The Keys We Keep — A 7-Day Sprint on AI Safety
MIT's Max Tegmark and Kevin Esvelt ask which powers humanity must never hand over to AI. This seven-day sprint takes that question personally — one idea per day, each with a reflection, an action, and links deeper into Voice Up's AI safety resources.
Day 1 — The Question Nobody Asks
Most AI conversations ask what the technology can do. Tegmark and Esvelt ask what it must never do — which decisions must remain human no matter how capable machines become. It is a different question, and it changes everything about how you think.
Reflect
Which decisions in your own life would you never delegate to a machine — and why those?
Do
Write your personal 'never delegate' list: three decisions that must stay human, with one sentence each on why.
Day 2 — The Insider's Warning
The people urging restraint aren't outsiders — they're the builders. When the engineers say slow down, it carries a different weight than when critics do. The symposium's credibility comes from the speakers' proximity to the work.
Reflect
When have you warned someone about something you were part of — and what did it cost you to speak?
Do
Read one AI safety news story today. Note who is speaking, what they're close to, and what they're risking by speaking.
Day 3 — Biology's Lesson: Containment
Esvelt brings a biologist's mindset: in biology, containment is a discipline — layers of safeguards, each assuming the others might fail. Applied to AI, it means never trusting a single guardrail with civilizational stakes.
Reflect
What does 'containment' mean for a technology that can copy itself — and who is responsible for building the layers?
Do
Explain the idea of containment to someone else in plain language — no jargon. Notice what gets clearer when you have to simplify it.
Day 4 — Precision Over Panic
Vague fear paralyzes; precise risks mobilize. The symposium models precision — naming specific powers, specific failure modes, specific lines. Anxiety becomes agency the moment it gets specific.
Reflect
What is one specific AI risk you can name precisely — not a vibe, a mechanism?
Do
Write a one-paragraph 'risk statement' in plain words: what could go wrong, how, and what would prevent it.
Day 5 — Who Decides?
The deepest question isn't technical — it's democratic. Who gets a say in AI's future? The symposium argues the answer can't be 'whoever builds it fastest.' Governance is a design choice, and it's ours to make.
Reflect
Who should have a say in AI's future — and do you currently have one?
Do
Find one public forum, comment period, or community conversation about AI policy. Read what's being said — or add your voice.
Go deeper
Day 6 — The Long View
Tegmark thinks in centuries. Most tech debates think in quarters. The symposium's time horizon is its superpower — it asks what today's choices look like from the year 2126.
Reflect
What will matter in 100 years about the AI choices made this decade?
Do
Write a short letter to someone living 50 years from now about AI — what you hope we got right, and what you fear we got wrong.
Day 7 — Your Hands on the Keys
The symposium ends where your life begins: with the keys you personally hold. You may not run a lab, but you decide how you use AI, what you delegate, and what stays human in your own work.
Reflect
What is one way you keep human judgment central in your own use of AI?
Do
Commit to one concrete practice — then look back over the week and note which day changed your thinking most.