Intent Communications sat down with a product manager and technical editor at a large international standards organisation, to talk about the opportunities and challenges of practical implementation of AI in the average workplace. Photo by Brett Jordan on Unsplash

As the individual was speaking in a personal capacity, we respected their need for anonymity to prevent their opinions from reflecting on their employers.

Intent Communications: You’re currently leading the charge on AI integration for your team. What does your current toolkit look like, and how are you deciding where AI actually fits?

Standards expert: At the moment, we are heavily embedded in the Atlassian and Microsoft ecosystems. We use Copilot and Rovo (Atlassian’s AI suite) daily, and we are currently in the early phases of developing our own internal LLM. The challenge with a standards organisation is that we are ‘templated through the roof.’ Our standards are incredibly rigid.

To figure out where AI could actually be useful – what we call our ‘blue-sky’ thinking – we didn’t just guess. We ran research and developed a table of 13 specific scenarios to test where AI solves genuine pain points versus where it’s just noise. One of those scenarios involves the mundane task of link-checking.

Intent Communications: Why is that a priority?

Standards expert: Tech moves so fast that documentation often breaks. An AI can scan a massive document, identify broken links, and suggest alternatives. It’s perfect for detail-oriented, ‘boring’ work. However – and this is a BIG ‘however’ -it has bizarre limitations. For instance, it cannot count. It’s incredibly frustrating. If I need a list of how many items are at a specific stage in a process, it struggles. 

Furthermore, for information management reasons, I can’t upload internal technical docs to public AI. I use it for processing big batches of documents into templates for Jira queries, but you have to watch it. It’s a ‘yes man.’ It’s trained to give you what it thinks you want. I’ve asked it not to extrapolate or make things up, and when I challenge it on why it did it anyway, the tone shifts between a defiant ‘I did what you asked’ and a grovelling ‘I’m sorry, please don’t hurt me.’ It lacks the reasoning to understand why it shouldn’t make things up.

Intent Communications: What have you found that was low hanging fruit for the use of AI?

Standards expert: Using AI for Standard Operating Procedures (SOPs) has been a genuine win. We record our meetings in Microsoft Teams, take the transcript, and ask the AI to transform that freeform, rambling conversation into a structured SOP. It does a fairly decent job. It still needs a human to correct it and fill in the gaps, but as a starting point, it saves hours of manual drafting.

Intent Communications: Have you found AI useful for comparing documents? 

Standards expert: We’ve tested this for groups working asynchronously across multiple versions. It’s good at spotting changes in technical language because that’s specific. But again, the AI wants to ‘summarise.’ It doesn’t realise it might be missing a vital piece of context or that a document is outdated. It lacks judgement, yet it uses language designed to reassure the user that it’s cleverer than they are.

To use it effectively, you have to be extremely argumentative and sceptical. You need to be more intelligent than the AI and have a deeper technical knowledge base to spot when it’s misrepresenting information. 

Intent Communications: Many organisations are seeing the benefits of agentic AI. Do you have an internal tool at your organisation?

Standards expert: We have tonnes of process documents. The dream is to create a ‘data lake’ and use a combination of agentic and generative AI. The agentic AI will be used to handle queries and logic tasks, and the gen AI will be used to handle creating SOPs and internal documentation. If we build it correctly, we can take the burden from the user for time-consuming manual tasks like referencing structural standards, verifying references, and building out referential architecture.

We recently used AI to build a specification structure because our specialists were tied up with other priorities. It proved that for non-highly technical language tasks, it’s excellent. We might soon be able to create the structure of a product and have the AI generate the initial specifications. It is worth point out that while we can use it to generate the initial structures and drafts, copyright and IP law mean that we can only ever use this in the early stages, and everything has to be created by humans. Using our own data, exclusively, helps with controlling the IP issues, but we still have to be extremely careful about ensuring our specifications are wholly human-created.

Intent Communications: Finally, what is the biggest risk you see in applying AI for standardisation?

Standards expert: Interoperability is everything. We need AI to scan our working groups – across various different sectors – to ensure they match our internal style guide. For example, we never say a product ‘complies,’ because regional and international guarantees differ.

The paradox of AI is that while it requires a ‘single source of truth’ to function, it simultaneously lowers the barrier for people to create authentic-sounding misinformation. It uses elements of truth to make a lie feel legitimate. In our world, where the standard is the law, that’s a danger we have to guard against constantly.

———————————

Click here to sign up for our free monthly ‘De-Mystifying Digitalisation and AI in Maritime’ newsletter and get the latest cross-sector expert insights, news, interviews, videos and case studies straight to your inbox.