Reflections & stories
Ideas, experiences and learnings about education, technology and coexistence.
Showing 1597–1608 of 1,735 articles
Safety begins when reporting an error is possible — before an automatic response: Everyday artificial intelligence
The safety of an automatic response depends on errors being reportable, correctable, and explainable without punishment. Everyday AI needs clear channels, human oversight, and repair.
Read article
What works for most people can exclude someone — before an automatic response: Everyday artificial intelligence
An automatic response designed for the majority case can leave out people who write differently, need more time, or do not fit its assumptions. Everyday AI requires alternatives and human review.
Read article
A rumor travels fast when it tries to care — before an automatic response: Everyday artificial intelligence
Rumours about AI responses often begin in care, but they can turn a doubt into alarm. A responsible response verifies, explains limits, and keeps a human route open.
Read article
Conflict is not resolved by adding another screen — before an automatic answer: Everyday artificial intelligence
When an automatic answer conflicts with a real need, adding steps and screens usually hides the problem. What is needed is explanation, review, and a human route with power to decide.
Read article
The exception reveals the quality of the system — before an automatic answer: Everyday artificial intelligence
Automatic answers often work for anticipated cases. Their quality is revealed when someone does not fit: an exception should open understanding, review, and a real human route.
Read article
A useful number can still lead us to the wrong place — before an automatic answer: Everyday artificial intelligence
A metric can organise work while also hiding what matters. With automatic answers, sound measurement requires looking at decisions, harm, and alternatives, not only speed or acceptance.
Read article
Accepting does not always mean being able to choose — before an automatic answer: Everyday artificial intelligence
Accepting an automatic answer does not, by itself, show that a free and informed choice was available. A suggestion from an assistant, a classification on a screen, or an answer generated during an application can arrive
Read article
Manifesto: the first question must arrive before the tool
A responsible automated response starts with the question that protects people, not with the tool that promises speed.
Read article
Participation changes the question and improves the answer: Screens and attention
Participation reveals what a screen demands from attention and lets decisions be corrected before everyday friction becomes exclusion.
Read article
Care also means not monitoring everything: Screens and attention
Caring for a tool means noticing harm and offering support without turning every pause, error, or doubt into surveillance data.
Read article
Supervision requires the power to stop and correct: Screens and attention
Supervising a tool means noticing when it exhausts attention and having real authority to stop it, correct it, or restore a safe alternative.
Read article
After launch, the important work begins: Screens and attention
Launching a tool does not prove it works well under pressure. The important work starts by observing how it distributes attention, uncertainty, and the ability to correct.
Read article