Article · 08 OCT 2026 · 4 min read
Why Understanding Is the Next AI Problem
Karina Strobl
Founding advisor
We use AI every day and there is something interesting about the way most of us still use it.
We ask a question, get an answer and then realise it doesn't know something important. So we add context, explain what changed, correct an assumption and clarify what we're actually trying to achieve. We also have to point out that something it found is outdated.
Eventually, the answer improves. But by then, we've done a surprising amount of the work ourselves... not the writing, searching or analysis, but providing the understanding that was missing.
We already carried context the system didn't have: the history, the objective, what mattered, what had changed and which assumptions no longer held.
The AI didn't, and that's where the problems start.
Problem 1: The question already assumes understanding
One of the most common pieces of advice about AI is to ask better questions.
Sure, that sounds easy enough but to ask better questions, you first need to know what's actually worth asking.
Say you want to enter a new market, launch something, understand a risk, fix a problem or find an opportunity. You don't usually start with a perfect list of questions. You start with an objective.
Then you need to figure out what you actually need to understand to get there. What matters? What doesn't? Where should we look? What's missing? How does it all connect?
By the time we formulate a really good question for AI, we've probably already done quite a bit of the hard thinking ourselves.
Problem 2: A good answer can still be hard to trust
Then there is the answer itself. AI can give us something incredibly convincing, but what is it actually based on?
What is the evidence? What was inferred? Is it relying on an assumption? How reliable is it? When was it true? Where are its limits?
Without that, we may have a great answer without really knowing how much we should trust it. So we're back to checking the sources, validating the assumptions and figuring out what's missing.
Problem 3: What was true yesterday may not be true today
Even well-supported evidence doesn't stay relevant forever. Markets move, regulations change, customers change direction, people leave, products launch or new information appears. This means that understanding has a shelf life.
And keeping it current isn't just about remembering new information. New information might change what everything else means.
A new regulation might completely change how attractive a market is. A customer changing direction might make something we thought was an opportunity irrelevant. One new piece of evidence might challenge an assumption that an entire conclusion was built on.
So it's not enough to know something new. We need to understand what it changes.
That's the difference between remembering new information and understanding what it changes.
Acting without understanding
All of this becomes more important as AI moves from answering questions to recommending, monitoring and acting for us. Because if we're still the ones providing the context, deciding what matters, connecting the dots and checking whether the information is still valid...
So how much does AI actually understand about the thing we're asking it to act on?
Maybe there is a layer we've skipped.
Objective → Understanding → Decision → Action
We've spent the last few years making AI quite good at the things on the right, but have we overlooked the importance of the left side? The next big problem might be the bit we've quietly been doing ourselves all along.
Understanding.