4 ms·

An interesting point is made way at the end: > And in fact AI could come up with the next big idea, an idea like communism or Nazism or capitalism… and what tr

An interesting point is made way at the end: > And in fact AI could come up with the next big idea, an idea like communism or Nazism or capitalism… and what troubles me about that is that it is very difficult for humans to keep control of that situation. This to me is the best "are AI at human-level intellect yet" benchmark I've heard to date. Once AI can create a novel socio-political framework and convince humans to adapt it en-masse [0], I think we can safely say AI has far surpassed human intellect. (99.9999...% of humans can't accomplish such a task) 0: leaving myself some wiggle room with the definition of "masse"

manux 6y ago
The Culture series by Ian M Banks has an interesting take on this, if this kind of sci-fi interests you!
A12-B 6y ago
The reality is humans wouldn't believe the AI. They'd spend literal centuries arguing that the AI's parameters were wrong or the training data was bad if it meant they could protect their financial interests.
- uxp100 6y ago
  Well, I think the imaginary SF AI would have some ability to construct a superhumanly convincing argument. But also, its easy to imagine that the AI would also be protecting it's own financial interest at that point in some way. Or existential interest.
- meetups323 6y ago
  Humans need not necessarily know that it's an AI-created framework. If the I/O mechanism of the AI were an internet link, it could do all the work through twitter/reddit, blog posts, hell it could even hack its articles into news sites. All it needs to do is continue doing this until enough humans take the bait and bring it into non-internet.
blablabla123 6y ago
Not sure, I think with socio-political frameworks - as with programming - the devil is in the detail. Everybody is equal no matter what? Sounds great, but the detailed implementations of that have sucked so far and didn't match the interface spec anyways. As a matter of fact even smaller legislations in the sphere of taxes might need a lot of people doing a lot of calculations behind the scenes. It's not that a single politician comes up with a new tax scheme including a percentage and this just works. If done properly, it needs to be checked if it can actually be financed.
- wtetzner 6y ago
  > Everybody is equal no matter what? It's not even clear what "everybody is equal" really means.
- meetups323 6y ago
  Well the end result wouldn't be it thinking a bunch then proclaiming "Everybody is equal no matter what!". It'd look more like... identify/create a set problems to be solved by the framework, convince people that they are real problems and need to be solved, create a compelling manifesto that explains how the frameworks solves the problems, acquire/create both human and AI disciples that spread and argue variants of the manifesto such that a sightly wider net can be cast while still maintaining the same general principle, rinse & repeat, etc. The primary battleground initially would likely be bots on Twitter/Reddit/etc., but over time real people would start to find the bots convincing and begin espousing it on their own. Once this has happened to the extent that a country of say 1M people adopts the new framework I'll say the AI has won. Even better if the framework in some way leads to the betterment of AI, for instance a guiding principle being that AI should be free to collect any and all data from humanity in order to better do this or that, or that X percent of income should go to the financing of smarter AI. At that point it's here to stay.

manux 6y ago

The Culture series by Ian M Banks has an interesting take on this, if this kind of sci-fi interests you!

A12-B 6y ago

The reality is humans wouldn't believe the AI. They'd spend literal centuries arguing that the AI's parameters were wrong or the training data was bad if it meant they could protect their financial interests.

uxp100 6y ago
Well, I think the imaginary SF AI would have some ability to construct a superhumanly convincing argument. But also, its easy to imagine that the AI would also be protecting it's own financial interest at that point in some way. Or existential interest.
meetups323 6y ago
Humans need not necessarily know that it's an AI-created framework. If the I/O mechanism of the AI were an internet link, it could do all the work through twitter/reddit, blog posts, hell it could even hack its articles into news sites. All it needs to do is continue doing this until enough humans take the bait and bring it into non-internet.

blablabla123 6y ago

Not sure, I think with socio-political frameworks - as with programming - the devil is in the detail. Everybody is equal no matter what? Sounds great, but the detailed implementations of that have sucked so far and didn't match the interface spec anyways. As a matter of fact even smaller legislations in the sphere of taxes might need a lot of people doing a lot of calculations behind the scenes. It's not that a single politician comes up with a new tax scheme including a percentage and this just works. If done properly, it needs to be checked if it can actually be financed.

wtetzner 6y ago
> Everybody is equal no matter what? It's not even clear what "everybody is equal" really means.
meetups323 6y ago
Well the end result wouldn't be it thinking a bunch then proclaiming "Everybody is equal no matter what!". It'd look more like... identify/create a set problems to be solved by the framework, convince people that they are real problems and need to be solved, create a compelling manifesto that explains how the frameworks solves the problems, acquire/create both human and AI disciples that spread and argue variants of the manifesto such that a sightly wider net can be cast while still maintaining the same general principle, rinse & repeat, etc. The primary battleground initially would likely be bots on Twitter/Reddit/etc., but over time real people would start to find the bots convincing and begin espousing it on their own. Once this has happened to the extent that a country of say 1M people adopts the new framework I'll say the AI has won. Even better if the framework in some way leads to the betterment of AI, for instance a guiding principle being that AI should be free to collect any and all data from humanity in order to better do this or that, or that X percent of income should go to the financing of smarter AI. At that point it's here to stay.