opment of AI, including the Future of Life Institute, 3
IEEE, 4 The Royal Society, 5 and the Stanford AI100
project. 6 Governments, intergovernmental organizations, and nongovernmental organizations, including the European Parliament7 (Goodman and Flaxman 2017) and the International
Telecommunication Union, 8 are holding summits
and passing sweeping regulations. Clearly, the culture and law of ethical AI development will be enacted over the next decade.
Areas of beneficence, fairness, explainable AI, and
other aspects of AI governance will be a focus in
round two of the competition. We look to feedback
from our advisory board and judges to adapt the
competition guidelines to ensure the ongoing execution of a competition process that is fair to competing teams and maximally impactful in the real world.
Competing AIXP teams are at the forefront of eth-
ical AI development through their pursuit of $5 mil-
lion in prize money. Their efforts support the move-
ment with applications of AI that are beneficial for
humanity, that demonstrate human and machine
collaboration, and that identify the greatest oppor-
tunities for AI to make an impact on society. While AI
techniques are developing quickly, we have an
opportunity to better understand where research
intersects with grand challenge applications to pro-
duce new opportunities. An open competition plan
has allowed teams from many backgrounds to tackle
hard problems with AI. As the competition proceeds
to year two, the XPRIZE team, along with the prize
sponsor IBM and other supporting ecosystem part-
ners, look forward to seeing the good an impassioned
group of AI developers can produce in the world.
First and foremost, the teams competing to make the
world a better place deserve special recognition for
their efforts. Next, IBM has shown great vision in
supporting such an open-ended endeavor.
The IBM Watson AI XPRIZE relies on an advisory
board including Yoshua Bengio, Francesca Rossi, Rob
High, Babak Hodjat, Neil Jacobstein, Subbarao (Rao)
Kambhampati, Peter Norvig, Tim O’Reilly, Jean
Ponce, Lav Varshney, and Manuela M. Veloso.
The judges perform the hard work of balancing
imagination and critical review. They include Gabriel
Skantze, Carla Gomes, Eric Van Gieson, Adam Chey-er, Robin Murphy, Danah Boyd, Ivan Laptev, Bistra
Dilkina, Alex London, Al Kellner, Erin Walker,
Madeleine Clare Elish, Franc¸ois Chollet, Sidney
D’Mello, David Kale, Danielle Tarraf, Xiaoyang
Wang, Evan Muse, Nicolas Papernot, Henry Kautz,
Risto Miikkulainen, Pascal Van Hentenryck, Mark
Crowley, Forent Perronnin, Bill Smart, Graham Taylor, Julien Mairal, Stefano Ermon, Antoine Bordes,
Jonathan Zittrain, Michael Gillam, Peter Eckersley,
Barry O’Sullivan, and Rayid Ghani.
Finally, the XPRIZE staff members Jennine Dwyer,
Yvonne Cooper, Katherine Schelbert, Michael Martin, Sean Beougher, Daniel Miller, Stephanie Wander,
and Ed McNierney have all been instrumental in
organizing the IBM Watson AI XPRIZE.
3. See the Asilomar AI Principles ( futureoflife.org/ai-princi-ples).
4. Such as the IEEE Global Initiative on Ethics of
Autonomous and Intelligent Systems ( standards.ieee.org/
5. The Royal Society issued a report on machine learning in
2017 ( royalsociety.org/topics-policy/projects/machine-learning).
6. The AI100 Project, a collaboration of AI scientists, issued
a report in 2016 called Artificial Intelligence and Life in 2030
7. See the Council of the European Union, European Parliament, Regulation (EU) 2016/679 of April 27, 2016 (
8. The AI for Good Global Summit 2017, www.itu.int/en/
38 AI MAGAZINE
We live in a world where more scientific discovery is underway than ever before — but the research process is plagued
with hard-to-justify inefficiencies, and among them, the
growing need to distill and filter through all the noise. Interdisciplinary exploration is vital to new discovery, but exploring a new field where one is not a domain expert can be
immensely time consuming.
Aiming to build an AI researcher for literature-based discovery, Iris.ai semiautomates the time-consuming process of
literature review. Their “exploration and focus” tools reduce
the time required to go from a problem statement to a reading list by 90 percent, while also increasing interdisciplinary
The Iris.ai team is focusing on extraction of a research
paper’s key concepts, together with an encoding technique
that can construct a document vector space based on the
available information. This strategy allows the building of
intuitively meaningful content-based indexes. The team’s
next steps are developing hypotheses-extraction techniques
and word-to-word graph representations of documents.
Evaluation has shown a reduction in time for research
teams augmented with the Iris.ai exploration tool. In building the document vector space, their WISDM metric shows a
consistent speed-up, while upholding precision of comparable models.
For more information, see iris.ai.