Few other future topics are as controversial as artificial intelligence. How will AI change industrial production? Or: How do autonomous cars reach their destination safely? Will AI be able to recognize, evaluate and process data for decision-making with the same level of complexity as humans do? And if so: Is there any chance that AI will one day take on a life of its own and take more intelligent decisions than humans?
Questions upon questions – these experts provide answers based on latest research:
AI research shows just how complex the interdependencies of applied human intelligence are. Our brains excel at dealing with uncertainty, enabling them to take the (presumably) best decision even in unclear situations. This is where AI still has deficiencies – it needs certainty. Researchers are working to give AI the plasticity it needs to make robots act ‘correctly’ even in ambiguous situations. This requires suitable algorithms. And tremendous computing power. Quantum computers and Deep Learning can help, as can extremely comprehensive recognition and simulations of the environment.
In addition to general considerations – including the ethics of artificial intelligence – the scientists will present specific use cases for AI-based applications. Example: Smart robots can freely navigate through unknown spaces and identify objects to be transported to the desired location.
In brief: The Building Intelligence session will provide first-hand insights into AI research – with a focus on industry and applications. Here you will learn how AI that replicates human intelligence can be implemented in objects such as robots or databases.
Artificial intelligence isn't. The reason is simple: most successful, applicable methods focus on supervised learning of human-annotated data sets. Nowhere close to how biology learns; how natural intelligence emerges.Unsupervised learning to the rescue? Modern latent-variable models can be used to implement predictive coding. It allows you to predict; to act like an intelligent system. We can make this work in complex dynamical systems, such as drones or robots; but we can't break this final frontier to intelligence quite yet. Once we can, how can we make our systems accountable? In the end, we need to discuss ethics of such algorithms. In my talk I will try to enlighten all of these issues.
Patrick van der Smagt is director of the open-source Volkswagen Group Machine Learning Research Lab in Munich, focussing on probabilistic deep learning for time series modelling, optimal control, robotics, and quantum machine learning. He is also a faculty member of the LMU Graduate School of Systemic Neurosciences and research professor at Eötvös Loránd University Budapest. He is the founding head of a European industry initiative on certification of ethics in AI applications (etami). Patrick previously directed a lab as professor for machine learning and biomimetic robotics at the Technical University of Munich, and leads the machine learning group at the research institute fortiss. he founded and headed the Assistive Robotics and Bionics Lab at the DLR Oberpfaffenhofen. Ages ago he did his PhD and MSc at Amsterdam’s universities on neural networks in robotics and vision. Besides publishing numerous papers and patents on machine learning, robotics, and motor control, he has won a number of awards, including the 2013 Helmholtz-Association Erwin Schrödinger Award, the 2014 King-Sun Fu Memorial Award, the 2013 Harvard Medical School/MGH Martin Research Prize, and best-paper awards at machine learning and robotics conferences and journals. He was founding chairman of a non-for-profit organisation for Assistive Robotics for tetraplegics and co-founder of various tech companies.
My research is driven by the puzzle of why humans can effortlessly manipulate any kind of object while it is so hard to reproduce this skill on a robot. Humans can easily cope with uncertainty in perceiving the environment and in the effect of manipulation actions. One hypothesis is that humans are exceptionally accurate in perceiving and predicting how their environment will evolve. Therefore, improving the accuracy of perception and prediction in robots is one way forward. In this talk, I would like to advocate for a different view on this problem: What if we will never reach perfect accuracy? If we accept that premise, then an important focus towards more robust robotic manipulation is to develop methods that can cope with a base level of uncertainty and unexpected events. In this talk, I will propose a set of such methods that will help to solve this great puzzle on how to enable autonomous robotic manipulation.
Jeannette Bohg is an Assistant Professor of Computer Science at Stanford University. She was a group leader at the Autonomous Motion Department (AMD) of the MPI for Intelligent Systems until September 2017. Before joining AMD in January 2012, Jeannette Bohg was a PhD student at the Division of Robotics, Perception and Learning (RPL) at KTH in Stockholm. In her thesis, she proposed novel methods towards multi-modal scene understanding for robotic grasping. She also studied at Chalmers in Gothenburg and at the Technical University in Dresden where she received her Master in Art and Technology and her Diploma in Computer Science, respectively. Her research focuses on perception and learning for autonomous robotic manipulation and grasping. She is specifically interesting in developing methods that are goal-directed, real-time and multi-modal such that they can provide meaningful feedback for execution and learning. Jeannette Bohg has received several awards, most notably the 2019 IEEE International Conference on Robotics and Automation (ICRA) Best Paper Award, the 2019 IEEE Robotics and Automation Society Early Career Award and the 2017 IEEE Robotics and Automation Letters (RA-L) Best Paper Award.
Recent advances in deep learning and GPU-based computing have enabled significant progress in several areas of robotics, including navigation, visual recognition, and object manipulation. This progress has turned applications such as autonomous driving and delivery tasks in warehouses, hospitals, or hotels into realistic application scenarios. However, robust manipulation in complex settings is still an open research problem. Various efforts at NVIDIA robotics research investigate how deep learning along with physics-based and photo-realistic simulation can be used to train manipulators in virtual environments and then deploy them in the real world. Our work shows promising results on different pieces of the manipulation puzzle, including manipulator control, touch sensing, object pose detection, and object pick and place. In this talk, I will present some of these advances. I will describe a robot manipulator that can open and close cabinet doors and drawers, detect and pickup objects, and move these objects to desired locations. Our current system is designed to be applicable in a wide variety of environments, only relying on 3D articulated models of the furniture and the relevant objects. I will also present an example from an industrial use case, where a manipulator detects, picks up, and stacks boxes. I will discuss lessons learned so far, and various research directions toward enabling even more robust and general manipulation systems.
Dieter Fox is a Professor in the Allen School of Computer Science & Engineering at the University of Washington. He grew up in Bonn, Germany, and received his Ph.D. in 1998 from the Computer Science Department at the University of Bonn. He joined the UW faculty in the fall of 2000.
Dieter Fox is sharing his time between UW and NVIDIA, where he is leading the Robotics Research Lab in Seattle.
His research interests are in robotics and artificial intelligence, with a focus on state estimation and perception applied to problems such as mapping, object detection and tracking, manipulation, and activity recognition. He is a Fellow of the IEEE, ACM, and AAAI, and recipient of the IEEE RAS Pioneer Award. He was an editor of the IEEE Transactions on Robotics, program co-chair of the 2008 AAAI Conference on Artificial Intelligence, and program chair of the 2013 Robotics: Science and Systems conference.
At IBM Research we are convinced that the fundamental practice of scientific research is on the threshold of a revolution, driven by information technology, which will greatly accelerate scientific discovery. As we develop the next generation of information technology infrastructure, IBM Research has envisioned and is building a future generation of computing capabilities, which not only process digital bits – which are at the core of most of today's computing systems – but also integrate neural accelerators that greatly speed up AI algorithms, and exploit quantum computing in a single infrastructure. One example of the impact of this is IBM RoboRXN for Chemistry, which couples AI to predict the outcomes of unknown complex organic chemistry reactions with robotics to physically make molecules from anywhere in the world, drastically reducing the time and cost of the discovery process, enabling complex experiments also in a time of social distancing and home-office working.
Dr. Alessandro Curioni is an IBM Fellow, vice president of Europe and director of the IBM Research lab in Zurich, Switzerland. In addition to leading the IBM Research activities in Europe, he is also responsible for the global research in IoT and Security.
Dr. Curioni is a world recognized leader in the area of high performance computing and computational science where his innovative thinking and seminal contributions have helped solve some of the most complex scientific and technological problems in healthcare, aerospace, consumer goods and electronics. He was a member of the winning team recognized with the prestigious Gordon Bell Prize in 2013 and 2015.
Dr. Curioni received his undergraduate degree in Theoretical Chemistry and his PhD from Scuola Normale Superiore, Pisa, Italy. He started at IBM Research – Zurich as a PhD student in 1993 before officially joining as a research staff member in 1998. His most recent position has been the head of the Cognitive Computing and Computational Sciences department.
In 2017 he was named a member of the Swiss Academy of Engineering Sciences.
“Building intelligence” will be moderated by Prof. Dr.-Ing. Eckehard Steinbach as Session Chair.
To reach the next level of AI and achieve broad, practical applicability, new machine learning techniques that include robust predictions about the reliability of their results without large, manually annotated datasets are required.