by uploading an existing paper or using their own problem proposal in this specific research field. The AI machine finds the most important words in that paper, identifies synonyms and hypernyms, and matches those to documents in the databases. In return, researchers get all the papers relevant to their research.
You can see a 4 minute video of the experiment and peer-reviewed paper about the tech.
Screen thousands of papers and patents
AI tools also speed up the reading and categorization of scientific literature.
At Iris.ai we conducted an experiment where a group of researchers, led by Computer Scientist Christian Berger at Chalmers University, did a research mapping study of autonomous vehicles. Ahead of the experiment, the researchers had screened 11,000 papers manually. Two years later they decided to repeat the query using Iris.ai, which returned another 11,000 papers (research being published at eye-popping speed! 👀).
Manually analyzing the additional 11,000 papers would have taken months, so researchers used Iris.ai to categorize and filter relevant papers, and the results showed that the AI machine speeded up this process by 78%. In the interview with Nature, Christian Berger said that results provide “a quick and nevertheless precise overview of what should be relevant to a certain research question”.
Extract key data
When R&D researchers have found the most important papers and patents, they spend time extracting key data, which helps them understand the competitive landscape and decide R&D direction.
One of our clients, the Head of R&D at a major steel manufacturer, told us that one researcher spends a full month extracting data from 60 patents, and he asked us how AI tools automate this process.
In short, we told him that the AI machine identifies all the domain-specific data in text, tables and images, then exports it to a desired output format (CSV or any other application with APIs).
So instead of one month manually extracting data from 60 patents, the machine does it in 4 minutes, at 90% accuracy.
Faster extracting key data means researchers can focus on analysing the most relevant competitive information and develop new products, instead of spending time collecting data.
💆 Time, headspace and innovation
R&D is one of the most important functions of a company, but researchers spend a significant amount of time searching for and reading scientific literature — at the expense of creating products and driving innovation.
Artificial intelligence tools free up researchers time, which means they have more headspace to understand the competitive landscape and can identify early opportunities for new products.
Would you like a demo of our tools?