AM-Conferences

Topics

• Adopting AI in Operations.
• Conversational AI designs.
• SiWarex, situational awareness in Maximo.
• Be a part of the team

Conclusions

The session, “Next-Gen AI – Enabling Situational Awareness in Maximo,” was delivered by Mats Gothe, Anu Bhamidipaty, and Mumtaz Mesania at Maximo World 2024, introducing a preview of new conversational AI in Maximo Application Suite (MAS) 9.0.

At Maximo World, we showed the progress made with generative and conversational AI in Maximo. In this session, we shared the ‘behind-the-scenes’ work across user research, concept design, and AI release that had led up to the new Conversational AI capabilities in Maximo.

• User research on how organizations using Maximo are adopting AI.
• Design concepts to establish a Maximo Conversational AI Assistant.
• AI research by the IBM research team on GenAI, Assistants, LLMs, and RAG architectures for Maximo

Conversational AI infuses an AI Assistant into the Maximo application, providing contextual and situational awareness into the business objects and workflows. The benefits are time-to-value, cost efficiency, and operational resilience. Conversational AI can transform questions into answers and actions, reduce knowledge gaps, diagnose issues and repairs, and summarize extensive operational and maintenance data.

Regardless of the organization’s size or industry, they all face similar needs: streamline maintenance planning, enhance asset or equipment reliability, and optimize workflows to improve productivity. Generative AI can give users what they want faster than ever before—but that convenience is only valuable when it’s built on a foundation of trust. Here are some of the top critical factors for trust indicated by users:

• Cybersecurity: 72% of interviewees say their organization lacks the governance and structures needed to manage GenAI ethical challenges and security, resulting in costly compliance fines. They believe AI benefits could come at a high cost.
• Business Value: Organizations are focused on short-term thinking, the ongoing effort needed for insourcing or outsourcing, refinement of models, testing, and patience required to see a return on investment.
• Foundational Data: Lack of proprietary and high-quality data is a challenge. 67% of participants cited data accuracy as their main concern. The data needs to be well-rounded; otherwise, the outputs or recommendations might be incorrect.

In our design discovery studies on GenAI and Conversational AI, we identified four groups of use cases:

• Conversational AI acting as an Assistant: Guiding on Maximo workflows and suggesting the next steps.
• Conversational AI answering any question you ask: Situational to Maximo data and workflows.
• Automation using a text prompt: Implies an action to be performed based on conditions and constraints (e.g., open, close, complete, set).
• Instrumentation using a text prompt: Directs a response to a dashboard card or a work queue query.

The IBM Research team have contributed the technology to provide situational awareness to the user by surfacing information in a contextual way, integrating it into their current experience, and responding and refining based on users’ questions. We call this technology Siwarex.

Let’s understand this by breaking it down. In industrial domains, there is a wide variety of data we work with. There is structured data in Maximo and textual information in the form of work orders, logs, failure reports, etc. There are also existing analytics, e.g., a health scoring capability. All this information is exposed as APIs.

How does it work? The key input needed for Siwarex is the representation of the domain metadata. Metadata consists of entities and their relationships, best represented by a graph structure. Each of the nodes in the graph contains information about how to access the entity, whether in a database or via an API.

When an actual text prompt comes from the user, it is processed in three steps: identifying the relevant metadata, decomposing the query into tasks, and executing the said tasks. LLMs assist in all of these steps. This is also referred to as agentic behavior, where an agent is not engaged in only single-turn Q&A but can plan a flow, backtrack, and execute.

Demonstration

The session ended with a demonstration of the SiWarex conversational AI capabilities by IBM Research.