IDEAS SoCal AI & Data Science Conference 2018

Alexa and other chat bots have taken the world by storm. However, the current reality is that natural language applications have severe limitations, failing on tasks that even a child could easily handle. They typically have no memory of what was said earlier, don't reason about their task, and have no common sense. Moreover, they are usually engineered to perform just one specific task -- i.e. they do not have any general intelligence.


Originally (the first wave), conversational agents were programmed manually using logic flow. Everyone is well aware of limitations in scaling and maintaining such system. More recently (in the second wave), big data and machine learning have automated task classification (intent identification) and some key parameter extraction. However, we now also understand the difficulty in obtaining massive amounts of manually tagged training data, and the problems with lack of transparency, reasoning, and real-time adaptability in these systems.


This talk will explain how a cognitive architecture approach can overcome many of these limitations and provide a much more intelligent platform for conversation agent. A practical example of this 'third wave' technology will be demonstrated.