Shashi Mudunuri is the founder and CEO of Research Square, a company whose mission is to “make research publishing faster, fairer and more useful.” He has degrees in Computer Science, Philosophy, and Business from Duke University.  Shashi recently spoke with IPR License about the role that artificial intelligence (AI) will play in research and scholarly publishing. His interview was conducted in conjunction with his keynote address at the annual International STM Spring Conference in Washington DC.


There seems to be substantial misunderstanding of what artificial intelligence (AI) really is. What is it exactly?

There are two types of AI: General AI, which is a purposeful mind; and Narrow AI, which is a single-purpose machine that solves one specific task. Think of Narrow AIs as masters of hierarchical or layered pattern recognition.  Narrow AIs are being used in companies today to help humans to do their work better, faster, and cheaper.

General AI requires that the computer pass the Turing Test, a test for intelligence in a computer, which requires that a human being be unable to distinguish from hearing a conversation which is the human and which is the AI.  General AIs that express intentionality won’t exist for decades.


Can you give some examples of Narrow AI that are currently in use?

Progress in AI is moving blazingly fast. This is true in the financial technology and automotive sectors and everything in between. The current annual investment in AI is over $5 billion. AI has toppled champions of games like Jeopardy and Go. A recent Science article reported that AI beats doctors in predicting heart attacks.

In medicine, radiology is another good example. What’s the function of a radiologist? Looking at an image. This is a very rich data space. You can clearly define the outputs in the training data. In other lines of work that are highly original or are not data rich, you cannot train a machine to perform the task in deep learning.  An example is the play “Hamilton.”  You could never create that play just by inputting the history of Playbills into it.


Can you explain how AI works?

Narrow AIs use deep learning, or artificial neural networks, to learn how large data sets map inputs to generate desired outputs; they require large amounts of computing capacity to do it. AIs can create layers of understanding based on a large number of variables.  Humans can make decisions based on a handful of factors.


What is necessary to make AI successful?

Self-driving cars are an example. Self-driving cars require a great deal of historical data to make smart decisions, as well as present data provided by sophisticated sensors.  They also require an incredible amount of computing power that will partially be provided by the car and partially by the cloud. The prevailing wisdom 15 years ago was that we needed to upgrade our roads with sensors and communication to tell cars where to go and how fast.  It remains to be seen whether cars can be self-driving without those infrastructure changes.


There is widespread belief that AI will disrupt labor markets and lead to substantial unemployment. Would you please address this concern?

Most of the world’s AIs will be Helper AIs, meaning they will help people perform their work rather than replacing human labor.  These Narrow AIs take on a few particularly time-consuming tasks for humans, and automatically suggest conclusions based on their mountains of training data.  The human will then decide whether to use the suggestion or not.  In time, some of these Helper AIs will turn into Worker AIs. It will take another five years to get to that point when they will be much cheaper and faster than a person.


How will AI affect scholarly publishing? What should publishers be exploring?

Finding the right articles to read: this is a hard place to compete. Google is already there. They have a lot of data and have more machine learning (ML) talent than anyone else. However, identifying reviewers, detecting plagiarism and figure manipulation are all good segments of the workflow process to explore. Measuring researcher impact and effectiveness is another promising area. AI will be a better predictor of researcher impact than the Science Citation Index impact factor calculation very soon.

Author and customer service is another important area. We really don’t do a good job at rejection letters, for example. Creating and formatting guidelines, virtual reality (VR) and augmented reality (AR) science conferences come to mind as well. These will all raise the standard of research communication.


Are there any other observations you would like to share?

Will you be an investor or partner in AI?  It’s still a big investment, but If you don’t then you could miss an opportunity. In five years, AI will be a part of the fabric of every institution. We are racing toward a world where AI will converge with advances in Robotics, VR and AR, and the Internet of Things (IoT) to create an incredible world of opportunities.


pv-gantz-042513Paula Gantz ( consults for learned societies in the U.S., Europe and China. Her focus is on new products, new technologies and innovative business models. She has over 35 years experience in scientific and consumer publishing and an MBA from The Wharton School.

Paula Gantz Publishing Consultancy