Making machines more intelligent


Manuela Veloso is the Herbert A. Simon University Professor in Computer Science and Robotics at Carnegie Mellon University. She was a guest speaker at an ABB event during the World Economic Forum’s 2016 annual meeting in Davos.

Q: How do you define artificial intelligence?
AI has the goal of trying to make a computing device perform intelligently, integrating perception, cognition and actuation capabilities. Some researchers see it as the study of how humans are intelligent, both in behavior and at the brain level, so areas such as neuroscience and cognitive science share goals with AI. Others, including myself, focus on achieving intelligent performance, independently of the human process. There are different forms of research. But the goal is having intelligent machines.

Q: In industry, what are some current applications?
Industry has traditionally addressed the automation of repetitive tasks, in particular using rule-based programs. Now industry is exploring less defined types of automation, using methods that can handle more uncertainty and can improve with experience and acquire knowledge from humans. A machine that becomes better through experience and then personalizes its behavior would be an AI product. In a factory, that could be a machine that can learn from instruction or be easily reprogrammed to do something new.

Q: Is that what your so-called “CoBots” do?
Yes, the CoBots at Carnegie Mellon are a good example of autonomous task service, navigation and learning performance. Robots can move around indoors, but our goal was to deploy them to be completely independent. The moment we let the robot go, without anybody following it, was a compelling and memorable one. Now it has traveled more than a thousand kilometers on its own, based on a robust and novel algorithm for localization and navigation. It also interacts with humans, who can make requests using natural language, like, “Please take this book to the lab.” The robot initially doesn’t know the names of locations or objects. But it can learn by asking people or going to the web. It keeps this information for future use.

Q: How does the CoBot know where it is and where it’s going?
CoBot makes a map of the building to capture the persistent features, like walls. When navigating, the robot localizes itself both by knowing where it has moved and by matching its map to its perception of these features. The robot perceives many other objects, like tables and chairs, which can vary in position, but they’re also used for localization in a novel approach contributed by my student Joydeep Biswas.

Q: Does the technology have any commercial use?
Robots like CoBot have clear commercial use. Its algorithms can be used in other autonomous mobile platforms. In addition, another one of my students, Richard Wang, introduced the idea of using these robots as mobile data-gathering platforms. As CoBot moves around, it gathers data, like Wi-Fi signals or temperature. Such data can help people make decisions about allocating resources in the building, such as where to put Wi-Fi access points. So a mobile robot can help improve data gathering just by being able to move around and acquire useful data.

Q: Should people be afraid of AI?
Humankind will always follow its own route with new discoveries and societal goals. So research and development will inevitably proceed towards machines becoming more intelligent, which brings enormous potential benefits. The challenge and opportunity is to make good use of the AI technology. The good news, as Herbert Simon said, is that we are not spectators, but actors in the future of technology.