Take AISafety.info’s 3 minute survey to help inform our strategy and priorities

Take the survey
Predictions about future AI

Timelines
Compute and scaling
Nature of AI
Takeoff
Takeover
Relative capabilities
Good outcomes
Catastrophic outcomes

How might AGI kill people?

The general problem with predicting the exact ways an AGI

could outmaneuver humans is that, while we can predict that an extremely smart and capable agent would be able to achieve its goals (since that's what smart and capable means), we can't predict exactly how it will achieve those goals. This is because it is likely to come up with plans and factors that we didn’t even consider. This Vingean uncertainty is worth keeping in mind when discussing ways an AGI could win in a confrontation with humans: you might think you have a counter-move for everything, and still reliably end up losing.

With this in mind, an internet-connected AGI that wants to harm humans could:

  • Hack our weapons systems and turn them against us

  • Create bioweapons and either deploy them autonomously or distribute them to rival nations or terrorist organizations

  • Socially manipulate humans in ways that incite conflict that lead to war

  • Hack into anything connected to the internet and cause car/plane/ship crashes

This list is of course not exhaustive. For many of the cases listed above, the AGI would need to be connected to the internet, but “boxing an AGI” might be harder than it seems.

Further reading:

Keep Reading

Continue with the next entry in "Predictions about future AI"
How likely is it that an AI would pretend to be a human to further its goals?
Next
Or jump to a related question


AISafety.info

We’re a global team of specialists and volunteers from various backgrounds who want to ensure that the effects of future AI are beneficial rather than catastrophic.