SBIR Phase I: Automating Conversational Systems For Enterprise Customer Support

The broader impact/commercial potential of this Small Business Innovation Research (SBIR) Phase I project will result from the development of a scalable, enterprise-ready conversational AI platform that enables businesses to automate customer support requests, optimize call routing efforts to get customers to the right human support agent, generate customer self-service resources like smart FAQs, augment human staff with bot capabilities, and enable personalized recommendations to support sales and purchasing decisions for customers. Such technology is beneficial to any enterprise organization, B2C (business-to-consumer) and B2B (business-to-business), who needs to communicate to customers, employees, and the public. With rising customer service and communication costs, automation is critical for businesses to maintain profitability and competitive advantages. On a societal level, conversational interfaces democratize access to information and resources worldwide. A significant portion of the world's population lives in regions where internet access is poor and device ownership is rare. Many also suffer from physical limitations, such as impaired vision or limited motor control. Accelerating technology access through conversational interfaces such as SMS and voice has already been shown to transform quality of life for disadvantaged groups through applications in telemedicine, personal finance, and education. This Small Business Innovation Research (SBIR) Phase I project aims to enable human-level conversational ability in machines. Teaching computers to communicate like humans is a critical step to achieving human-level machine intelligence (HLMI), also called artificial general intelligence (AGI). Current state-of-the-art approaches to automated dialog systems use a combination of supervised learning and reinforcement learning implemented with deep neural network architectures. These approaches have shown promise on toy problems in academic research, but have mixed performance in enterprise settings. This Phase 1 project extends current deep learning approaches by developing a repeatable methodology for annotating and preparing enterprise data for use in dialog systems and a scalable, grounded neural network architecture that can reference external knowledge sources with minimal manual engineering. This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.