AI and the Ship of Theseus
- index
- 5 days ago
- 8 min read
When Every Part of a Policy Has Changed, Is It Still the Same Truth?
The Ship of Theseus is an ancient thought experiment about identity, change and continuity.
It asks a simple question that becomes increasingly awkward the longer you think about it.
Imagine a ship is preserved over many years. As parts decay, they’re replaced one by one. A plank here. A rope there. Then the sails. Then the mast. Then the rudder, the deck and eventually every part of the original ship has been swapped out.
So, is it still the same ship?
If you replace one plank, probably yes.
If you replace half the ship, maybe still yes.
If you replace every piece of it, one at a time, it gets trickier.
At what point did the original ship stop being the original ship?
It’s the sort of philosophical question that starts as a harmless conversation and ends with everyone regretting they didn’t just talk about the weather.
I found myself thinking about it this weekend while restoring a galleon replica I bought.
It’s a beautiful model, full of sails, rigging, flags, tiny ladders, wooden detail and enough delicate string that made me question every life choice that led me to this moment.
Restoring it wasn't simply a case of “make it look nice again”. I had to work out what’s original, what’s structural, what’s decorative, what’s missing, what’s been repaired badly before and what absolutely must not be touched unless I fancied spending the following three hours untangling miniature rigging while muttering things not suitable for a corporate blog.
And somewhere between adjusting a sail and deciding whether a rope was functional or decorative, the connection landed.
Enterprise knowledge has a Ship of Theseus problem.
Knowledge changes plank by plank
Most organisational knowledge doesn’t break all at once.
It changes gradually.
A policy is updated.
A paragraph is rewritten.
A legal clause is added.
A regional team adapts the wording.
A PDF version is saved elsewhere.
A training deck quotes an earlier version.
A contact-centre script gets amended separately.
A chatbot knowledge source is refreshed from one system but not another.
A “temporary” workaround becomes permanent because temporary, in corporate language, can sometimes mean “until the sun burns out”.
No individual change feels dramatic.
Each one probably made sense at the time.
But over months and years, the organisation ends up with several versions of the same knowledge, all carrying the same flag but sailing in different directions.
The document still says Customer Refund Policy.
The process still has the same name.
The knowledge article still sits in the right folder.
But is it still the same truth? Or is it a familiar-looking ship built from different parts, different eras and different interpretations?
That’s the problem.
AI doesn’t know which ship is seaworthy
This matters because AI doesn’t automatically understand the history of your knowledge estate.
It doesn’t instinctively know that one version replaced another.
It doesn’t know that a PDF is obsolete if the PDF still exists and looks official.
It doesn’t know that a regional exception only applies in one market.
It doesn’t know that the training deck was copied from a policy that has since changed.
It doesn’t know that the chatbot source is technically current, but missing the operational caveat everyone in the contact centre knows by heart.
AI sees content.
If several pieces of content look relevant, it may retrieve them, summarise them and present an answer with great confidence.
That answer may sound clean, polished and helpful.
It may also be stitched together from the knowledge equivalent of spare planks, old sails and a suspiciously modern rudder.
That’s not really an AI problem.
It’s a knowledge lineage problem.
The AI hasn’t failed because it found information.
The organisation has failed to show which information still has authority.
The real issue is authority, not just access
A lot of AI projects begin with the ambition to “connect AI to our knowledge”.
That sounds sensible.
But the more important question is: Which knowledge?
The approved policy?
The archived PDF?
The customer-facing version?
The internal version?
The regional version?
The version support teams actually use?
The version legal recognises?
The version that appears first in search because it’s been opened 12,000 times and refuses to sink?
Access is not the same as trust.
An AI system can access a document without knowing whether that document is still valid, complete, current or authoritative.
That’s the difference between a content repository and a knowledge system.
One stores things. The other helps people and AI understand what can be relied upon.
You can’t fix the whole ocean at once
This is where organisations often get stuck.
They look across SharePoint, Confluence, ServiceNow, PDFs, intranets, wikis, training decks, local folders and chatbot sources and realise the knowledge estate is huge.
Then the temptation is either to boil the ocean or to do nothing.
Neither is helpful.
Trying to fix everything at once becomes overwhelming. It turns into a vast transformation programme with steering committees, spreadsheets, workshops, heroic optimism and a delivery date that quietly moves into the next financial year.
Doing nothing is worse.
The better approach is nautical, not oceanic.
You don’t try to repair every ship in every sea at the same time.
You start with one harbour.
One knowledge base.
One category.
One critical business issue.
One set of policies that are causing customer friction, compliance risk, operational cost or AI unreliability.
You make that area safer, clearer and more trustworthy.
Then you move to the next harbour.
That’s how knowledge improvement becomes manageable.
Not because the problem is small, but because the approach is staged, sensible and focused.
Restoration requires judgement
Back to my galleon.
When you restore a model ship, you don’t just replace everything that looks old.
Some worn parts are still doing their job.
Some old features are part of the character.
Some broken pieces need careful repair.
Some replacements are necessary.
Some additions from a previous owner may be useful.
Some, frankly, make you wonder whether they were installed during a power cut.
The point is that restoration requires judgement.
Enterprise knowledge is the same.
Not every old document is wrong.
Not every duplicate is useless.
Not every local variation is a contradiction.
Not every workaround should be deleted.
But every important piece of knowledge needs context.
Who owns it?
What does it replace?
What depends on it?
Where else does it appear?
Is it current?
Is it approved?
Is it safe for AI to use?
Is it still part of the ship, or just floating nearby looking official?
Without that context, organisations end up with knowledge estates that look impressive from a distance but become fragile when people, customers or AI try to rely on them.
How index helps organisations navigate it
index is built to help organisations understand and improve the knowledge sitting beneath AI, search, customer service and operational decision-making.
We don’t begin by trying to boil the ocean.
We begin by helping organisations identify the harbour that matters most.
That might be a customer-service knowledge base causing inconsistent answers.
It might be a compliance category where policy drift creates risk.
It might be a product area feeding unreliable chatbot responses.
It might be a critical operational process that exists across too many versions, formats and teams.
Through Scope, we define the area that needs attention and why it matters commercially. We look at the systems involved, the business impact, the users affected and the outcomes the organisation needs.
Then Scan analyses that focused knowledge area to surface the issues that are usually hidden below deck:
Duplicate and near-duplicate content
Contradictory guidance
Outdated documents
Broken links
Missing ownership
Unclear terminology
Poor machine readability
Superseded content that still looks active
Knowledge that may be unsafe or unreliable for AI
Scan helps answer the question:
What is this harbour actually full of?
Then Solve supports structured remediation.
That means resolving contradictions, consolidating duplicates, assigning ownership, improving content quality, clarifying canonical sources and retiring information that should no longer influence decisions.
Not as a chaotic clean-up exercise.
Not as a heroic attempt to fix the entire enterprise in one go.
But as a controlled, evidence-led programme of improvement.
One harbour at a time.
Finally, Sustain measures and manages progress.
It shows whether knowledge quality is improving, whether risks are reducing and whether the content estate is becoming more reliable for people and AI.
Because leaders don’t just need reassurance that work is happening.
They need evidence that the organisation is becoming more trustworthy.
The question every AI project should ask
The Ship of Theseus asks whether a ship remains the same when all its parts have changed.
Enterprise AI asks a similar question:
When knowledge changes over time, how do we know which version is still true?
That question matters because AI will not wait patiently at the dock until your knowledge estate is perfect.
It will use what it can find.
And if what it finds is fragmented, duplicated, outdated or contradictory, it may produce confident answers from uncertain sources.
That’s when businesses discover that their AI problem is really a knowledge problem wearing a captain’s hat.
Final thought
Restoring the galleon reminded me that keeping something seaworthy isn’t about replacing everything.
It’s about understanding the structure, respecting what still works, repairing what doesn’t and making sure the whole thing holds together.
Organisational knowledge works the same way.
You don’t need to fix the whole ocean at once.
You need to know which harbour matters first.
You need to understand what’s in it.
You need to repair the vessels that are most critical.
And you need a way of proving they’re safe to sail.
Because AI can move quickly.
It can answer quickly.
It can scale quickly.
But if the knowledge underneath it is made from conflicting parts, missing context and old versions drifting around in the fog, it may still sail confidently in the wrong direction.
Possibly with beautiful rigging.
Possibly with an impressive dashboard.
But still, unfortunately, towards the rocks.
index helps organisations detect, resolve and measure the knowledge problems sitting beneath AI and everyday operations.
Through Scope, Scan, Solve and Sustain, we help organisations improve their knowledge estate in a focused, manageable way, one harbour at a time.
Set a Course for AI Readiness
We’ll be exploring this in more detail in our upcoming index webinar, where we’ll look at how organisations can deploy AI successfully without losing sight of the knowledge quality, governance and operational foundations that make it work.
Our next session is August 7th:
Deploy AI Priorities in Your Organisation Successfully
You can sign up here:
We’ll be running more of these sessions too, because AI readiness isn’t a one-off conversation. Like the knowledge estate itself, it needs regular attention, practical steps and a sensible route forward. One harbour at a time.
Find out more at www.index-ai.net
Contact us at contact@index-ai.net
A Note on the Galleon
In case you were wondering, the ship itself appears to be a Spanish galleon replica, likely based on the San Francisco II, a model representing the kind of armed ocean-going galleon associated with Spain’s maritime power in the 16th and 17th centuries.
The Castile and León-style flags, high sterncastle, multiple masts, decorative shields and overall build all point in that direction.
I’m not claiming it’s a perfect reconstruction of one fully documented vessel, but it’s very much in the world of Spanish treasure ships, long voyages, imperial trade routes and the sort of nautical history that makes you want to say “set a course” even when you’re only moving it from one shelf to another.
Comments