Robotland Bookstore

Thursday, October 22, 2009

ShanghAI Lecture 2 - Embodyment

The topic of lecture 2 was Cognition as computation: Succes and failures and the need for embodied perspective on intelligence.
Prof. Pfeifer explained the challenging problem of what a theory of intelligence might look like and how a set of desiign principles, complemented by a general framwork in the form of a number of meta-pronciples, would be suitable.

The original idea of Artificial Intelligence (AI) was that "evert aspect of learning or any other feature of intelligence can in principle be so precisely described that a mchine can be mode to simulate it." (John McCarthy et.al., 1955)

The so-called physical symbol systems hypothesis (PSSH) by Allen Newell and Herbert Simon states that for “general intelligent action” it is a necessary and sufficient condition that the system be a “physical symbol system”, i.e., a system that can build and manipulate symbol structures and has a physical implementation
(e.g., a brain or a computer).

Classical AI reseach areas have been problem solving, knowledge representation and reasoning, acting logically, uncertain knowledge and reasoning, learning and memory, communication, perceiving and acting.

Classical AI successes are search engines, formal games (chess), text pprocessing systems, data mining systems, restricted natural language systems, appliances, control systems.

Classical AI failures have been more natural forms of intelligence such as vision/perception in the real world, speech, moving, manipulation of objects.

There is a great difference between the real and virtual reality, industrial environments and real world. Industrial environments are well-known, little uncertainty, high predictability. Real world environments have limited knowledge and predictability, are rapidley changing and have a high degree of uncertainty.

More will come.....

No comments: