TUTORIAL #1: Social and Emotional Robots as a Playground for Early-Stage AGI Systems
Organizers: David Hanson and Ben Goertzel, Hanson Robotics
In this tutorial, the presenters will
- Explain the key ideas behind social and emotional robotics
- Describe and demonstrate the platform for social/emotional robots created by Hanson Robotics, including (1) a highly realistic humanoid robot head, with realistic facial expressions and sophisticated software for social engagement, and (2) an open-source software framework using ROS, Blender and other tools to provide a 3D simulated humanoid robot head with realistic facial expressions and sophisticated software for social engagement. This framework can be downloaded by anyone and used freely in their own R&D
- Describe and demonstrate the use of the OpenCog AGI software framework for controlling the physical and simulated robot head. The audience will see a visualization of the most important knowledge and procedures in the OpenCog Atomspace, as the associated OpenCog system controls the robot.
- Outline the API via which AI researchers and developers can connect their own software to this platform
The potential of social and emotional robots – including physical and simulated robots, and including both heads and full human bodies – as an environment for testing and evaluation of early-stage AGI systems will be discussed. The crafting of appropriate tests and metrics for measuring AGI progress in this context will be highlighted as an open problem, and discussion with the audience on this and other topics will be solicited.
TUTORIAL #2: Incremental Machine Learning in AGI
Organizer: Eray Ozkural
We review the incremental machine learning, or transfer learning problem which was defined by Ray Solomonoff as one of the three major open problems in AI, and the solutions proposed so far. We explain the incremental learning approach with a genetic programming approximation to universal induction. We revisit incremental grammar learning algorithms which are relevant to set induction.
We then review the adaptive universal search approach and explain the Ordered Optimal Problem Solver system using FORTH and the dynamic program probability update approach in detail. We then show how the OOPS approach was extended to LISP-like languages, and the update procedures for stochastic grammar representation of program probabilities.
We then review the transfer learning methods for deep-learning/RNN training systems, and conclude the tutorial with a comparison of existing methods and a discussion of future research.
TUTORIAL #3: Universal Artificial Intelligence: Practical Agents and Fundamental Challenges
Organizer: Tom Everitt, ANU
Foundational theories have contributed greatly to the scientific progress in many fields. Examples include ZFC in mathematics and universal Turing machines in computer science. Universal Artificial Intelligence (UAI) is an increasingly well-studied foundational theory for artificial intelligence. It is based on ancient principles in the philosophy of science and modern developments in information and probability theory.
- an accessible explanation of the UAI theory and the optimal agent AIXI,
- three approaches to approximating AIXI effectively.
TUTORIAL #4: NARS Overview (Non-Axiomatic Reasoning System)
Organizer: Pei Wang, Temple University, and the OpenNARS Team
NARS (Non-Axiomatic Reasoning System) is a general-purpose reasoning system, and it has ability to learn from its experience and to work with insufficient knowledge and resources. NARS attempts to uniformly explain and reproduce many cognitive facilities, including reasoning, learning, planning, reacting, perceiving, categorizing, remembering, decision making, and so on.
The current implementation is an open source project Open-NARS (https://github.com/opennars/opennars/wiki), carried out by a development team whose members are in several countries.
The tutorial will consists of two parts:
- A lecture on the conceptual design of the system, including its theoretical foundation, its logic (Non-Axiomatic Logic), the system architecture, working cycle, control strategy, etc.
- Hand-on experience with Open-NARS. Working examples will be provided and explained, and from there participates can create their own examples to explore the functions of the system.