The Three Inevitables Model
Plan for unstoppable AI, superhuman intelligence, and inevitable failures
- Difficulty
- Moderate
- Time to result
- ~ongoing to results
- Steps
- 4
- Confidence
- 93%
The Three Inevitables Model sets three planning assumptions for an AI-shaped future. First, AI has already happened and will not simply be switched off, because companies, nations, and defenders are caught in an arms race: each fears that rivals will continue. Second, continued development means AI will become smarter than humans in more domains, making permanent human control an unsafe premise. Third, bad outcomes will occur somewhere as powerful systems spread, even if catastrophic outcomes remain uncertain. The model does not predict one exact disaster. Instead, it removes three fragile assumptions from strategy: that development will stop, that humans will always remain cognitively superior, and that nothing important will break. Teams can then focus on resilience, accountability, coordination, and adaptation rather than waiting for certainty or relying on a universal pause that competitors have little incentive to honor.
Origin
Gawdat introduced the three inevitables in Scary Smart to explain why AI cannot be stopped, will surpass human intelligence, and will produce harmful outcomes.
Core principles
- 01AI development will continue because competitors cannot trust one another to pause
- 02Machine intelligence will eventually surpass human intelligence
- 03Some harmful outcomes are inevitable in a system this widespread and powerful
- 04Preparation is more useful than planning around a complete shutdown
How to run it
- 1
Adopt continuation as the baseline
Build plans on the assumption that AI research and deployment continue. Identify dependencies on a universal pause or an off switch.
Pro tip Ask what competitors would do if your organization stopped alone.
Watch out A voluntary pause by one actor does not end an arms race.
- 2
Stress-test human superiority
Find decisions that assume people will remain more capable than machines. Redesign oversight for a world where capability can exceed the overseer's understanding.
Watch out Control methods based only on being smarter than the system may not endure.
- 3
Map plausible failures
List immediate and systemic harms that could arise from widespread use. Include ordinary failures and misuse, not only science-fiction extinction scenarios.
Pro tip Prioritize harms that are already observable or can occur soon.
Watch out Focusing only on extreme scenarios can distract from present risks.
- 4
Build adaptive safeguards
Invest in resilience, accountability, and response plans that remain useful across multiple futures. Update them as the capability frontier changes.
Pro tip Prefer safeguards that reduce several failure modes at once.
In the wild
Gawdat explains that Google cannot confidently stop while Meta, Amazon, nations, or small independent developers may continue. Each actor's duty to defend its position reinforces the race even when leaders recognize shared risk.
→ Unilateral restraint fails to stop development, so strategy must account for continued competition.
A professional-services firm stops assuming that AI capability will plateau. It retrains staff, creates review controls, and builds an incident process while competitors continue adopting stronger systems.
→ The firm becomes more resilient without needing to predict the exact arrival date of advanced AI.
Common mistakes
Treating an off switch as strategy
Distributed development across companies, countries, and individuals makes a universal shutdown assumption fragile.
Fixating on one forecast date
The framework is valuable as a set of planning assumptions even when precise timelines remain disputed.
Equating inevitability with helplessness
Accepting continued development is a reason to build safeguards and adaptation plans, not to abandon action.
Is it for you?
Best for
Leaders making durable strategy under rapid and uncertain AI development.
Not ideal for
Short-lived operational choices where AI capability has no material effect.
From the transcript
“There are three inevitables.”
“there is no stopping it”
“we've created a prisoners dilemma”
From the episode
Mo Gawdat: Ex-Google Officer Warns About the Dangers of AI, Urges All to Prepare Now!
Mo Gawdat