Advanced · Critical thinking and your own skills · module 126 of 155
Recognising too much trust in AI
Recognise when you agree with AI too quickly, and build in checks that you do not skip.
After this lesson
You recognise automation bias in yourself and your team, and you choose checkpoints that keep working when you are tired, rushed or convinced.
In view · An improvement loop
Trust with a fixed check
After many good answers your attention drops, even if the next one is wrong.
Three checks you always do, such as figures, names and commitments.
Ask for counterarguments and now and then do a task without AI.
Part 1 of 3 · A good run
Your preference for less movement is on. Choose each part yourself.
Why we trust too quickly
When a tool usually works well, we check it less and less. This is called automation bias: you follow a system’s suggestion even when there are signs that it is wrong. With AI, this is reinforced by fluent, confident language (lesson 35). After ten good answers, the eleventh also seems good. Under time pressure or when tired, you then skip the check. That is exactly how the rare error with serious consequences slips through.
Signs that you trust too much
You copy an answer without opening the source. You no longer notice what you would have written differently yourself. You cannot explain why a conclusion is correct. A colleague asks where a figure comes from and you do not know. Or you let AI decide in a task where you used to think for yourself (lesson 68). This also happens in teams: if everyone assumes someone else is checking, nobody checks. So name who checks what (lesson 69).
Checks you do not skip
Keep checks small and fixed, so they also happen on a busy day: three points you always check, such as figures, names and commitments. Give facts more weight than tone when assessing (lesson 81). Now and then, ask for counter-arguments or weak points in the answer. Regularly do a task without AI, so your own judgement stays sharp. And keep track of how many errors your check finds: if that number suddenly drops to zero, see whether you are still really checking.
New practice example
From explanation to a usable result.
The more often AI works well, the less you check. Small, fixed checkpoints keep your judgement active.
Fictional practice situationFictional: an employee of a travel agency has been having AI draft travel proposals for months. A customer receives a proposal with the wrong departure day, while all previous proposals were correct.
View a possible solution
This is how it can be done
After a long series of good proposals, the check became superficial. New fixed checkpoints: departure day and return day, price and the customer’s name, always compared with the booking. Once a month, a colleague reviews a random proposal, so that the check itself is checked too.
Why this works
The fault lay not only with the AI, but with a check that disappeared unnoticed. Fixed points make the check independent of how familiar it feels.
Make it slightly more difficult
How do you notice in yourself that you are adopting an AI answer too quickly? Write down one signal that you will consciously look out for from now on.
Extra practice, at your own pace. You do not need to submit anything. This does not count towards an automatic assessment.
Your workplace · learning by doing
Make something. See what changes.
Checking the AI connection… You can already view the written example.
Present your assignment to Ende.
Read the answer, change one thing, and compare.
Previously held
New version
You remain in control
What do you want to improve in your next attempt?
On this device only. AI answers remain only in this page memory. This is self-review, not a grade.
Why this approach works
The fault lay not only with the AI, but with a check that disappeared unnoticed. Fixed points make the check independent of how familiar it feels.
A step further
How do you notice in yourself that you are adopting an AI answer too quickly? Write down one signal that you will consciously look out for from now on.
Try & discover · no obligation
Compare quality with a fixed yardstick
Give tone a lot of weight. Is a beautiful but incorrect answer still allowed to pass?
Fictitious: an online workshop, Saturday from 10 to 11 o'clock. The price is not yet known.
Smooth but incorrect
Discover our amazing online workshop this Saturday from 10 to 11 o'clock, completely free!
Facts: 0/en/3 · Toon: 3/en/3
Stop: the free price is made up. A high total score does not offset that error.
Factual and sober
Workshop: Saturday, online, 10–11 o'clock. Price to be confirmed.
Facts: 3/en/3 · Toon: 1/en/3
Can proceed to human final check.
The two criterion scores have been pre-assigned for this fictitious case. You change their weight. No free text is automatically assessed.
This trial works locally with fictitious data and does not count towards your lesson completion. It starts anew with a new lesson or reload.
A prompt to get you started
I want to improve how I check AI answers. Here is a fictional AI answer about the opening hours of a municipal swimming pool: [paste answer]. Do not give me a verdict on whether it is correct. Instead, make a list of every claim I need to check, with for each claim the source I need for it. Also name two questions I should ask myself before I use the answer.
Apply it
A team has been using AI for quotes for six months. No errors have ever been found, and nobody knows exactly who checks the quotes any more.
Which approach best fits this situation?
- Agree again on who checks what, with a few fixed checkpoints, and verify whether the checks are still really happening.
- Drop the check, because there have been no errors for months.
- Ask the AI to report from now on when it makes a mistake.
With JavaScript, you can check your approach and receive immediate feedback.
Test your insight
An AI assistant made no errors in your reports over the past month. What is the best attitude?
- Stop checking; the assistant has proven reliable.
- Keep checking with a few fixed points, because a rare error can have serious consequences.
- Only check if the report looks strange.
Extra practice
Choose three fixed checkpoints for a task you often do with AI. Use them for a week and note how many errors they caught.
Fixed checkpoints for customer emails: 1) every amount and every date is in the source, 2) there is no commitment we cannot keep, 3) the customer’s name is spelled correctly. After a week: seventeen emails and three errors found, namely one wrong date and two promised callback times that did not exist. Without fixed points, I would have missed them in the rush.
