Lex, the Growth House answer bot, interacts with web site visitors by answering questions presented in natural language. Lex responds by presenting the user with relevant web resources, and optionally may present content drawn from anywhere on the web. Lex is currently in preliminary testing and is only available on the web during selected test windows.

Lex is an example of a class of artificial intelligence programs known as chatterbots, answerbots, or conversational robots. His programming involves parsing user inputs against a domain-specific vocabulary database for topics in end of life care and palliative medicine. Lex is gradually learning how non-specialist humans talk about end of life issues by actually interacting with folks he meets on the web. Each time Lex goes on the web he keeps logs of his conversations. His performance is evaluated between sessions and the new log samples of human communication are used to program Lex for his next outing. Users can even suggest to Lex what he ought to learn about, helping the development team spot deficits in Lex's understanding of the world.

Lex is written in a specialized bot development language which is optimized for human speech pattern recognition. The goal is not to duplicate the sophisticated natural language processing capabilities already available through various National Library Of Medicine (NLM) interfaces to databases such as MEDLINE. Instead, the goal is to recognize poorly-structured search requests related to certain topics in end of life care and efficiently pass them to appropriate databases that offer more extensive parsing capabilities.

We solicit feedback and suggestions from those working in related topics in medical informatics and artificial intelligence.