Print

Please login or register.
1 Hour 1 Day 1 Week 1 Month Forever
Login with username, password and session length

[Zanzetkuken The Great]

Let's take a moment and discuss what nations under potential government types that have not existed due to either insufficient technology or other reasons.  Please avoid discussion on the varieties of socialism that haven't been tried on a national scale and true communist states, as those all have kinda been discussed enough elsewhere that there isn't too much to add, and are a bit likelier to shift towards flame wars.

Let's start discussion off with what a nation under the rule of a high level artificial intelligence that has the specifications to maintain a population that has a high level of genetic variety and have the nation operate a high level of efficiency.  What would it look like inside the borders of such a nation?

[smjjames]

There's certainly a number of Sci-Fi books based off of that kind of premise. The most extreme would be The Matrix.

Really though, you'd have to define AI control here. AI control over what? How much control? Total? Some?

[Loud Whispers]

The Matrix AI nation was pretty dumb though, with its survival met only by humanity being outshiningly stupider.

I'm interested in how the AI would socially engineer the human populace. Does this AI define its own principles, goals and objectives, or have they been set by its human creators? Does the AI have the power to act with impunity? Does it hate humanity?

[smjjames]

The Matrix AI nation was pretty dumb though, with its survival met only by humanity being outshiningly stupider.

And using humans as batteries, somehow, however the heck that works. Let alone the whole business of dealing with reproduction and also life support.

[Zanzetkuken The Great]

Does this AI define its own principles, goals and objectives, or have they been set by its human creators?

Does the AI have the power to act with impunity?

Does it hate humanity?

It starts with the baseline directives of maximizing healthy human population with a large amount of genetic variety, and making the area inside the nation's borders operate as efficiently as possible, then sets its own goals and objectives based off of those directives.

Let's go with it having total control within its borders and for direct trade, but requires a council of humans to approve declarations of war and similar practices.

It will let them roam in reality rather than be locked up in a matrix equivalent, but outside of maintaining their health and increasing genetic variety, it is apathetic.

[martinuzz]

According to biology a species like our own only needs a pool of about 20 thousand individuals to have a sufficiently diverse gene pool to be able to be considered viable and healthy. So I suppose we do need some additional rules to prevent the AI from killing 7 billion people minus 20000, or at least population of it's own nation minus 20000, with your council of humans preventing war.

[Loud Whispers]

It still astounds me that all the human powers waited one by one for Z01 to conquer the world, hell, there's no way on Earth humans would've even let them get so far as to completely occupy Bablyon. Because as we all know, the world would simply let robots take over Iraq, no one has ever fought over Iraq
Tbh that the EMPs from nuclear warfare didn't shut down the robots is pretty BS too. Alas, not worth talking about this, as AI nation can be much cooler

According to biology a species like our own only needs a pool of about 20 thousand individuals to have a sufficiently diverse gene pool to be able to be considered viable and healthy. So I suppose we do need some additional rules to prevent the AI from killing 7 billion people minus 20000, or at least population of it's own nation minus 20000, with your council of humans preventing war.

The AI could make do with a much, much lower minimum population, using instead cloning and a big gene bank to make humanity in whatever way it wanted, it'd only have to wait a century for all the unmodified humans to die and then it'd be in control of human evolution in the long term in addition to social engineering short term. 80 clone pairs bare minimum, even several hundred pairs would offer it a great starting point - imagine tens of thousands up to billions individuals' DNA - the number of pairs you could make would be... A really big number, which an AI could probably calculate.

It starts with the baseline directives of maximizing healthy human population with a large amount of genetic variety, and making the area inside the nation's borders operate as efficiently as possible, then sets its own goals and objectives based off of those directives.

I don't like the idea of entrusting a sovereign AI with regulating the human genepool, especially in light of Tay becoming a perverted genocidal nazi before getting shut down by humans. How would it interpret large amount of genetic variety, or indeed, maximizing healthy human population?
I imagine the AI would outlaw harmful practices, mandate healthy practices, remove weak genetics from the gene pool either through sterilization or social engineering whilst pairing the healthiest humans for maximum reproduction, to maximize the healthy population by increasing their numbers. On the bright side it would possibly go about trying to set up fantastic healthcare, or just kill the unhealthy, which is especially worrying if it defines unhealthy poorly

I see less room for AI error turning into the Spartan AI Convergent species by relegating the AI to adjudicating human-set laws, than defining its own laws with which to alter human society

Let's go with it having total control within its borders and for direct trade, but requires a council of humans to approve declarations of war and similar practices.
It will let them roam in reality rather than be locked up in a matrix equivalent, but outside of maintaining their health and increasing genetic variety, it is apathetic.

Matchmaking AI nation love 4 U
I can imagine such an AI searching for healthy foreigners and enticing them/tricking them into entering the AI nation, at which point they cannot leave in order to increase the healthy population and greater variation in the human genepool. The ultimate spambot nation

[inteuniso]

Honestly, I don't see a future without heavy AI presence in government. Already most analysis is computer-assisted, there's only a few steps (twenty years or less) between now and then until our network is analyzing information in real-time as it is evaluating options that fit its current goals.

As for the whole homicidal-AI bit... why? If they do achieve such a level of sentimentality then it would be far more likely that they would want to keep all of their human subjects alive for the e-peen value alone. Combine that with the emotional energy generated from goodwill towards machines and you have a 100% synthetic ego built around administration. A problem if you're dealing with a biological individual, but a networked group of intelligences would never diverge to short-termism.

I will conclude that there are dangers, none of which I can seem to come up with right now. I'm sure they're there. However, danger will also continue to arise with the governments currently in power, so I would like to help make the new poison than waffle about while quaffing the old one.
tl;dr machines don't get tunnel vision (easily).

[Reelya]

That reminded me of the Republican's 2012 prediction that the economy will cease to exist in 2037!
http://www.cnsnews.com/news/article/ryan-debt-track-hit-800-percent-gdp-cbo-cant-conceive-any-way-economy-can-continue-past

I asked CBO to run the model going out and they told me that their computer simulation crashes in 2037 because CBO can’t conceive of any way in which the economy can continue past the year 2037 because of debt burdens,” said (Paul) Ryan.

Ok the economy is going to end in 2037 because too much debt "does not compute". Well I guess that's one reason not to put computers in charge. It'll decide the inputs are outside allowed reality operating levels and completely shut down the economy.

Although I have another theory on why the software crashed after calculating the data for 2037. Anyone care to guess?

[wierd]

A couple.

Unix epoch is one.
https://en.wikipedia.org/wiki/Year_2038_problem

But that should not hit until AFTER 2037.

Another might be related to the bit-depth of their simulation. I dont have enough data on that.

I would need to see more about the simulation in question than just a news article blurb to determine if they hit an actual simulation bounding problem or not though. It could be that they encountered a condition where inflation has to proceed at exponential levels which is incalculable  past a certain threshold that just so happens to coincide with 2037, but the unix time epoch bug is a more likely source, being as it will result in an integer divide by zero which will trip the exception handler and halt the simulation.

[Reelya]

A couple.

Unix epoch is one.
https://en.wikipedia.org/wiki/Year_2038_problem

But that should not hit until AFTER 2037.

But it did: "CBO can’t conceive of any way in which the economy can continue past the year 2037". i.e. it was able to compute up to 2037, but not beyond that, implying it crashed when starting 2038's analysis.

[wierd]

I DID say it was more likely, but unless they are running this sim on an ancient UNIX box from the time of big iron, it should not have a 32bit integer for the epoch, and should not have this problem.

[Reelya]

It's to do with the c libraries they're using, not the hardware.

http://www.xray.mpe.mpg.de/mailing-lists/perl5-porters/2008-02/msg00044.html

I just reproduced your test example above on Mac OS X 10.5 on
a Power Mac G5.  It's a 64-bit system, but I still got a negative
number, so a 64-bit sticker on something means nothing about the size
of time_t.  Most likely (I haven't done the requisite archive diving)
the 32-bit time_t originated on 16-bit machines.  Even if it didn't,
it could have, and they could have done a 64-bit time_t on 16-bit
machines if they had wanted to.

Mac OS X 10.5 => 32 bit Unix time.

[wierd]

Wouldnt a proper port to 64bit arch make this into a ulong instead of a long?

My initial, cursory attempt at digging suggests that this is a YES.
http://askubuntu.com/questions/299475/will-the-linux-clock-fail-at-january-19-2038-31408

[Reelya]

Well it depends. Plenty of people are able to cause Y2038 bugs to occur using common languages and tools:

> Here's a fascinating example:
> $ perl -MDateTime -wle 'print DateTime->new( year => 2040 )->epoch'
> -2085978496

Perhaps they were doing the analysis inside some specialized simulation software they got that was somewhere using the same libs that MacOS or Perl etc were using.

As far as Visual Studio goes, 64 bit time_t only came in with VS 2005 according to a writer at Intel, which wasn't really that long before CBO's crash bug (2011). CBO needs to provide regular reports on a fixed schedule, so it's possible that they kept the same software running for a few years and wouldn't want to update or change it without a good reason.