Beyond grammar: AI is getting content savvy

ChatGPT and similar tools already demonstrate how well artificial intelligence can write texts or programs within seconds. When used correctly, the output meets the right quality standards. This enables industries to benefit from reliable systems and new features. But how does it work?

Combine grammar with content knowledge base

By working with large language models, AI already produces good formulations. It is trained with exceptionally large volumes of language text examples and so it learns not only to form grammatically correct sentences but also to recognize connections between words. But without a concept of contextual meaning, LLM-based technologies still struggle with content. You can address this issue using techniques designed to map content correlations.

For example, in a domain-specific knowledge database for system maintenance, error codes can be stored alongside repair instructions, spare parts databases, and customer accounts. It also includes information about how these contents are related. This includes specifying which spare part is needed for which error pattern, the supplier from whom it can be sourced, and the relevant order number. The AI can then build on this information and context instead of having to combine them from sample texts.

Say hello to Siemens Industrial Copilot

In collaboration with Microsoft, we developed an AI-powered assistant. By combining the capabilities of generative AI with an industrial knowledge database Siemens Industrial Copilot can write code for machines in factories and support employees with troubleshooting. The information for the industrial context comes from operating instructions, user manuals, maintenance logs, and more. Additionally, we integrate machine data such as machine status, error codes, and process data. This allows us to analyze the behavior and condition of the machine in natural language and provide corresponding recommendations. By linking the industrial context and machine data with the generic capabilities of LLMs, we make systems more user-friendly and accessible. Personnel in production, as well as in repair and maintenance services, can now perform tasks that were previously beyond their capabilities.

“We empower employees in production to talk to the machine. This pays off, especially in critical scenarios: for example, when production is at a standstill and every second counts” Michael Lebacher, Product Business Owner at Siemens Digital Industries.

From natural language to programming language

LLMs are not only fluent in natural languages, but also in programming languages. The challenge is to ensure that the programs always run correctly at runtime. Rare errors are difficult for the AI to detect. The key to good software therefore lies less in testing a finished piece of code for a long time and more in systematic processes and well-thought-out architectures.

Machines in factories are usually controlled by programs that define the sequence of work steps and the machines' responses to certain events. Although the code generally has a simple structure, it consists of many individual steps and is therefore often tedious to program manually.

The Siemens Industrial Copilot helps to speed things up: programmers can communicate with the system and give instructions in normal language. The AI uses this information to develop a proposal. Since the code is not particularly complex, trained employees can usually easily understand and thus review the AI-generated code. Additionally, there is a simulation function that allows for functionality testing. This enables us to efficiently develop reliable software with the help of AI.

Let′s boost industry standards

Our Industrial Copilot shows how you can enable effective communication and operational support by integrating AI with industrial databases. We also call this “industrial-grade AI,” a quality standard for artificial intelligence that boosts productivity and efficiency across the industrial lifecycle. But this is just the beginning because the Siemens Industrial Copilot shows how much potential AI still has in production and maintenance

This article was initially published on Siemens Stories.