The Existential Risk Basket
Judge a technology by whether it shrinks the whole risk basket, not one scary scenario
- Difficulty
- Moderate
- Time to result
- ~weeks to results
- Steps
- 6
- Confidence
- 82%
Hoffman's answer to AI doomerism is a reframe of the question. Critics ask whether you can guarantee AI will not produce Terminators; the honest answer is no, and that answer ends the conversation before any real analysis happens. Instead, treat existential risk as a basket containing nuclear war, asteroids, pandemics, climate change, and everything else. Then ask the only question that can actually be answered: does building the technology make the whole basket better or worse? AI may add a new negative possibility to the basket, but Hoffman argues it is also the only route he can see to solving pandemics and is already advancing climate technologies. Net, the basket improves. The mechanism turns an unfalsifiable guarantee-demand into a comparison of two states of the world, and it forces the cost of delay onto the ledger rather than treating inaction as free.
Origin
Hoffman developed this while fielding what he calls every single pessimist argument against AI as an investor and co-founder, and formalized the reasoning in his books Impromptu and Super Agency.
Core principles
- 01Risk is a portfolio, never a single line item
- 02"Can you guarantee no harm?" is an unanswerable question, so it is the wrong question
- 03The correct comparison is the whole basket with the technology versus the whole basket without it
- 04Refusing to build also carries a cost, and that cost is usually invisible
- 05Delay is a decision with its own body count
How to run it
- 1
Surface the guarantee-demand
Identify the exact question the critic is asking. Usually it takes the form of "can you guarantee X won't happen?" Note that no one can guarantee it.
Pro tip Concede the point immediately. Hoffman says "no, I can't guarantee" — conceding is what lets you move to the real question.
Watch out Do not try to win the guarantee argument. You cannot, and trying makes you look evasive.
- 2
Enumerate the full basket
Write out every risk in the same category, not just the one being debated. For existential risk that means nuclear war, asteroids, pandemics, climate change, and more.
Pro tip Force the list to at least five items. One-item baskets are how the framing trap works.
Watch out A basket you assemble to win an argument is not a basket. Include risks that cut against you.
- 3
Score the basket without the technology
Establish the baseline. What does the risk profile look like if this technology is never built or is indefinitely delayed?
Watch out The status quo is not a zero-risk state. Treating it as one is the most common error.
- 4
Score the basket with the technology
Add the new risks the technology introduces, then subtract the risks it mitigates or solves. Hoffman's example: AI adds the Terminator possibility but is the only path he can see to solving pandemics.
Pro tip Ask what the technology uniquely unlocks — not what it merely accelerates. Unique unlocks dominate the comparison.
Watch out Do not let the new risk count double just because it is novel and vivid.
- 5
Price the delay
State the cost of not building, in concrete human terms. Hoffman's line is that doomers delay a future in which every person has a medical assistant and tutor better than the average doctor, and that delay has a huge cost in human suffering.
Pro tip Quantify per unit of time. "A year of delay costs X" is far more decision-useful than "delay is bad."
Watch out This step is where motivated reasoning creeps in. Be as honest here as in step 4.
- 6
Decide on the comparison, then plan the transition
Make the build/don't-build call on the basket comparison. Then separately plan for the transition difficulties, which Hoffman treats as real and worth focusing on rather than denying.
Pro tip Separating "is the destination good?" from "is the path hard?" prevents transition pain from vetoing a net-positive outcome.
Watch out Concluding the basket improves is not permission to ignore transition harm. Hoffman calls navigating the transition one of the things he most focuses on.
In the wild
A critic asks Hoffman to guarantee that AI will not deliberately or accidentally produce killer robots. He concedes he cannot guarantee it. The critic treats that as case closed under a precautionary principle. Hoffman's move is to point out this only follows if killer robots were the only item in the basket. Once you add nuclear war, asteroids, pandemics, and climate change to the list, the question becomes whether AI makes the aggregate better or worse. He argues it clearly improves it, because AI is the only route he can see to solving pandemics and is already helping advance climate technologies.
→ The unfalsifiable guarantee-demand becomes a tractable comparison, and the precautionary conclusion no longer follows automatically.
Hoffman notes that we already have line of sight to putting a medical assistant better than the average doctor, plus a tutor for every subject and every age, on a smartphone — available to all eight-plus billion people with a phone. Applying the basket, the doomers are not choosing zero risk; they are choosing to delay that deployment. He puts the cost of that delay on the ledger explicitly: the delay itself has a huge cost in human suffering.
→ Inaction stops being scored as the safe, costless default.
Common mistakes
Treating the status quo as zero risk
Not building is scored as safe by default, so the basket without the technology is never actually tallied. This makes any new risk look like pure addition.
Answering the guarantee-demand on its own terms
Arguing that a catastrophic outcome is unlikely concedes the framing that one scenario decides the question. Concede the guarantee, then change the question.
Building a basket that only contains convenient risks
If you assemble the basket after deciding the answer, the comparison is theatre. The method only works if the enumeration is honest.
Is it for you?
Best for
Anyone evaluating a general-purpose technology, policy, or investment where critics anchor on a single catastrophic scenario.
Not ideal for
Narrow, well-bounded decisions with one dominant, quantifiable failure mode where portfolio thinking adds nothing.
From the transcript
“Existential risk is a basket. It's not just are there killer robots or not. It's also nuclear war, asteroids, pandemics, climate change, a bunch of…”
“We may add the Terminator robots as a negative possibility, but is the basket of risk get better or worse? And I think it just…”
“All the doomer and gloomers essentially are delaying that future and that delay has a huge cost in human suffering.”
From the episode
Reid Hoffman: Unlocking Limitless Human Possibilities With AI
Reid Hoffman