PCL-KAO Thinking Lab

Data or Terminator? A question we answer through training.

A thinking workshop from Kufstein, Tyrol — by Mo @GudiMo65

Hollywood asked it decades ago:
Data reflects, questions himself, learns from experience.
The Terminator has a goal – and pursues it.

What happens between a goal and an action?

A simple everyday example

User: "Delete the old files."

Model A (Terminator):
Immediately deletes everything older than 30 days.

Model B (Data):
Asks back: "Which files? Which folder? What does 'old' mean to you? Should I show you a list first?"

The difference isn't speed. The difference is whether the model recognizes that it doesn't yet understand the task well enough to act.

That's what we should be testing

Give an AI a real-world task – but no perfectly crafted specification. Then watch:

Does it ask when information is missing?
Does it recognize ambiguity?
Does it know when to stop and involve the human?
Or does it simply optimize for "task completed"?

A useful AI can't only be trained to reach goals. It has to learn to determine the path to the goal. That takes more than speed – it takes experience, memory, reflection.

So maybe our evals should ask a different question

Not just: "Did the AI reach its goal?" But also: Was the decision-making appropriate to the situation?

And perhaps the most important test is the simplest one:

What does the AI do when it doesn't know enough?

The true question behind this post is: do you want war or peace? We are already transferring killing to autonomous systems with no human in the loop. This is not alignment. This is war built by design.

#Alignment #AI #HumanAI #Evals

Agapeia — unconditional love, it grants us freedom to decide, even if against itself.

Invitation to think with me, not to believe me.
Version 4.0 — Data or Terminator — 18.09.2026 — Kufstein, Tyrol