Professor Mary "Missy" Cummings - "The Future of Artificial Intelligence and Logistics"

 

PROFESSOR MARY "MISSY" CUMMINGS
Professor, Duke University and Director, Humans and Autonomy Laboratory

The Future of Artificial Intelligence and Logistics

Mary "Missy" Cummings, Professor at Duke University Pratt School of Engineering, the Duke Institute of Brain Sciences and Director of the Humans and Autonomy Laboratory and Duke Robotics, began her career as a naval officer and military pilot from 1988-1999, and was one of the Navy's first female fighter pilots. She is the co-chair for the World Economic Forum's Council on Artificial Intelligence and Robotics, as well as a member of the U.S. Department of Transportation's advisory committee for autonomous transportation. She will outline how to conceptualize and leverage the strengths and limitations of both humans and artificial intelligence such that humans harness the raw computational power of computers, while applying inductive reasoning for potentially creative, out-of-the-box thinking for the increasingly complex logistics challenges of the future.She led the conversation with a collection of congressional staff members who work on the issue. She researches human-unmanned vehicle interaction, human-autonomous system collaboration, as well as the ethical and social impact of technology.

Professor Cummings was a keynote at the EDGE 2017 Conference held in Atlanta, Georgia, and CSCMP recently reached out to her inquiring about what she has been working on since September 2017, and asked her to share her opinions on recent news involving autonomous vehicles:

You were a keynote at CSCMP’s EDGE 2017 Conference in September 2017.  What has been the focus of your work since then?

My lab is still very much looking at the intersection of humans and complex systems. Recent projects we have started include developing a drone detection system that can help prison wards detect incoming drones trying to drop contraband. We are also working on developing models of remote control centers for driverless cars.


Have there been any noteworthy advancements in the regulatory issues regarding drone usage since your keynote address at EDGE 2017?

There have not been any major changes in drone regulations since then, but there may be some new ones soon.

The supply chain industry has seen some advancements in drone package deliveries in recent months.  Can you give an update on drone technology overall and what you expect to see in terms of drone advancements in the next 6 months?  12 months? 

Drone businesses, in terms of those building new platforms, have been struggling, with most businesses going to services (i.e., imaging from the air, surveying, etc.).  I think we will see solidification of business models in this area, with fewer companies competing. Some progress will be made in drone delivery but it still will not be broadly available.

In your interview with Jon Stewart, he pressed you a bit about the perverted uses of drone technology.  At the time of the interview, you spoke about positive daily uses, like package delivery, but in recent news, you have been recruited by the North Carolina prison system to assist in detecting drones that may be dropping contraband into prisons.  Can you share what you are working on to defend against this usage of drones?

See above. There are some people trying to drop cell phones, drugs, wire cutters etc into prison yards.  We are trying to develop cheap and mobile sensors that can detect a nearby drone and alert prison officials that there is a problem.

During your EDGE 2017 keynote, you shared an example image that demonstrated the current status, at the time, of driveress car computer vision and spoke about this in terms of AI inductive reasoning.  The image showed the computer incorrectly categorizing a single image on the back of the car as four separate images.  How far has the technology for driverless car computer vision come in the last seven months, since your keynote?

To my knowledge, there has not been any improvement and indeed, more problems have surfaced. The biggest issue that has come up is the ability to trick such algorithms into seeing images completely incorrectly, i.e., making a car computer vision system see a 45 mph speed sign instead of a stop sign. This problem highlights just how immature this technology is and how much more work needs to be done.

Have there been notable advancements in AI inductive reasoning, particularly for the technology used in driverless cars?

See above.  I would say there have been notable advancements in how brittle AI is. Moreover, the deaths of the Tesla driver in Mountain View and the pedestrian in Arizona show just how difficult it is to develop safe autonomous systems.

How do you feel about the recent issues with autonomous vehicles? For example, the recent accident in Tempe, Arizona, when a women was unfortunately killed, or the incident in California where the car was on “ autopilot” and, supposedly, the driver was notified to take it off but didn’t?

See above. They both perfectly illustrate the problems with brittle AI that I have been highlighting these past two years.

Are these two recent incidents technology issues or operational issues?  

Both. They demonstrate how unreliable this technology is but they also highlight operational immaturity and the fact that we do not yet know how to effectively test such nascent technologies to know that they are “safe enough,” particularly in public settings.

What kind of regulations are currently in place to support driverless cars? 

None at the federal level and various states have tried to address this. This is an evolving and dynamic issue, so there is no broad consensus. California has said they will allow driverless cars on the road with no safety driver, which I am completely against. Other states that allow such cars still require a safety driver in the car, which I think is still needed, but the Tempe accident demonstrates how unreliable people are. There is movement in Congress for the AVSTART legislation, which is supposed to put together a federal framework that will address testing as well as deployment of these vehicles, and I expect it to pass before the summer break.

What additional regulatory and safety standards would you recommend before advocating for widespread use and acceptance of driverless vehicles?  

I have been strongly advocating for a vision test for driverless cars. We insist that people have vision tests before they get a driver's license and given that we know driverless car computer vision technology is very brittle, it makes sense that these cars should also have to pass a test that demonstrates that they can see in typical driving environments that include rain, snow, fog, etc. Read more about this.

Learn more about Professor Mary "Missy" Cummings.

There is space available at the 35th Annual Seminar, "Automation and Technology in the Supply Chain," being held in Oak Brook, Illinois, on May 17, where Cummings will be presenting live. Register for the event.