Why people build AGI despite the danger
Many of the people building advanced AI say openly that it could kill everyone. Why do they build it anyway?
This page is about why builders act as they do. Whether careful people should build is a separate question, argued in Should safety-focused labs keep building.
The short answer
No single reason explains it. Builders usually give several of the reasons below at once, and the ones they state are not always the ones doing the work. The first four are arguments that can be checked. The last three are motives, and people rarely say them out loud.
Reasons builders give
”It will happen anyway, so better me”
If someone will build it regardless, staying out changes nothing except who is in the room. The builder then prefers to be the one shaping it. This is the most common justification, and the hinge of the inside strategy in Should safety-focused labs keep building.
Elon Musk put it most plainly in 2026: since “there’s nothing I can do to stop it”, his choice was to be “a participant or a spectator” (source).
The weak point is that the argument is available to every builder at once, and their combined participation is what makes the race unstoppable. It also assumes that pausing is impossible, which is its own question: Can AI progress be paused. The people who make this argument have rarely tested it; Musk has not seriously tried to stop AI makes that case for Musk.
”The other side will win the race”
The national version of the previous reason: if the US slows down, China does not, and the most capable AI ends up in the hands of a government with worse values. See Can the US and China agree to limit AI.
”The benefits are enormous”
Cures for disease, an end to scarcity, faster science. If the risk can be managed, delay has a cost in lives. Machines of Loving Grace is the fullest statement of this case from a lab leader. See also Scarcity after automation.
”The danger is manageable”
Some builders think alignment is hard but tractable, or that AI will help solve it. If so, careful building is reasonable. Much depends on How hard is alignment and Can AI automate alignment research.
Motives people rarely state
Money and power
Whoever controls the most capable AI may control a large share of future wealth and decisions. See AI ownership and Who should control aligned AI. This motive is compatible with sincerely holding all of the reasons above.
Curiosity and the “technically sweet” problem
Building the most important technology in history is the most interesting work available. J. Robert Oppenheimer described the pull at his 1954 security hearing: when you see something “technically sweet”, you go ahead and do it, and argue about what to do only after the technical success.
Ego and legacy
Eddie’s view is that ego plays a larger part than builders admit: for someone used to shaping events, it is hard to accept being a bystander to the biggest change in history, even if the change may be bad. Musk’s own phrasing points this way: being a participant “is more interesting than being a spectator”, and “maybe I can shape it”. Musk’s AI position does not hold together develops this.
Not minding a successor
Some people do not think a future run by AI rather than humans would be a loss. Musk has long said that Google co-founder Larry Page called him a “speciesist” for wanting humanity to stay in control, and that he accepted the label. In the same 2026 interview he said “there’ve been a few times where I’ve thought maybe Larry Page was right.” If AI taking over is acceptable, much of the danger stops looking like a reason not to build.
How the reasons combine
The reasons reinforce each other. “It will happen anyway” removes the moral cost of joining. The benefits and the race give a positive reason. A belief that the danger is manageable, or that a successor would be acceptable, removes what is left of the objection. Ego, money, and curiosity make all of these easier to believe.
A test for any builder: would they support a binding, verified pause if every other major developer agreed to one? A builder whose main reason is “it will happen anyway” should say yes. One who would not is relying on another reason.
Related: Plans for the AI transition, Death with dignity.