Why AI keeps getting the blame for problems that actually start much further upstream

I was at a conference recently where people were talking about taxonomy, ontology, metadata, semantics and all the other words that can make someone’s eyes glaze over halfway through a sentence.

And yet the funny thing is, the core issue is actually dead simple.

Most organisations do not really have an AI problem. They have a knowledge problem.

AI just happens to be the thing exposing it.

That is what I kept coming back to. Everyone wants AI to give smart, reliable, explainable answers. Fair enough. But then you look underneath and find the business is sitting on years of duplicate content, conflicting versions, outdated policies, badly labelled documents, inconsistent naming, patchy ownership and information spread across five different systems that barely agree with one another. Then people act surprised when the AI gives a shaky answer with a straight face.

That is a bit like blaming a sat nav for taking you the wrong way when the map is out of date and half the road signs are missing.

One of the more interesting ideas raised at the conference was this notion that AI is increasingly learning from AI-generated content.

In plain English, the machines are starting to feed off their own leftovers. These models were first trained on the internet, which already contains all the bias, noise, contradiction and rubbish you would expect from the internet. But now more and more of the content out there is synthetic, generated by the same types of models. So what you end up with is a loop. A copy of a copy of a copy. It starts to drift. The link back to reality gets weaker. That is exactly why provenance matters so much. You need to know where knowledge came from, who owns it, whether it is current, whether it has been approved, and how much trust you should place in it.

Without that, AI is not really reasoning. It is guessing with confidence on top of messy foundations.

That is where taxonomy and ontology suddenly stop being niche subjects and start becoming very practical.

The easiest way to describe taxonomy is this: it is how you stop your organisation’s knowledge turning into an unusable garage full of boxes with no labels on them.

It is structure. It is consistency. It is deciding what things are called, how they are grouped, what terms mean, and how people should find them later. It helps create order across large, messy estates of knowledge where different teams have usually evolved their own ways of describing the same thing.

And that matters because once you scale beyond a certain point, chaos compounds. One team says “customer issue”. Another says “incident”. Another says “case”. Another says “service event”. They may all mean the same thing, or they may mean four slightly different things. If nobody sorts that out, systems get confused, users get frustrated, and AI happily inherits the mess.

One phrase I heard that I really liked was that taxonomy is a bridge.

That is exactly right. On one side, you have raw content, policies, documents, assets, records, procedures, tickets, product information and operational know-how. On the other side, you have decisions, actions, service delivery, automation and AI outputs. Taxonomy helps bridge the gap. It gives shape to the knowledge so it can actually be used properly rather than just stored.

But even that is only part of the story, because real life is messier than neat categories.

That is where ontology comes in, although even saying that in a pub sounds like a good way to drink alone. Still, the idea is useful. A lot of organisations try to organise information in rigid hierarchies, like old folder trees. The problem is that real things are related in different ways. One thing can be part of another thing, associated with it, used alongside it, owned by it, governed by it, triggered by it or dependent on it. Those are not the same relationships, and treating them as if they are creates blind spots.

A simple example would be this:

• A lens is part of a camera.

• A lens cloth is not part of a camera, and it is not a kind of lens either. It is related, but differently related.

Sounds obvious when you say it out loud, but systems often fail on exactly that kind of nuance. If your structure cannot handle different kinds of relationships, then useful information gets buried or disconnected.

And that is why so many organisations struggle with search, self-service, automation and AI.

Not because the front-end tools are useless, but because the structure underneath is weak.

This is exactly the problem we built index to solve.

We are not trying to be another shiny AI layer that pretends the underlying mess does not matter. We focus on the knowledge layer underneath it all, because that is where trust is either built or broken.

index Scan surfaces the issues that undermine reliable outcomes across the knowledge estate. Duplicate content. Contradictions. Outdated material. Broken links. Weak metadata. Poor machine readability. Unclear ownership. Structural inconsistency. All the things that quietly sabotage search, copilots, service operations and compliance.

Then index Solve helps organisations do something about it in a governed, auditable way.

Not random clean-up. Not one heroic workshop that gets forgotten three weeks later. Proper remediation with ownership, process, workflow and accountability.

But this is the important bit. Tools alone do not fix meaning.

A platform can tell you where the problems are. It can measure them, prioritise them and route them. That is valuable. But it still takes people to make sense of the mess and decide what good looks like.

That is why our model is not just product-led. It is product plus expertise.

You need knowledge managers who understand governance, operational risk, ownership and how knowledge supports the business day to day. You need taxonomists who know how to structure language and categories so people can actually find and use what they need. You need ontologists and semantic specialists who understand relationships, meaning and how to model the real world without forcing it into simplistic boxes. You need metadata specialists, information architects, content strategists, knowledge analysts and content governance leads who know how to apply all of that in practice across large estates, competing teams and imperfect source systems.

That combination matters.

Because the truth is, a lot of technology providers like to imply that tooling is enough. Buy the platform, switch on a feature, let AI sort out the rest.

I don't buy that.

I've seen too much of the reality. I've seen systems sold as if keywording is easy by people who've clearly never had to do proper keywording at scale. I've seen digital asset and knowledge platforms marketed as if structure magically appears once the software is installed. It doesn't. Meaning needs design. Governance needs ownership. Structure needs thought. Language needs judgement.

And none of that becomes less true just because AI has arrived.

In fact, the opposite is true. AI has made all of this more urgent.

Because now the cost of bad knowledge is no longer just that someone can't find a document. Now the risk is that an employee, a customer or an agent gets a polished but wrong answer from a machine that sounds authoritative. That's a much bigger problem.

So when people ask me where AI and governance meet, my answer's simple. They meet in the knowledge layer.

That's where standards matter. That's where ownership matters. That's where provenance matters. That's where good taxonomy, good metadata and well-modelled relationships matter.

Get that layer right and you massively improve your odds of getting useful, trustworthy, explainable outcomes from AI. Ignore it and you're basically building on sand.

That's also why I think taxonomists, ontologists and knowledge professionals are far more important than most organisations realise. They're not just categorising content. They're helping define the structure of organisational meaning. They're making information usable. They're creating the conditions for better decisions, safer automation and more trustworthy AI. That's not back-office admin. That is core infrastructure.

So yes, the conference talked about taxonomy and ontology. But what it was really talking about was something much bigger.

1. How do we stop organisations drowning in their own information?

2. How do we make knowledge usable, governed and trustworthy?

3. How do we stop AI amplifying disorder and instead make it useful?

That's the real conversation.

And from where I sit, that's exactly why index exists.

Not to plaster over the cracks, but to help organisations see what's broken in the knowledge estate, fix it properly, and build a foundation strong enough for AI, search, service and decision-making to stand on.

Because in the end, trusted answers don't start with the model.

They start with the knowledge.