Future of scientific research with artificial intelligence

More than 3,000 research papers are published every single day (!) by researchers around the world. Yet, only a disturbingly small portion of them is put into practice or ever even read. These papers, and the connections they make, hold a great potential for new discoveries.

Wouldn’t it be neat to have a brain that could read and understand all that information? There’s this startup, Iris AI, that works to fix this problem by building an artificial one. This Singularity University backed team that wants to build an AI capable of highlighting new trends and interconnections of discoveries. This could help researchers in startups, corporations and research institutes to implement what is already out there.

I first came across them via TED where they had processed my 2011 (gosh a lot has happened since) TED talk into their awesome system. 

Resulting in this science-cloud.

One of the first versions of Iris was just tailored to our e-health conference, Our Future Health. My guess was this could hugely benefit not only for researchers but also conference attendees. We started exploring with the Iris-team.

This baby AI, as the team calls it tenderly, provides a visual shortcut to millions of open access research papers related to the different talks delivered at the conference. The tool enables the audience to navigate a multitude of research fields with talks that they think are the most fascinating as the starting point. It adds a totally new dimension to scientific conferences, with attendees now having access to a permanent digital resource to explore topics further and deepen their knowledge both during and post conference.

Now, let’s scrutinize the results provided by this young AI science assistant. Take the talk delivered by the great design thinker Job Vogel, for example. Iris [1] got most of the topics right, although the tool today still struggles with elements such as anecdotes, metaphors or ironic remarks built within scripts. The concept extraction and paper matching displayed already add value to users, despite the big room for improvement through additional ongoing training efforts.

The team aims to optimize their machine learning algorithms through a combination of unsupervised, supervised and reinforcement learning techniques. You can join their AI training program signing up here [2].

So, could it be that we’re finally on the brink of pulling science out of dusty drawers into the real world? That remains to be seen. However, in the meanwhile, as we wait for this baby to grow up and scale, do check out all the result maps at ourfuturehealth.org. They should be there by early next week!

See their great talk at our conference :

One of the next stops we want to try is feeding the transcribed text automatically into Iris.ai and exploring the use within our Radboud University Medical Center, so stay tuned !!

[1] https://api.iris.ai/shared/6800/
[2] https//iris.ai/join/