Print

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

[Zanzetkuken The Great]

Hey, focus.  We're talking about how a society under AI control might wind up looking.  You heading a bit off topic from that.

[Reelya]

But it is on topic. what happens when software has a "blind spot" but instead of being a side-show it's running the main deal? Things can go very wrong. But not because the machine is malicious, that's attributing emotions to the machine. The machine only fulfills what it's design allows it to do, it doesn't know or care about outcomes.

[wierd]

Not really OT.  Any AI that controls society would need to have integer sizes picked that are sensible, or it would suddenly crash and nobody would know or understand why, leading to widespread panic as humans attempt to seize power in the resulting vacuum. (Many would feel it is unsafe to allow any improvements at all to the master computer, as they would not feel comfortable ceding such power over even a little.)

A 128bit integer size would make this kind of thing unlikely to ever happen in the forseeable future. Humans could evolve into something else before you hit an overflow.

[Reelya]

Actually 128-bit time is still lacking. Let me explain why. Your AI system should have a time keeping system that's more accurate than any possible technology so that it's completely future-proofed. Therefore the base unit of time is the Planck Time. This means that 2^128 would only cover up to 1/30000th of a second. But then you'd have options: a 64-bit time would be understood to be that many multiples of 2^128 Planck Times past the current epoch (that would give you a 20 million year window at 1/30000th second accuracy), 128-bit would mean that same accuracy but up to 3*10^26 year past some reference event (dating the big bang?), and 256-bit would be a full-accuracy time clock, how many Planck times since the Big Bang.

[Tack]

I couldn't trust it, frankly.
If they switched out the AI every four years, yeah that would make sense

[Reelya]

Well they could take that AI mentioned in another thread that designs other AIs according to criteria. Basically random AIs would be made for each town, then bred together. And estimates would be taken for such things as death rate, crime, road fatalities and the like, and rogue AIs would be culled to be replaced with new mutants. e.g. if you local AI decided that in your house, specifically, the temperature should be set to 80 degrees Kelvin, via pumped Liquid nitrogen at 3am in the morning, we'd lower that AI's health score, and it wouldn't be replicated unless it also scored relatively highly in other areas.

But it's important not to completely cull the psycho or aberrant AIs, if ultimate wellbeing is your goal. Because you can't know whether the current "best we've got" is the peak, or merely a local maximum in the equation of possible choices vs desired outcomes. If we limit the AI such that it can only make "safe" choices that people are happy with, that ultimately dooms the project. No real changes will be made.

One important thing to think about for the longterm would be to avoid premature optimization. Basically that's when the system is geared to make small changes which make life better, but it ends up putting evolution in a cul-de-sac. An AI that optimizes a community has a real danger of being trapped in that. e.g. in a shitty stepford-wives type world, but any incremental change from that point, specifically makes things measurably worse. Sometimes you have to plough through worse conditions because you know or hope there's a better reality on the other side. Typical "optimize this" type AIs tend to get stuck in ruts, because change hurts people, and if the AI is designed to always min/max happiness and sadness, it's going to get stuck in some rut or other.

[inteuniso]

Typical "optimize this" type AIs tend to get stuck in ruts, because change hurts people, and if the AI is designed to always min/max happiness and sadness, it's going to get stuck in some rut or other.

In other words, we need AI that evolve in pseudo-biological survival of the fittest, with AI evaluating whether or not those evaluations are correct, and essentially mammalian redundancy/recursive neural networking. All happening at the speed that would make these relevant.

I think the largest change needed for the singularity to occur/succeed is an increase in the amount of information available. There is no way, for example, to measure the flow of water through a river without disrupting it, nor is there a way to check the insides of doors/walls/trees without actively destroying the outside. An increase in telemetry, provided with electronically conductive building materials (refer to technology thread for how I plan to do this), will improve conditions for people and their creations.

[martinuzz]

Buy the new Intel Overlord Alpha 0.49! It's brand new 32 bit empathy subprocessor should prevent prison overflows and morgue stack limits that were reported in the 0.48f alpha from happening again. Now with extra drone comforts!

[Tack]

Because the people are going to explicitly trust an AI which goes 'This is good for you' as roads are demolished across the globe.

[Reelya]

"Where we're going we don't need roads".

The AI will solve every dispute by using machine learning to drag up witty movie quotes that are relevant to the complaint. That way you'll feel like it's one of you.