Recently, I’ve been thinking about the implications of a world where general-purpose robots are developed before AGI. (In such a world, I imagine, trillions of robots would permeate society, replace factory workers, work in homes, etc.)
One implication could be increased AI takeover risk. Intuitively, widespread robots might make takeover substantially easier for a variety of reasons, such as by giving a ASI readily available physical actors and making it easier to secure datacenters / other physical infrastructure crucial to AI. On the other hand, a superintelligence may have an unfathomable number of other ways to gain power or cause extinction. If takeover is thus already likely through other routes, perhaps pre-existing robots add only a negligible amount to total risk.
TL;DR: Does anyone know of good discussions on whether widespread general-purpose robots before ASI would meaningfully increase takeover risk?
I’ve found scarcely anything on this; discussions from Paul Christiano and Nick Bostrom, for instance, argue that AI could take over without general-purpose robots, but don’t address how much additional risk pre-existing general-purpose robots would create.
If anyone knows of relevant readings or discussions, would really appreciate it. Also, I understand the effect is difficult to quantify, so am just trying to get a sense of how people think about its likely magnitude.
Progress in robotics has been drastically slower than progress in AI. Everyone has access to Claude, ChatGPT, Gemini, and so on, and they use them for practical things, not cooing over how well the dog walks on its hind legs. No-one has access to a robot housekeeper, because nothing close to that exists.
Much depends on the economics, and on the range of behaviors these robots can perform autonomously. I suspect there's a long road to get them cheap enough to really replace human workers in most of the supply and maintenance chain, let alone "trillions of robots" (100 per human?). I could be wrong here - once they CAN replace all humans in the mining, refining, manufacturing, etc. of themselves, it could very rapidly take over. Much like AI self-improvement could very well foom.
Recently, I’ve been thinking about the implications of a world where general-purpose robots are developed before AGI. (In such a world, I imagine, trillions of robots would permeate society, replace factory workers, work in homes, etc.)
One implication could be increased AI takeover risk. Intuitively, widespread robots might make takeover substantially easier for a variety of reasons, such as by giving a ASI readily available physical actors and making it easier to secure datacenters / other physical infrastructure crucial to AI. On the other hand, a superintelligence may have an unfathomable number of other ways to gain power or cause extinction. If takeover is thus already likely through other routes, perhaps pre-existing robots add only a negligible amount to total risk.
TL;DR: Does anyone know of good discussions on whether widespread general-purpose robots before ASI would meaningfully increase takeover risk?
I’ve found scarcely anything on this; discussions from Paul Christiano and Nick Bostrom, for instance, argue that AI could take over without general-purpose robots, but don’t address how much additional risk pre-existing general-purpose robots would create.
If anyone knows of relevant readings or discussions, would really appreciate it. Also, I understand the effect is difficult to quantify, so am just trying to get a sense of how people think about its likely magnitude.