MLM Home Page - José Ferreira de Castro

The main purpose of this page is to make available the work in progress related to the Measurement Logic Machine (MLM) and the corresponding Measurement Logic (M-Logic). The latter is a theoretical framework for a machine that includes a fast online reinforcement learning method that rapidly discovers and uses the regularities of the surrounding world to perform inference in non-stationary environment. The MLM provides a simple and intuitive basis for a unified theory of cognition, with objectives similar to those of SOAR, ACT-R, CLARION, EPIC, ICARUS, LIDA, etc. So far, the machine only addresses the non-verbal processes of cognition, but already gives hints to explain some aspects of language learning.

I encourage the interested reader to download and try the latest Python 3.4.3 (Anaconda) version, found in the Source Codes and Tutorials subpage. There are specific and updated explanations there, including the most up-to-date Power-Point presentations. If you need further explanations, just send me an e-mail to CastroJFGF@gmail.com.

A preliminary theoretical M-Logic draft document, a pool of ideas and research notes, can be found in the attachments hereafter, along with two older Power-Point presentations. They further highlight some of the M-Logic Machine main ideas and applications.

Over the years, the M-Logic Machine concept has been implemented and tested with several programs written in XSB Prolog and Python (versions 2.7.2 and 3.3.4). A Magabot is also being currently used to test the concept with a real robot.

The current M-Logic Machine offers straightforward answers to questions like: How can reflex actions and learning be integrated? How are cinematic memories used for inference? Can machines know and believe? What is the use of dreaming? Why do we need a short-term memory? What generates "superstitious learning"? Why do we learn nouns before verbs? Can we do AI with irrational and incoherent machines?