Embodied artificial intelligence

from Wikipedia, the free encyclopedia

Embodied artificial intelligence ( English embodied artificial intelligence ) denotes a flow in recent research in the field of artificial intelligence (AI). Following the theory of embodiment , it is assumed that intelligence must be understood in the context of physical agents that behave in a real physical and social world. The design and construction of robots should be guided by this core belief. The research field of "embodied artificial intelligence" is characterized by a high degree of interdisciplinarity and employs researchers from the fields of engineering , philosophy , psychology , computer science , biology , neuroscience , biomechanics , materials science and linguistics .

history

In the early research on AI, thought processes were conceptualized as abstract symbol manipulation or arithmetic operations. The focus was consequently on algorithms and computer programs , whereby the underlying hardware was considered largely irrelevant. The Australian computer scientist and cognitive scientist Rodney Brooks was one of the first to question this perspective. In his 1999 lecture Intelligence without Reason , he criticized the then common practice of focusing on the emulation of human problem-solving and reasoning skills in the development of AI systems using a top-down approach . In his opinion, intelligence models developed in the context of traditional AI research, which are strongly based on the functioning of the computers available at the time, show almost no similarity with the modus operandi of intelligent biological systems. This becomes clear, for example, from the fact that the majority of the activities that people pursue in their everyday life are neither problem solving nor planning, but routine behavior in a relatively benign, but highly dynamic environment. As an alternative, he called for intelligence to be approached from a bottom-up approach by focusing on agents who have a body and behave in a real-world environment.

Embodied artificial intelligence today

In the meantime, the view of embodied artificial intelligence is gaining increasing attention both in science and in the public eye: “No longer static and software-controlled robotic arms or knowledge-based systems oriented towards dialog, strategy or expert heuristics are making the headlines, but the insect-like six-legged creature called Genghis that moves in unfamiliar spaces, the Honda robot that can climb stairs, or the Cog that hops around like a small child ”( Werner Rammert : The future of artificial intelligence: embodied - distributed - hybrid) This is not least the fact owing to the fact that systems that were developed based on the traditional AI paradigm are far from approximating human generalized intelligence. Intelligent systems outperform people today with regard to a large number of often highly specialized activities (cf. for example Deep Blue and AlphaGo , which defeated the human world champions in the field of chess and Go games). At the same time, however, they are unable to perform tasks such as using a bicycle or pen that a child can easily do. The agents developed by representatives of embodied artificial intelligence form a counterpoint to the island-endowed systems that emerged from the traditional AI approach. Their intelligence should arise from the interaction between software, body and environment and thus be as similar as possible to human intelligence. In this context, the Swiss computer scientist and robotics expert Rolf Pfeifer takes the position that intelligence is exclusively embodied in agents, i.e. H. can be ascribed to real physical systems, the behavior of which can be observed in interaction with the environment. The American entrepreneur Jeff Hawkins , another prominent figure in the field of embodied artificial intelligence, believes, however, that an intelligent machine does not necessarily have to have a physical body, but only be able to change what it perceives through movement :

"For example, a virtual AI machine could" move "through the Web by following links and opening files. It could learn the structure of a virtual world through virtual movements, analogous to what we do when walking through a building. "

“For example, a virtual AI could“ move ”through the Internet by following links and opening files. She could learn the structure of a virtual world through virtual movements, analogous to what we do when we walk through a building. "

- Jeff Hawkins in What Intelligent Machines Need To Learn From the Neocortex

Individual evidence

  1. ^ F. Iida, R. Pfeifer, L. Steels, Y. Kuniyoshi (eds.): Embodied artificial intelligence. International Seminar Dagstuhl Castle, Germany, July 7-11, 2003. Revised selected papers. Springer, Berlin / Heidelberg 2004.
  2. ^ A b R. Pfeifer, J. Bongard: How the body shapes the way we think. A new view of intelligence. The MIT Press, Cambridge, Massachusetts, USA 2007.
  3. ^ RA Brooks: Intelligence Without Reason. In: Proceedings of the Twelfth International Joint Conference on Artificial Intelligence (IJCAI-91). Sydney, Australia 1991, pp. 569-595.
  4. ^ A b J. Hawkins: What Intelligent Machines Need to Learn From the Neocortex. June 2, 2017. (spectrum.ieee.org)
  5. ↑ W. Rammert: The future of artificial intelligence: embodied - distributed - hybrid. In: TUTS-Papers, TUTS-WP-4-2003. 2003.