BOT2001 Report
Managing Discourse with SmartBots

By Brian Proffitt

As the technology of bots matures, the terminology surrounding this new frontier is constantly changing. Chat bots are becoming more known as chat bots, and a conversation between a human and a bot is no longer a chat so much as a discourse.

The methods of handling discourses between Homo Sapiens and bots is a pressing issue for bot and intelligent agent developers these days. This theme was certainly prevalent during this year's BOT2001, and most certainly the subject of the talk given by Artificial Life, Inc.'s VP of Technology Development, Noel Bush.

Bush, who's Boston-based company has developed the SmartBot product line, comes at this topic with a remarkable fervor. While he acknowledges that there are many other companies within the same technology space as Artificial Life, how each of these companies deals with bot/human interaction is what will set them apart from each other.

Naturally, Bush places great faith in Artificial Life's methodology. And with good reason. Clearly this is a company that has done a lot of homework in what Bush terms as "discourse management."

In simplest terms, an example of a well-managed conversation with a bot is when the bot never tells the human user "I don't know." During an efficient human conversation, if one party does not know the answer to the other's question, then often a new initiative is created. "I don't know that answer, but let's see if we can find out together," would be an example of this new initiative.

When bots can deftly handle conversational disconnects well, then they are doing their job, Bush explained. The best way of handling these disconnects, he added, is through effective modeling of the context of the conversation.

Traditionally, there are two ways in which context is modeled. The plain, ordinary way is creating a database of a simple dialog, which allows the bot to remember the conversation as it goes and ideally refer back to previous points in the conversation when need be.

Artificial Life's bots use a different methodology, Bush said. In their case, topics are kept as atomic as possible, which allows the SmartBots to create associative links from topic to topic and help keep the discourse running more smoothly.

SmartBots can also recall topics and events from earlier conversations. Bush highlighted an example where a user could tell one of the SmartBots he or she wasn't feeling well that day, and the next time the bot conversed with the same user, it would ask how the user was feeling.

Another way the conversation can keep going is through the SmartBot's management of a series of goals, which the bot's owner has stipulated. For instance, the owner of a sales site may want the SmartBot to gather a minimum amount of information about the customer. These goals can then be interwoven into the conversation in an easygoing, non-threatening manner.

SmartBots can also adjust to the emotional state of the user, Bush said. Artificial Life has taken a slightly different approach to handling emotions. Instead of reflecting emotions back to the user, SmartBots keep a virtual eye on certain trigger words and will, in the case of a very angry user, escalate the dialog to a human agent.

This level of sophistication of SmartBots, and other bots in the eCRM field, puts bots into a very exciting time in their development. In a few years, Bush sees the entire technology as becoming as ubiquitous as PCs are today.

"It's due to become a pervasive technology type," he said.