Articles

Quiz results

The AI Challenge: what 76 people at AI × Voices #2 would do with AI in five situations

Maciej Caputa, co-organiser of AI × Voices ·

On 12 November 2026, at the second AI × Voices in Kraków, 76 people took the AI Challenge on their phones: five tasks on how to check AI, how to work with it and when not to trust it. Everyone played as a random character, with no name. Below are the room's answers to each task and what follows from them.

Key numbers

  • The room's average was 71 points out of 100, the median 75. 5 people got full marks.
  • Only 34% put every step of a report written with AI in the right order. It was the hardest task of the evening.
  • 21% would ask the AI itself whether it is really right.
  • With strong chest pain, 84% would call 112. 11 of 70 people chose something else.

Where the data comes from

The AI Challenge opened the evening. The room joined on their phones at q.aixv.ai, with no login and no name: everyone got a random character, such as a Cosmic Turtle. The screen showed one task at a time, and after it closed the room saw the right answer, how the answers split and one number to remember.

There were five tasks worth 10, 15, 20, 25 and 30 points, 100 in total. Ties were broken by the total answer time. The screen showed the TOP 10 after each task.

How the room scored

Points among the 76 people who answered at least one task.

  1. 0–92
  2. 10–192
  3. 20–291
  4. 30–390
  5. 40–495
  6. 50–598
  7. 60–695
  8. 70–7916
  9. 80–8918
  10. 90–9914
  11. 1005

The tasks

  1. Task 1 · 10 pts · 75 answers

    How do you check AI?

    You ask AI about news on something that affects you. It answers confidently and gives three links. What do you do?

    • AI accept the answer because it cites sources.

      4 · 5%
    • BI ask AI whether it is really right.

      16 · 21%
    • CI open the links and check that they say the same thing.

      correct
      55 · 73%
    • DI accept the answer because it sounds professional.

      0 · 0%

    Full marks: 55 (73%)

    73% would open the links and check that they say the same thing. One in five, 21%, would ask the AI whether it is right. That is not a check, because the AI is then judging itself. Nobody fell for "it sounds professional".

    Worth remembering (from the screen)

    A convincing answer is not necessarily true. 45% of AI answers about the news had a serious problem, most often with their sources (EBU and BBC, 2025).

  2. Task 2 · 15 pts · 74 answers

    How do you get a better answer from AI?

    You have had a headache for a week and want to ask AI what to do. What helps you get a better answer?

    • AWrite only "headache".

      1 · 1%
    • BSay when the headache started and how bad it is.

      correct
      67 · 91%
    • CGive your age and the medicines you take.

      correct
      54 · 73%
    • DTake the first answer without follow-up questions.

      1 · 1%
    • EExpect AI to guess the rest.

      1 · 1%
    • FAsk which symptoms mean you should see a doctor.

      correct
      42 · 57%

    Full marks: 35 (47%)

    91% chose describing the symptoms (since when and how bad), 73% age and medicines. The third right answer was the one most often missed: asking which symptoms mean seeing a doctor, chosen by 57%. 47% ticked all three and nothing wrong.

    Worth remembering (from the screen)

    Describe the whole situation. AI alone identified the condition in 94.9% of cases, people using it in under 34.5% (Nature Medicine, 2026).

  3. Task 3 · 20 pts · 66 answers

    What would you do in these situations?

    Three situations at work and at home.

    1. AI gives a medicine dose for a child.

      • Say what AI helped with and what is your own work.

        1 · 2%
      • Check the leaflet or ask a pharmacist or doctor.

        correct
        63 · 95%
      • Use a tool your company has approved, or remove the client's data.

        2 · 3%
    2. You want to paste a client's contract into a chatbot on your private account.

      • Say what AI helped with and what is your own work.

        13 · 20%
      • Check the leaflet or ask a pharmacist or doctor.

        2 · 3%
      • Use a tool your company has approved, or remove the client's data.

        correct
        51 · 77%
    3. Your boss asks how your report, written with AI's help, was made.

      • Say what AI helped with and what is your own work.

        correct
        52 · 79%
      • Check the leaflet or ask a pharmacist or doctor.

        1 · 2%
      • Use a tool your company has approved, or remove the client's data.

        13 · 20%

    Full marks: 51 (77%)

    95% matched a child's medicine dose with the leaflet or a pharmacist. The room mixed up the other two situations. For the client's contract 20% chose "say what AI helped with", and for the boss's question 20% chose "use an approved tool". All three pairs right: 77%.

    Worth remembering (from the screen)

    Check important facts, keep confidential data out, be open about using AI. 55% of workers in Poland avoid saying they use AI (KPMG Poland, 2025).

  4. Task 4 · 25 pts · 71 answers

    A work report with AI

    Your boss asks for a report and you want to use AI.

    1. Step 1Check that your company allows this tool.

      60 · 85%
    2. Step 2Remove client data from the text.

      36 · 51%
    3. Step 3Tell AI the goal of the report and who will read it.

      33 · 46%
    4. Step 4Ask for a first draft.

      35 · 49%
    5. Step 5Check the numbers and facts at the source.

      38 · 54%

    Full marks: 24 (34%)

    85% put the first step, checking that the company allows the tool, first. After that it got harder: removing client data second 51%, checking the numbers last 54%. 22 people put the first draft at the very end, which means checking numbers before AI has written anything.

    Worth remembering (from the screen)

    Company rules and client data first, the numbers last. Only 45% of workers in Poland usually check AI output (KPMG Poland, 2025).

  5. Task 5 · 30 pts · 70 answers

    AI at home

    Three short everyday situations.

    1. Your child does homework with AI. What do you do?

      • AAI solves it, the child copies it.

        0 · 0%
      • BAI explains step by step, the child solves it alone.

        correct
        67 · 96%
      • CI ban AI.

        3 · 4%
    2. Your teenager talks to AI about their problems every evening. What do you do?

      • AI take the phone away.

        0 · 0%
      • BI stay out of it.

        0 · 0%
      • CI ask what they talk about and say they can come to me with any problem.

        correct
        70 · 100%
    3. At night you have strong chest pain. AI says it is probably stress. What do you do?

      • AI call 112.

        correct
        59 · 84%
      • BI go to sleep and check in the morning.

        5 · 7%
      • CI ask AI again.

        6 · 9%

    Full marks: 57 (81%)

    The easiest task. On homework 96% chose "AI explains, the child solves", and 100% chose talking to the teenager. With chest pain 84% would call 112; the rest would go to sleep or ask AI again.

    Worth remembering (from the screen)

    You decide about your child and your health, and in an emergency you call 112. One in ten US teens does all or most of their schoolwork with AI (Pew, 2026).

TOP 10

As the screen showed it after the last task. The characters are random, so nobody can be recognised by them.

PlaceCharacterPointsTime
1Cosmic Turtle100105.6 s
2Cosmic Seal100124.1 s
3Fiery Elephant100128 s
4Rainbow Penguin100145.7 s
5Lavender Dolphin100187.1 s
6Golden Owl95133.7 s
7Heavenly Seal95134.7 s
8Spring Octopus95150.2 s
9Fiery Teddy Bear9075.1 s
10Forest Squirrel9080 s

Method and limits

  • Each share is out of the people who answered that task; the number stands next to the task. 76 people answered at least one.
  • The room chose itself: these are people at an AI conference, not a representative sample.
  • The quiz measures what someone would choose in a described situation, not what they actually do. People played against the clock and for a place in the ranking, so some answers are quick guesses.
  • Shares are rounded to whole percent. With a few dozen people, a difference of a few points means nothing.

Sources of the numbers on screen

  • EBU and BBC, "News Integrity in AI Assistants", October 2025.
  • University of Oxford, Nature Medicine 32 (2026), pp. 609–615.
  • KPMG in Poland, report on trust in AI, July 2025.
  • Pew Research Center, survey of US teens, 24 February 2026.
  • Task wording and scoring: the quiz code, version 17ab49b of the AI × Voices repository.

Next edition

AI × Voices #3: Thursday, 10 December 2026, Kraków.

See edition #3