What is the optimal interface between a human and a machine?
In the Zero Pressure Podcast, Helen Sharman is joined by Commodore Michael Brasseur and Professor Sameer Alam to explore the concept of ‘human-machine teaming’, how we manage the relationship between the human, the machine, and the interactions and interdependencies between them.
The concept of machine learning is becoming increasingly visible and valuable in society, but understanding how to optimise the human and machine interface and how they work together is key to the success of this technology.
So say Commodore Michael Brasseur and Professor Sameer Alam, two AI experts working in unmanned maritime applications and air traffic control respectively, in the latest episode of Imperial College London and Saab’s Zero Pressure podcast.
Available at your favourite channel:
The role of maritime robotics
Zero Pressure’s host, UK astronaut Helen Sharman, hears first from Michael Brasseur about his role as leader of the US Navy’s new Task Force 59. The team aims to rapidly integrate unmanned systems and artificial intelligence into operations in the navy’s 5th Fleet area of operations in the Arabian Gulf. The technology is eventually expected to be used in the likes of counterpiracy and maritime interdiction actions.
Commodore Brasseur tells Helen that autonomous capabilities can have great value in reducing the load on human operators, especially in processing a huge amount of data or in multi-tasking operations under pressure. However, if humans are to trust machine algorithms, we need to be sure the machine will do what we expect it to do.
“For example, we wouldn’t want one of our un-crewed missions with a machine violate someone else’s territorial waters. We need to understand how machines behave when potential adversaries try to disrupt our communications or GPS,” he says.
“It’s a really difficult problem. The only way to build trust is by operating in a communications-contested environment, by going through the paces, doing several exercises where the machine is being used what it’s been designed to do.”
Ultimately, Commodore Brasseur believes that one of the great values of maritime robotics is to be able to utilise the fact that they are less expensive than manned solutions, such as more destroyers or cruisers, and therefore have more available for use.
“There’s value in getting more sensors out on the water to enhance our maritime awareness,” he says.
He paints a picture of one person controlling 12 or 13 maritime robotics above, on and below the water, with a supply of comprehensive data that builds up a pattern of life, so a machine can learn to determine when something is abnormal, highlighting it for the operator to take a closer look. However, he adds that his ideal picture of the future is to always have humans in the loop, particularly for high-risk operations.
Machine learning for air traffic control
Air traffic management is further along in its embrace of machine learning research than its maritime counterpart, partly due to the less harsh physical environment and an embedded culture of automation.
Professor Sameer Alam, Deputy Director of Air Traffic Management Research Institute and Co-Director of SAAB-NTU Joint Research Lab in Singapore, has 20 years’ experience of researching machine learning for air traffic management, and leads a team of 20 research scientists and seven PhD students at the Singapore lab.
For Professor Alam, the main focus areas include solving problems such as how the machine, or AI agent, perceives its environment, takes action, whilst also evaluating what the repercussions of that action down the line. And, how to get to the stage where the machine has collected enough data on human actions to recognise established patterns of behaviours, upon which it can start basing its own decision-making.
“This makes the algorithms very powerful because now they are taking a collective human knowledge that has evolved over time,” he says.
Other research areas include Explainable AI, when the machine not only advises you of a decision, but also advises you of the logic behind a decision. And does so in an understandable and thus trustworthy way.
The importance of the user interface
Interestingly, both Commodore Brasseur and Professor Alam highlight the importance of the interface technology in building trust between the human user and the AI agents.
In the air traffic control scenario, Alam points to augmented reality and virtual reality equipment, including trials with the Microsoft Hololens, where the controller can work from home, with no need to come to the control centre.
In the maritime case, Brasseur says that a piloting trial of an unmanned surface vessel was being controlled by an X-Box controller. “It’s a well-known interface among the operators, and makes the transition easy to operate,” he says.
Another important benefit in improving interface technology and utilizing autonomy is reducing the cognitive load on a decision maker or operator. “We’re in a period of sensor overload and the human brain can get overloaded, it gets fatigued, whereas the machine learning can process tonnes and tonnes of information and make sense of it,” adds Commodore Brasseur.