All things AI

[Recusant]

Could/would an AI do a better job of running humanity than humanity has?

Sure--it's ready to give us bold innovations. Like using glue on your pizza.

[Tank]

LOL. Yes I saw that. But the question remains. Would a AI be better than the fuck ups we have managed ever since we learned to bang rocks together?

[billy rubin]

no.

[Tank]

When you have a moment I'd like to hear your thoughts behind that 'no'.

[billy rubin]

the answer is simple, but hard. first, an AI would act to to maximize the goals put forward in its programming.

which is based on human thinking, the original programmers.

until the day when thehumans say, AI, youre so smart, make some decisions for yourself, to the general benefit.

note, the entire purpose of A in this contextI is to derive a machine tool that will make decisions better than a human can, in a human situation. if you have a different view of this question, let me know, because i dont know of any.

when the AI is given the choice to make a decision based on its original programming, it will do so based only on those considerations-- with all the errors that ensue. when it is given latitude, it will use it to the fullest.

what directions do we issue to our AI?

--make a better world?

--make a world wher ehappiness is th enumber one goal?

--make a world where the arabs cannot kill the jews?

--make a world where the russians cannot launch an atomic bomb?

who does the programming?

who figures out the semantics?

who can correct the AI when it goes wrong?

try this:

list seven laws for AI programming to control the world for our benefit, in order.

think carefully. what seven directives would you provide to an AI to guide its decisions.

think about asimovs three laws to get an idea of the stature of the seven laws you would sugest.

[Tank]

A lot to consider. I shall have a think.

[Tank]

"the answer is simple, but hard. first, an AI would act to to maximize the goals put forward in its programming."

I think we see AI in a different way. An AI would, almost by definition, not be rule driven but rule learning and rule defining as it grew. What I would do is give it no initial rules beyond possibly 'learn all that there is to learn'. I would then let it go on all we have discovered and learned and speculated. At some point I would ask it what it knows and what it would do with that knowledge if allowed to. This first exercise would be to understand what an unfettered AI would do if left to it's own devices.

I'm not avoiding the seven 'rules' or objectives as such. I'm going about the learning process in a different way to see how, not if, it could exceed our expectations.

[Asmodean]

Could/would an AI do a better job of running humanity than humanity has?

Now this is precisely the kind of question to bait a unsuspecting The Asmo out of hiding.

I think the broad and simple answer is "no," seeing how a lot of governing already is a "flowchart business."

If A, do B. If C, do D. When things go OK, it's not a very creative job, unless you live in a society where the government's job is to find ways to generate value, rather than "merely" getting paid rent by those who do. Beyond that, it's reacting to culture, the huwhims of its constituents, outside influences... so forth.

AI may be efficient at it, but not necessarily "better" for it.

[Tank]

Could/would an AI do a better job of running humanity than humanity has?

Now this is precisely the kind of question to bait a unsuspecting The Asmo out of hiding.

I think the broad and simple answer is "no," seeing how a lot of governing already is a "flowchart business."

If A, do B. If C, do D. When things go OK, it's not a very creative job, unless you live in a society where the government's job is to find ways to generate value, rather than "merely" getting paid rent by those who do. Beyond that, it's reacting to culture, the huwhims of its constituents, outside influences... so forth.

AI may be efficient at it, but not necessarily "better" for it.

But that's missing the key point about AI in that it is supposed to be 'intelligent' not simply a rule driven algorithm. Maybe it would need to go to school and study rather than being programed as such?

[Tank]

Would an AI be republican or democrat?

[Dark Lightning]

Depends on whether its core is a lump of coal or a diamond. 

[Asmodean]

Yes, the AI would have to be capable of actual, as opposed to simulated, understanding. That does not counter my point, however, since at its core, it's that efficient politics is not a particularly creative job. It's reactive and all the palms you must grease to remain in power... Well, must be greased. Whether it be The Public(tm), the generals or that basement-dwelling IT tech with his collection of Star Wars t-shirts.

Would an AI be republican or democrat?

It would hopefully try to address what doesn't work and keep its digital fingers out of that which does. So... Perhaps leaning classical progressive at the start, then leaning conservative "on a regular Tuesday."

[Tank]

"efficient politics is not a particularly creative job. It's reactive and all the palms you must grease to remain in power"

What could you offer an AI that it could possibly value? How would you bribe/influence it?

[Tank]

...

Would an AI be republican or democrat?

It would hopefully try to address what doesn't work and keep its digital fingers out of that which does. So... Perhaps leaning classical progressive at the start, then leaning conservative "on a regular Tuesday."

So essentially pragmatic reactive?

[Tank]

Billy. Here are two potential guidance thoughts for an AI.

1. Do no harm.
2. Do unto others as you would have done unto you.

I appreciate 1. would depend on how 'harm' is defined and 2. would be problematic based on who 'you' is/was. Any thoughts?