Welcome to Reckonings, a free weekly newsletter for readers of At a Cost and The Quiet Cost. Each issue gives you a snapshot of the biggest AI stories and studies from the past week, then looks beneath the obvious benefits and harms to ask what may be changing in us.
We will also offer games and exercises that keep memory, attention, judgment, writing, and connection in practice. The goal is not to reject AI. The goal is to keep the whole trade in view.
01
When the help disappeared, people quit sooner.
A preprint reporting three randomized experiments with 1,222 participants found that people who used AI assistance performed worse when they later worked without it. They also showed less persistence. When the help was removed, they gave up sooner.
The important detail was how the AI helped. Hints that required people to continue doing the work were far less corrosive than complete answers.
Read the original preprintThe useful distinction is not AI or no AI. It is support or substitution. Help can preserve the work that forms us, or quietly remove it.
02
The careful answer to “Is AI making us stupid?”
A peer-reviewed review in Trends in Cognitive Sciences examined the evidence on cognitive offloading. Its conclusion was more measured than the headlines. Specific skills can erode when they are repeatedly handed away. Foundational capacities appear more resilient. Design often decides which direction we go.
Tools that ask us to think, choose, and revise can protect ability. Tools that complete the thinking for us can make the skill easier to avoid.
Read the peer-reviewed reviewCapability is not one switch that AI turns off. It is a collection of capacities that remain strong when they are used and become optional when they are not.
03
The bot sounded more empathic. Only the human reduced loneliness.
In a randomized trial with 296 participants, researchers compared supportive conversations with a chatbot and conversations with a human stranger. The chatbot delivered more measured empathy than the human participants.
But only the human conversations reduced loneliness. The chatbot moved the measure not at all.
Read the original studyEmpathy as language may not be the active ingredient. Sometimes what matters is not simply receiving the right words. It is knowing that another person chose to offer them.
Practice before the next issue
Do the first five minutes yourself.
Choose one task you now give to AI that you once did yourself. Set a timer for five minutes. Begin without assistance. Notice which part of you the old version of the task exercised.
Play at The Quiet Cost PracticeYour turn
What do you want to keep?
Choose one task you now give to AI that you once did yourself. What part of you did the old version of that task exercise? How could you preserve that part without giving up the help?
Write for five minutes before asking any machine to help.
Sources and further reading
- Liu, Christian, Dumbalska, Bakker, and Dubey, “AI Assistance Reduces Persistence and Hurts Independent Performance.” Preprint, 2026.
- Cash, Kelly, Macnamara, and Risko, “Is AI making us stupid?” Trends in Cognitive Sciences, 2026.
- Li, Folk, Singh, Ungar, and Dunn, “Is a random human peer better than a highly supportive chatbot in reducing loneliness over time?” Journal of Experimental Social Psychology, 2026.