Cassandra is our “Figure of Speech”, a title that nicely represents who and what she is.  ejTalk’s mission from the very beginning has been to focus on the basic elements of conversation instead of developing products to accomplish a singular task.  Most conversational agents today serve a particular purpose.  They talk about specific topics but do not communicate in very conversational ways.

Present day finite state based systems are straightforward and logical for developers but lead to awkward and unnatural use for human participants.  A few years ago we were only allowed to command our machines with one word directives, and though these applications have since evolved, there is still a fundamental flaw in their design.  Each specific automated voice system can only talk about a narrow subject field, and if they tried to talk about anything else, a new framework would be needed to accomplish this new task.  With a product oriented model, each agent’s knowledge and world experience is built from scratch; there is no underlying architecture that allows for the easy reuse of basic conversational elements.

These single application domain approaches can be sufficient for simple products, but are stunted by their rigidity if we want our agents to talk about more than one thing at a time.  Unlike Cassandra, they are plagued with having to anticipate all possibilities of a conversation instead of having a mechanism to maintain consciousness in multiple domains.  Cassandra’s architecture was built with the explicit intention that one day she would talk to us about many different things.

Like a toddler, it will take a while for Cassandra to comprehend the topics of her conversations as we humans do, but once she can talk…well, the possibilities are endless.  What wouldn’t you want to talk to her about?