Scientific Research in the Age of Artificial Intelligence
You should probably be thrilled to be alive these days.
Every week brilliant researchers present new breakthroughs in scientific papers that extend our understanding of who we are (groundbreaking accuracy of CRISPR/Cas9 genome editing), how the universe began (the recent discovery of gravitational waves), and what computers can learn (phenomenal advances in deep-learning techniques).
Yet, as a regular consumer of scientific research, I’m frustrated that the current knowledge comes in puzzle pieces: a research paper presented in a conference here and another recommended by a friend there. Existing tools aren’t particularly helpful when it comes to solving this problem. Finding relevant academic studies can take days, sometimes even weeks as Google Scholar, Microsoft Academic, and other existing tools give millions of results, the vast majority of which are not valuable to your search. This reflects on the low utilization rate of current research. One study estimates that up to half of all studies are only read by their authors, editors, and peer reviewers.
We will be far from harnessing the full potential of academic studies as long as navigating the vast body of that knowledge feels like running an ultramarathon. Luckily, the exponential growth of digitally available data and advances in cloud computing allow us now, for the first time in history, to build machines that can help us overcome this challenge.
My co-founders and I tapped into this opportunity by building an artificial intelligence, Iris, to make navigation of scientific content fast and easy. Our long-term goal is to build a tool capable of highlighting new trends and connections between discoveries to unleash the full potential of scientific knowledge for innovation.
The first version of our tool leverages neural networks-based algorithms. This approach refers to algorithms that mimic the ways in which neurons and synapses in the brain reformulate when fed with new information. It allows us to develop a tool that can find contextual meaning in extensive texts, including research papers.
Our journey towards the big goal started with TED Talks. We just launched the first version of our tool at https://ted.iris.ai. Iris provides a visual shortcut to 2 million open access research papers related to over 2,000 TED talks, allowing the global TED audience to explore research related to the talks from big picture overviews to in-depth detail covering a multitude of research fields.
Here’s how it works:
For now Iris relies fully on computer intelligence and high quality texts — the full body of over 2,000 TED talks. In the future, to keep up with her impressive rate of learning she needs to read a lot more scientific studies and start accepting inputs from additional sources. She will also need to learn from people sharing our passion for making science more accessible. Join our AI fellowship program if you’d like to help us do that!
With Iris, a process that could take several days can be completed in a matter of minutes. Using any article as an input helps users bypass the need to know key terminology when performing a search. Iris also helps people manage the risk of tunnel vision by mapping connections across currently siloed research disciplines.
If we succeed, anyone interested in building a new scientific venture will be able to easily navigate the full extent of existing research discoveries and connect the puzzle pieces of science to put it to good practice.
