AI via Neural Processing
- Speaker: Dr Chris Huyck (Middlesex University)
- Host: Steve Furber
- 8th October 2008 at 14:15 in Lecture Theatre 1.4, Kilburn Building
In this talk I'll discuss the approach that I and my collaborators have been following to developing real AI through simulated neural processing. The approach is to follow known neural and psychological processing as closely as is reasonable. It also requires the neural agents to work in and learn from an environment. This is a long term (years or decades) approach to developing AI that is capable of solving tasks such as the Turing test. This talk will include a description of our neural model, fatiguing Leaky Integrate and Fire neurons, and Cell Assemblies, a central concept in our approach. The talk will describe several projects including a Natural Language parser, sequence learning, rule learning, and our recently developed neural games agents, CABot1 and CABot2. These agents take natural language commands from a user, sense the visual environment, and follow simple plans to execute the commands.