AI Takes Our Willingness to Try Again

Originally written in Traditional Chinese: AI 最先拿走的，不是你的答案

My take: the first thing AI takes from us is not our answers. It is our willingness to try one more time.

On October 9, UC Berkeley News reported on a study by researchers from Berkeley, Carnegie Mellon, MIT, Oxford, and UCLA. Across three randomized experiments with 1,222 participants, people worked on short tasks. Some had an AI assistant in a sidebar and could get an answer just by asking. After roughly ten minutes, the AI was taken away.

In the first experiment, participants solved 15 fraction problems. After problem 12, the AI disappeared without warning. On the remaining problems, people who had worked alone solved 73% correctly. People who had used AI solved 57%. Their skip rate was 20%, compared with 11%.

The finding that stopped me came from the second experiment. Among participants in the AI condition, 61% said they mostly took direct answers, and 27% said they mostly asked for hints or clarifications. Once the AI was gone, the direct-answer group solved 65%. The hint group solved 76%. Same tool. Perhaps the difference is in what we ask for.

In an interview earlier this year, I said that anyone trained by an exam system will go straight for the answer when they use AI. We were trained to find the single right answer in the shortest possible time. AI just makes that shortcut shorter.

The researchers suggest that AI conditions people to expect immediate answers, removing the chance to work through difficulty on their own.

It's not our answers AI takes first. It's our willingness to try one more time.

I know what being stuck feels like. I started competing in public speaking in 2008. From 2008 to 2014, I did not win a single national championship. I am not someone who always wins. I am someone who kept getting through setbacks, again and again. What matters is not what came after 2014. It is everything I put in from 2008 to 2014.

To be clear, this study does not show that AI makes people less intelligent. Participants were adults recruited online, the tasks were short, and the answer-versus-hint groups were self-reported, so the researchers caution that this comparison is not necessarily causal. Nor is this an argument against using AI. The authors argue AI should be designed to build long-term capability, not only to finish the task in front of us. The minutes we spend stuck are not wasted. They are what learning looks like.

Three things you can try today:

  1. Hold on for three minutes. When you get stuck, set a timer before opening AI, and write one sentence: where exactly am I stuck?
  2. Ask for a hint, not an answer. Start with: "Give me one hint. I will find the answer myself."
  3. Remove the AI and try one more. After AI helps you with a problem, close it and solve a similar one alone. If you can, that one is yours.

For parents: when your child says "this is too hard," ask one question first: "Which step are you stuck on?"

For educators: on the next assignment, ask students to do the last question without AI, and see how many choose to skip.

For professionals: after AI drafts your next email, close it and write one paragraph yourself. The part you cannot write is the part to practice.

In the experiment, "skip" was just an option on a screen. In real life, it often sounds like: "Never mind, I'll just ask AI."

Get the Human-First practice card

Four steps, ten minutes, with copy-ready AI prompts. The card and my weekly letter are currently written in Traditional Chinese.

👉 Subscribe free to 碩學苑 (Traditional Chinese)

Follow in English

I also post in English on LinkedIn.

👉 Follow Haishuo Lee on LinkedIn

References