Main Subcommittee Category: Emerging Concepts and Innovative Technologies

How would you label your submission?: Industry

Abstract:

Since ChatGPT marveled the world with its capability to generate interesting answers, fascinations (and fears) around generative AI (“genAI”) have been compounding as new genAI capabilities from researchers and the industry frequently made headlines. Venture capitals poured $ 21.8B into genAI startups last year, and 36 companies hit the unicorn status. Across industries, including defense, cybersecurity and healthcare, leaders are fascinated by genAI’s potential to not only surface insights in multi-modal data sources (structured, text, image, video), but also interface with humans in natural language. Their fears range from safety and privacy issues to irresponsible applications that lead to unethical decisions or cyber vulnerability exploitations. Industry leaders clamor for AI governance as organizations from EU and NIST to the Whitehouse published their evolving guidelines.

On the application front, multiple gaps exist toward realizing the power of genAI in a typical workflow that goes beyond Q&A (e.g. submitting a request). A Large Language Model (“LLM”) or Foundation Model (“FM”) doesn’t know an organization’s workflow, the data required from the user, and the enterprise system(s) to interact with to complete the request. A typical LLM/FM also lacks the ability to conduct a multi-turn conversation to gather the information to complete such request.

We have worked on closing such gaps to produce an AI system that is truly conversational, configurable to understand an organization’s workflows, and integrating into the organization’s IT environment, as authorized, to gather insights across systems and complete work on a user’s behalf. It leverages multiple LLMs/FMs as required.

In this paper/talk, we describe the gap-closing components together with their complementing LLMs/FMs in an architecture compatible with the Zero Trust security model and following AI governance guidelines. We would also demonstrate how this AI system takes Human-Computer Interaction (HCI) to the level an LLM alone cannot, effectively enhancing the mission effectiveness of the workforce.

Keywords: AI;ARCHITECTURE;AUTOMATION;AVATAR;COGNITIVE;DISRUPTIVE INNOVATION;ENHANCING PERFORMANCE;INTEGRATION;NATURAL LANGUAGE PROCESSING

Menu

Mr. Cheong Ang