Technology behind Sasha
The current state of conversational AI technology
Artificial Intelligence can seem like a foreign idea. In order to understand where we are going, let’s review the past and current status quo.
Finite-state dialog systems
Chatbots built using finite-state dialog systems enable experiences analogous to those found in a phone tree or IVR system.
- System initiated (often presently frustrating user experiences)
- Brittle (often get stuck)
- Don’t scale
Chatbots built using retrieval-based systems (e.g. Siri, Alexa, and Cortana) are limited to:
- Shallow conversations (think ‘what’s the capital of Alaska’ or ‘where’s the nearest hotel’)
- Poor task completion
Both methods don’t meet the natural flow of human conversations. These limitations lead users to become frustrated and reduce engagement and adoption of new technology. Examples of these limitations include telling a chatbot to “start over” when booking or having to rephrase your question multiple times. These frustrations stem from the technology missing natural language, the understanding of intent, and the ability to change course as quickly as our minds do.