In his science fiction books, author Iain M. Banks created a utopian society called the ‘culture’ that is governed by tremendously powerful but benevolent AIs. These almost god-like ‘minds’ are capable of holding simultaneous conversations with millions of other beings, both biological and artificial, and this while they subtly manipulate the geopolitical intrigues of neighbouring, inevitably inferior, societies. Science fiction indeed; but is it a feasible scenario? In the field of Artificial Intelligence today, opinions differ markedly. Today’s AIs are programmed to do a specific task; they cannot do anything else. But to create a Culture Mind, we’ll first need to create a generalised, self-learning, self-improving AI. For this article we spoke to the pioneer of self-referential, self-improving AIs - one of the optimists, Professor Jürgen Schmidhuber, who is co-director of the Swiss AI lab IDSIA in Lugano and head of the Cognitive Robotics Lab at the Tech University Munich.
All current AIs are limited to specific, although increasingly complex, environments. Today’s AIs that perform well do so in a specific task; they have clearly defined and pre-programmed specific problem solvers. But if you were to drop a chess playing AI in a game of Go it would collapse. Similarly, you cannot ask a car-driving AI to fly a plane, or vice versa. That is the difference between AIs and human intelligence: contrary to today’s machines, we are able to adapt to different environments and learn as we go. Human beings are constantly confronted with new problems, relevant to the situation they find themselves in, and somehow create new strategies or behaviours or programs to solve those problems. In other words, we have a rather general type of intelligence.
You state that you want to build an optimal scientist and then retire. But is it at all possible to create a generally intelligent AI? And even if it is theoretically possibly, is it practically feasible to build one in the coming decades?
Historically, the field of AI has been split in two different methodological ‘schools’ of thought: the deductive and the statistical approaches. Are we still pendulum-swinging from one approach to another or are these approaches finally beginning to merge?
Can you give us some examples of your more practical work?
About Jürgen Schmidhuber
Jürgen Schmidhuber is Director of the Swiss Artificial Intelligence Lab IDSIA (since 1995), Professor of Artificial Intelligence at the University of Lugano, Switzerland (since 2009), Head of the CogBotLab at TU Munich, Germany (since 2004, as Professor Extraordinarius until 2009), and Professor SUPSI, Switzerland (since 2003). He obtained his doctoral degree in computer science from TUM in 1991 and his Habilitation degree in 1993, after a postdoctoral stay at the University of Colorado at Boulder. He helped to transform IDSIA into one of the world's top ten AI labs (the smallest), according to the ranking of Business Week Magazine. He is a member of the European Academy of Sciences and Arts, and has published more than 200 peer-reviewed scientific papers (some won best paper awards) on topics such as machine learning, mathematically optimal universal AI, artificial curiosity and creativity, artificial recurrent neural networks (which won several recent handwriting recognition contests), adaptive robotics, algorithmic information and complexity theory, digital physics, theory of beauty, and the fine arts.
You're exploring 'Tech', The Fifth Conference on the impact of technology on business and society
About The Fifth Conference
The Fifth Conference is an innovation platform for people who like to think. We publish a journal, host events and make this website.