01
THE CHOICE
Would you press again?
A possible breakthrough. A possible irreversible loss. Who should decide what risk is acceptable?
Go deeper +
Would your answer change if the benefits went to you, but the risks fell on someone else?
100 BUTTONS. ONE FUTURE.
If one button could
end humanity,
would you
press any?
This board has ten.
AI’s future affects all of us. We should all have a voice in it. Play to understand the stakes.
MAKE YOUR CHOICE
Not pressing is a decision too.
You chose not to take that risk. AI development continues. Who should have a say in the risks taken on everyone’s behalf?
Bring this question to the conversation →BEFORE YOU PRESS
- 01Pick one button.
- 02Read the idea it reveals.
- 03Find a mine? The round ends.
Make your choice.
10 hidden mines. The first one ends this round.
OPENED / 100000
HIDDEN MINES010
Your move. Every unopened button is a choice.
● Idea✚ Possibility✳ Mine
All 100 questions answered. Talk it through →
A Minesweeper-inspired thought experiment.
90 learning cards. 10 imagined irreversible outcomes.
THE ESTIMATE BEHIND THE BOARD
A warning in their own words.
Evan Hubinger gives a personal estimate of over 10% within a decade. Jacob Coxon warns that the people building advanced AI fear it could kill everyone. Leaders at OpenAI, Anthropic and Google DeepMind have also signed a statement naming extinction risk as a global priority, without agreeing on one percentage.
Read the signed statement ↗
Evan Hubinger on X ↗ · Jacob Coxon on X ↗
ONE MAP · DIFFERENT PREDICTIONS
AI → AGI → ASI
- 2026 WE ARE HERE
AI
Artificial intelligenceSystems that recognise patterns, generate content and act with tools.
AGI already? Jensen Huang“I think we’ve achieved AGI.”
23 Mar 2026 · NVIDIAMeaning + source
He was discussing an AI-created billion-dollar business, potentially a short-lived viral app, not human-level ability at every job. Original interview · 1:56:31 ↗
- ~2030 FORECAST · ±1 YEAR
AGI
Artificial general intelligenceBroadly human-level thinking and learning across many different tasks.
Demis HassabisAGI around 2030, give or take a year.
27 May 2026 · Google DeepMindMeaning + source
His forecast depends on solving gaps in memory, consistency and learning. AGI’s definition is disputed. Original interview coverage ↗
- ~2034–35 FORECASTS / VISION
ASI
Artificial superintelligenceFar beyond the best human thinkers across a wide range of tasks.
Junichi Miyakawa · ~2034ASI within ten years of his 2024 speech.
Masayoshi Son · ~2035A major ASI economy in his 2025 vision.
Meaning + sources
SoftBank’s Miyakawa gave a possible arrival window; Son described a business vision, not the first ASI’s arrival. Miyakawa · Oct 2024 ↗ · Son · 2025 report ↗
AI is here; AGI’s arrival is disputed. These are selected views, not agreed dates. Arrows show a possible path, not an inevitable sequence or a time scale. AGI and ASI are also AI.
What could drive the transition?
Better algorithms, training and computing power could expand capability. AI helping improve the next AI could accelerate this: recursive self-improvement. Greater ability does not automatically mean safe goals or human control.
01 / TALK IT THROUGH
The next move
is a conversation.
Try one card with your team.
No technical background needed.
02
THE UPSIDE
What is worth building?
Healthier lives, better cancer treatment, less routine work. Which benefit would matter most to you?
Go deeper +
What evidence would show it works, and who could actually access it?
03
THE POWER
Who holds the controls?
The lab builds the model. Governments set rules. Organisations decide where to use it. Where should people have a say?
Go deeper +
Think of an AI system at work. Who can pause it, question it or reverse a decision?
04
THE EVIDENCE
What would change your mind?
Name one finding that would make you more worried, and one that would reassure you.
Go deeper +
What would count as evidence rather than a promise or a viral post?
Running a team session? A simple 15-minute format +
- 3 minutes: play one round. Ask how the first mine changes the feeling.
- 5 minutes: discuss the cards you uncovered. What did you learn about the stakes?
- 7 minutes: choose a conversation card. Separate facts, forecasts and imagined stories.
People can choose not to play. No leaderboard, no timer, no pressure to agree.
02 / CONNECT THE DOTS
A chat window is
only the beginning.
Two short visual guides to what AI can do, and what makes it possible.
CAPABILITY + ACCESS + CONTROLHow does AI move
How does AI move
beyond a chat?
A chatbot is already AI. What changes the stakes is what it can do, what it can reach, and who stays in control.
It writes an answer.
A language model generates text. Think of advice on a screen: the answer alone does not operate a machine.
Select a stage. More access changes the consequences.
THE PHYSICAL WORLD BEHIND THE CLOUDWhat is AI
What is AI
running on?
“The cloud” is real equipment, in real buildings, using electricity and cooling. Follow the chain.
A chip does the calculations.
Hardware means physical computer parts. AI chips, including GPUs, perform many calculations in parallel. Think of a large team doing maths together.
Select a layer. Software needs physical infrastructure.
Explore the linked definitions in the learning library. Diagrams illustrate relationships, not inevitable outcomes.
KEEP ASKING