AI: Who controls the conversation? (and why it should be you)
by Robin Huggins, Head for Data

A few weeks back, thanks to the good folks at CeeD, I was asked to speak at the Scotland Manufacturing & Supply Chain Conference & Exhibition.

I'm not a technical person. I don't build AI models, I don't write code and, despite working for a data, automation and AI company, nobody at Head for Data would let me anywhere near the actual technology. Quite right, too.

To give you an idea of my technical capability, I can play my PlayStation, know a wee bit of my smartphone but I am expressly forbidden to touch the buttons for the TV at home.

And yet, I've spent more than 25 years working around very technical people, generally occupying a slightly odd space between business and technology. I'm the translator, which mostly involves asking annoying questions like "What does that mean?", "Why does it do that?", "Why have you used that word?" and, most importantly, "What does this actually mean for the business?"

I'm genuinely like an annoying child when I get going. I think more of us need to be.

We've made AI sound more complicated than it needs to

AI has a language problem.

Spend any time around it and you'll hear about frontier models, prompt engineering, sandbox environments, Retrieval-Augmented Generation, Agentic AI and a growing dictionary of other terminology.

Some of it is perfectly legitimate. Technical people need technical language when they're talking to other technical people about technical things, but the rest of us are supposed to be using this stuff too.

So, at the conference, we translated some of it.

A frontier model is one of the newest and most capable versions available. Prompt engineering is refining the questions and instructions you give AI. A sandbox is an isolated, safe space. RAG, or Retrieval-Augmented Generation, is giving AI the right reference material to work with. Agentic AI is software that can take steps and do things for you.

None of that sounds quite so clever when you say it like that, does it?

That is why I turned part of my presentation into a deliberately cheesy 1970s gameshow called "AI'll Name That Jargon!" There was a contestant, there were prizes and, whenever another piece of AI terminology appeared, the audience had to shout:

"Big chap - what does it mean?"

Ridiculous? Absolutely, but there was a serious point behind it.

Who controls the language?

I studied sociology at university and became really interested in language and power. One of the thinkers I came across was Pierre Bourdieu, who wrote about the relationship between language and authority.

You don't need to disappear down a French sociology rabbit hole to recognise it. Loads of professions have their own language - medicine, law, finance, academia, the military and the civil service all spring to mind.

Sometimes they need to. Precision matters, but specialist language can also exclude people and, sometimes, I think that's the point.

There. I've said it.

I think some of the language around AI is unnecessarily protected, some of it makes people sound clever and, controversially, there's commercial value in making something appear more complicated than it really is.

If I convince you that AI is terribly complicated and I'm one of the special people who understands it, there's a fair chance you'll stop challenging me and start trusting me to make decisions on your behalf.

Now I control the conversation.

And that's more important than simply being irritated by jargon. When technologists control the language, they can end up controlling the question too, and the first question often determines what follows.

"What can AI do for us?"

I hear variations of this all the time: "We need to get AI into the business", "Everyone else is doing AI", "Where should we be using AI?" and the classic, "What can AI do for us?"

STOP IT!

Ask "What can AI do?" and you've already decided AI is the answer, but we haven't even agreed what the problem is yet.

I'd much rather hear "What needs to improve?"

Now we're talking, because you know the answer to that.

You know your business better than the AI expert does

This is one reason I loved having this conversation with people from manufacturing and supply chain businesses.

You know where work waits, where mistakes happen, where information gets stuck and where skilled people waste time doing repetitive jobs. You know the spreadsheet everybody is terrified of touching because only one person knows how it works, the process everyone bypasses because the workaround is easier and the bit where somebody says, "Ask Dave, he knows how to do that".

And you know what happens when Dave goes on holiday.

That's the valuable knowledge, not knowing what RAG stands for.

Technologists have expertise too, obviously, but it's different expertise. We need both.

At Head for Data, we call our approach Business Reverse Engineering. It's a grand name for something deliberately straightforward: tell us what actually happens.

Not what the process document says happens or what management thinks happens, but what the people doing the job every day say actually happens.

Then find the friction. What's slow? What's annoying? What keeps breaking? What's duplicated? Where are people wasting time? What information can't you see when you need it?

Then decide what better looks like.

Only after that do we talk seriously about technology. Sometimes the answer is AI and sometimes it isn't.

Sometimes the solution is cheaper, simpler and considerably less exciting than AI, and that's absolutely fine. The objective was never to "do AI", it was to make something better.

Ask the stupid question

Actually, they're not stupid questions.

I've watched intelligent, experienced business folk become hesitant because somebody walked into the room speaking a language they didn't understand. Rather than risk looking stupid, they nod along and start giving away control of the conversation.

Don't.

If somebody proposes an AI solution, ask what problem it solves, what decision it improves, what process it changes, what risk it creates and what evidence says it's worth doing.

If the answer contains another three words you don't understand, ask what those mean, too.

You're not being difficult, you're doing governance.

AI needs site rules too

There's something else manufacturing and supply chain people understand extremely well: you don't mess about with safety.

When I visit a manufacturing site, I don't get to wander through the front door because I've announced that I'm innovative. There are rules, I might need a safety briefing and PPE, there are places I can go and places I can’t, and somebody accompanies me the whole time.

Nobody considers that anti-visitor. It's just sensible.

Why should AI be different?

Before introducing it into a business, decide what it's allowed to access and what it isn't. Think about the data it's using, who checks what it produces and what happens when it gets something wrong.

Because it will.

I use AI every day. I used it to help build the presentation for the conference and to help write this article. It makes mistakes, assumptions and, occasionally, does something completely mad.

My job is to remain in charge.

I don't particularly like the phrase "human in the loop" because it makes me sound like I'm part of the machine's process. I prefer human in the lead - it's working for me, not the other way round.

So, who controls the conversation?

If we're serious about AI adoption across Scotland's economy, the answer can't be "the people who understand AI terminology".

We need the people who understand the work at the table, too.

That's the operations manager who knows where the bottleneck is, the engineer who knows why the official process doesn't work in reality, the person who spends Friday afternoon manually reconciling something that shouldn't need manually reconciled and the person who's been saying for three years that there's a better way to do this.

Those people don't need to become AI experts. We need to bring technology into their conversation.

Those of us who work around AI have a responsibility, too. If we genuinely believe this technology should be accessible and useful across the economy, we can't build a linguistic wall around it and then charge everyone else admission.

Maybe AI literacy isn't about teaching everybody to speak like technologists. Maybe some technologists need to learn how to speak like everybody else.

So, next time somebody tells you they're going to "leverage frontier model capabilities through an agentic RAG architecture", or whatever this week's phrase happens to be, don't nod politely while wondering what the hell they're talking about.

Ask them:

"Big chap - what does it mean?"

And if the answer still doesn't make sense, ask again.

If AI is going to have the impact on our economy we're constantly being told it will, understanding it can't be the preserve of a small group of people who know the secret language.

The rest of us need to be part of the conversation, too.