AI is not intelligent — we have to be

Reflections from yesterday's BI webinar "More Than Words: How to Know How AI Can (and Can't) Help You"

Yesterday's webinar from BI with Auke Hunneman and Jan Ketil hit an important point:
AI creates great value – but only when we understand the limitations.

And I got to experience it myself, the same day:

I asked an AI model to create a picture of me in an office setting. The result?
It became "me"… but not me.
Something was off. The face, the expression, the form - everything was somehow almost right, but not right enough. And precisely this is the core of what many who use AI should understand and challenge:

1️⃣ AI predicts - it does not understand
LLMs (and image generators) guess the most likely outcome based on huge amounts of data. They don't know what I look like. They don't understand identity

Therefore, there is often something that looks similar, but does not match.

2️⃣ AI gives poor ROI when the problem does not fit
A distinction is made between three types of challenges:
• Problems (predictable): forecasts → AI is strong
• Secrets (hidden patterns): channel drivers → AI + expertise
• Mysteries (unpredictable): interdisciplinary campaigns → people must coordinate

To create a real picture of me?
It falls between Secret and Mystery:
AI does not have enough information → the result is uncertain → the human must evaluate

3️⃣ The Cynefin framework
Clear: Tasks with summary → automate
Complicated: Analyzable, but demanding → AI + experts
Complex: Must be tried out → experiment
Chaotic: crisis, little time → act fast
Aporetic: unclear problem → stop and ask new questions

Getting the AI ​​to "understand Marius"?
→ Aporetic + Secret: There is no facit for my face in the model

4️⃣ Intelligence is more than words - and more than pixels
AI can deliver text and images. But it cannot:
- read the room
– understand motivation
- build teams
- create culture
- recognize me in a picture
Here man is superior.

5️⃣ Trust determines the result
When people see a picture that is "supposed to be me" but doesn't look like it, they lose confidence. The same applies in organisations:
Without trust in both numbers and each other, collaboration stops - no matter how advanced the AI model is

What does this mean for managers – and for everyone who uses AI?
It is a core competence to:
• understand when AI fits
• understand when human judgment must take over
• combine data + people for real value
• challenge inconsistent AI results

If AI can't recognize my face - why should I blindly trust what it suggests?