Creating Robots with Toddler-Level Intelligence Using the OpenCog AGI Architecture Research Proposal for the Epstein Foundation Ben Goertzel --- February 12, 2013 Executive Summary The core R&D goal of the proposed project is to use the OpenCog system to make a humanoid robot that can achieve a high score on the WPPSI preschool IQ test, based on engineering and education not specifically oriented toward that test This test includes tasks involving English question answering, visual pattern recognition, and manipulation of blocks and puzzles. Achieving this goal involves taking OpenCog functionalities currently being prototyped for the control of virtual-world characters and using them in a robotics context It also involves implementation of some recent, fundamentally new ideas involving the integration of symbolic Al systems (such as OpenCog's probabilistic logic engine) and subsymbolic Al systems (such as the DeSTIN computer vision system). This is viewed as an initial step toward the larger goal of creating a "robot toddler" with the general cognitive capabilities of a 3 year old human child - which in turn is viewed as a step in a longer project aimed at creating software and robotic systems with general intelligence at the human level and ultimately beyond. If successful, the proposed research will constitute a major step in the direction of these more ambitious goals, and will constitute a dramatic step forward in generally intelligent robotic communication, learning and interaction. This would be an augmentation of a similar but less ambitious OpenCog/robotics project that has recently been funded for two years (2013-2014) by the Hong Kong government, in the total amount of HKD$2.6M (roughly USD$342,000) including overheads (and requiring a matching USD$38K corporate contribution). This proposal seeks additional funding to complement the government funding, so as to accelerate the project and improve the odds of success and the quali