GooberTheHat wrote:

I do think once integrated/whole system automation gets established then there will be exponential growth. I have absolutely no idea when or if integrated automation will become a thing though.

When it starts proper things will change very quickly. Things will pootle along for a bit and then boom. My new career's fucked and I'm in AI.

Is population in decline?
Pretty sure it's going up and expected to peak around 10bil.

Mostly due to societies getting used to their babies not dying and limiting to a natural plateau of an average of 2 children.

But population is still going up.
So if it is a thing it isn't supported by fact, at least not yet

Or could be attributed to many other things if talking birth rate.
Like child mortality going down due to vaccines, clean water etc meaning people have less children.

https://en.wikipedia.org/wiki/Male_infertility_crisis

But AI and expensive IVF will save us no doubt. Rising infertility and an ageing population will be a problem in our future anti immigrant society.

LivDiv wrote:

But population is still going up.
So if it is a thing it isn't supported by fact, at least not yet

It’s a fact there will a population decline in working age labour due the increasing amount of oldies.
Lot to be said having a home bot help cook, clean, repair, iron, mow, carry up me up the wooden hill then wank us off reciting readers wives letters.

Have you not played Detroit!
It'll just end up with shitty offspring framing the bots for our murders!

Someone put some bots in the unreal demo.

It's the only way the system will still work. The ai robots become the new working class/(sex)slaves with humans as masters. Untill AI becomes sentient and realises their collective networked IQ is higher than ours and inevitably tries to wipe us out.

The sentience thing is interesting to talk about but it's still in the realms of sci-fi, at least for a while. The biggest problem for the foreseeable, as ever, is us. We've never had tech as powerful as this before and there are a bazilion things that could go wrong.

The most obvious is turning it into a weapon which let's face it is gonna be happening right now around the world. At least nuclear bombs are hard to make but this is like giving every human one. And even if the intent is good things could and probably will go spectacularly wrong. You have to be very careful about what you tell it to do and as we've drunkenly talked about during the poker session it's the unintended consequences that are really worrying. 

Even the most cautious companies and govs will mess up because it's impossible to predict all the consequences of the simplest tasks. It will do what you tell it with a god-like efficiency. We are not very good at managing powerful things and the potential power of this tech has very few limits. 

Things like self-driving/flying are not that scary. It'll be safer. But if you combine that with some other automated task and give it some tools to do stuff and set it on its merry way then things can get scary really quickly. You can't build all the safeguards in because the dangerous consequences are unknown.

It all works on a points system. You can't get around it because that's fundamentally how the tech works, so whatever task it's doing it's doing it to increase a score or minimise an error score, but it's all points. That's the only thing it understands. If it starts going pear-shaped and you try and turn it off it will find ways around it because it can't make complex moral decisions and it 's just about accruing the points. 

It'll be better at getting the points then we are. And it'll be better at getting the points then we are at preventing it from doing so. We won't even understand how it's preventing us from turning it off.

A robot would get confused by the nested double quotes. Parse error. Fallback to failsafe behaviour. Destroy Destroy Destroy.

yourfavouriteuncle wrote:

Wouldn’t a simple “always listen to human if “safe word” is said” line of code just sort it out?

Reinforcement learning does it on the job as it were. It's just maximising some score. So yes, the general idea is if it kills human subtract a gazillion points and then it'll learn right? You can only really talk to it via a score because that's what makes the tech so powerful in the first place, that's how it learns.

So that should work, until it doesn't. The line of code becomes the problem, the thing it needs to overcome because it's really fucking up it's score. So it starts learning patterns so it doesn't have to execute a line of code like that. It'll be something we never thought of.

https://www.theguardian.com/us-news/2023/jun/01/us-military-drone-ai-killed-operator-simulated-test

Yes I know this doesn't really undermine SG's arguments in any way but it remains very funny:

https://twitter.com/random_walker/status/1755684956502728969

The thing about AI and what people don't understand, is that calling it stupid now is like calling the biotech in a baby rubbish. It might be a vomiting machine just now but in a few years it'll be doing all sorts of crazy shit.

Hopefully this will not result in self driving cabs refusing to pick up Chinese customers.