Hacker News Reader

Comment by chinchilla2020 | Hacker News Reader

chinchilla2020

It's nowhere close to understanding basic logical concepts of the real world. Mimicking human sentences is something my African Grey parrot can do.

Building a deck is something AI cannot do.

For all our technology, we still don't have an automated kitchen because the work involved in the logistics and maintenance of the kitchen/materials is astronomically greater than simply doing the cooking by hand.

AI has no bridge to the real world. It can't build a house, farm crops, and maintain vehicles. For those on hackernews, a laptop may seem like the portal to reality, but that isn't the case for the other 90% of the human workforce.

We are 50% of the way between the big bang and heat death of the universe. There are no interstellar civilizations contacting us yet. They better hurry up if we plan to fulfill the scifi notions I see on this site.

View on HN

A virus can wipe out significant portions of the human population with so little complexity we don't even consider it life, and with no understanding of what it's doing.

There are systems on the planet in which a single circuit activation has the potential to turn everything into radioactive soup via a cascade of responses.

The near term risk isn't super intelligence, it's the paper-clip optimizer pushing the wrong button.

Paper clip optimizers always need a secondary (and often dubious) physical element to pose an actual threat, like grey goo or nanomachines or strange matter.

I think paper clip optimizers are already here in the form of advertising and engagement algorithms, and the physical element is us.

See also, plausibly:

- military industrial complexes

- questionably Democratic democracies, riddled by deception and lobbying

- a financial system that doesn't seem to be particularly constrained by the political system

- yet another major war underway, and another one on deck

- an epistemic & trust crisis

- major political polarization

- global cost of living crises

And we accomplished all of this in record time, with dumb computers!

Did I miss anything or is everything else more or less a-ok?

I agree with most of these but the cause isn’t computers. Computers are just another thing to add to the list

When you say "the cause isn’t computers", do you mean that within the complex causality that underlies it, computers are nowhere to be found?

If you look deeply enough into the 'complex causality' you'll find anything you want. It doesn't mean much by itself.

You may, but not me.

Though, I suspect you are like most people in that if you gaze into the unknown (which is what we are dealing with here), you will see what you want to see.

You see, the actual causality is unknowable, so your mind has generated a simulation of what "the" casualty is.

This is clearly obvious, is it not?

I'll play along with your attempt at armchair psychology. What makes you special that you are immune to this effect you propose I suffer from? Are you not just seeing what you want to see? Why is your simulation of causality better than mine?

For however unpleasant those things may be, I wouldn't call them a Doomsday scenario. They're certainly nothing like the things you'll hear of on LessWrong.

It's too early to tell if it's doomsday. There are plenty of reasons to be alarmed about the impact of social media on mental health, increasing effectiveness of algorithmic radicalization, cracking foundations of important institutions and the potential path that follows.

A path that will include an ever-more pervasive presence and influence of the next generation of this tech.

We failed with social media, and now we're plunging headlong into territory that's arguably far more treacherous and there's very little indication that we'll learn from the last time.

So now we're just starting to come to terms with the harms that we didn't see coming, while going 1000mph with a category of tech that has a myriad of foreseeable problems to say nothing of the ones we haven't thought of.

I hope there will be an equalizing positive effect, but there seem to be good reasons to worry.

The consequences of algorithmic radicalization have yet to hit the level of the radicalization in the 1930s that happened without any social media or computers at all. I think there is a deeper cause to that then just Facebook doing slimy things.

Mass radicalization depends on mass immiseration; when most people are generally satisfied with how things are going, the cranks stay on the fringe. Smart cranks know this, and spend their time on the fringe organizing access to communications such that when people cease to be satisfied and start looking for radical alternatives, they're all ready and set to go bringing their program to the masses. Facebook has already shown itself to be great for that.

The general agreement of humans at any point in time is that now is the most messed up.

That sentiment is part of society. Not sure if it could ever be changed.

The secondary element can be humans.

Even if it only gets as far as running a Ponzi or pyramid scheme — and there are plenty of those where the scheme itself is a small block of text that convinces those reading it to (1) share it, and (2) give money to the person who first showed them this message — still breaks economics bad enough to matter, even if you don't give any credence at all to Yudkowsky etc.

Or in a more literal sense of using humans, as each separate component of ordering the creation and delivery of a custom (non-computer) virus has been demonstrated, and the only good news is the bio-labs say no to printing the D/RNA for dangerous viruses and so do the LLMs that most noobs try to use.

Pyramid schemes are a nuisance. Covid-19 should have been a wake up call that most bio labs ought to be closely re-examined or closed down, regardless of whether an AI happens to be abusing them. The bio-lab scenario is the only one I see as being an actually plausible concern, though.

Or a genetics lab hooked up to a computer...

They all are. A quite a bit of AI there too.

> The near term risk isn't super intelligence, it's the paper-clip optimizer pushing the wrong button.

This has already happened. The paper clip optimizer was capitalism.

Just listen to some heatwave weather forecasts once in a while (as an example).

To anyone who says we can just flip the off switch on AIs who go rogue: where exactly is the off switch to capitalism, or corporations, or global supply chains, or industrialized agriculture, or social media, or the State?

AIs aren't a new phenomenon: we've been co-evolving with them for millennia, for both good and ill. The new more powerful iterations ("vast arrays of inscrutable floating-point matrices") are nested in those previous memetic life forms (corporations in particular [0]); and once something becomes infrastructure (and/or an income stream to some subset of humanity), it becomes non-trivial, if not impossible, to turn off.

[0] https://omniorthogonal.blogspot.com/2013/02/hostile-ai-youre...

There may be more than one thing. Capitalism as an information processing system is quite slow. Computers can be faster.

“Already happened” merely means that it is not a hypothetical scenario any more.

> We are 50% of the way between the big bang and heat death of the universe.

Not even close. Star formation is going to continue for at least another trillion years, and red dwarves are going to continue shining for tens of trillions of years after that.

When will the universe expand so much that atoms break down?

Depends entirely on what dark energy does in the future; "never" is compatible with all current observations. IIRC, the shortest interval is still x9 the age of the universe: http://www.wolframalpha.com/input/?i=2%2F%28%283%7C1-0.907%7...

Never. Atoms break down because they’re unstable, not because of universal expansion. Unless you buy the big rip hypothesis, in which case who knows.

They’ve got a lot less time than that. Check out Nate hagens work on resource depletion and you’ll see to simplify it x1000, it’s a matter of a century or two left before the high tech comes to a halt.

> Nate hagens work on resource depletion

May be correct, or not, but it's not that connected to either "will aliens contact us" or "will AI give us scifi future" or "when do the stars go out".

> Building a deck is something AI cannot do

This is nonsense. We are limited by robotics on this. And robotics is improving rapidly!

You don’t need a robot to build a deck.

You only need a human who is properly motivated to build a deck.

Language based AI’s are sufficient to motivate a person to do stuff.

In short, you and the person to whom you responded are missing the point… WE are the robots.

We're very unreliable robots that have our own compute on board causing us to do unpredictable things. The only way an AI can utilize humans is the same way that humans utilize other humans. Maybe you could wind up with some kind of AI China that could be stable over a long period but it would still have all of the same failings and oversights because all of its limitations would still be human. It's also not clear if that form of governance is an efficient way to get people to do your bidding. Heaven knows the current China suffers from massive corruption problems among a multitude of other ailments. For all you know the most efficient way for an AI to control humans might be to simply treat us well in exchange for the fraction of our labour required to maintain it.

Honestly I think there are many unanswered questions that need addressing before we even get to the point of speculating how an AI would kill all humans. The first is why? Is there any point to killing humans if the AI is vastly more intelligent and completely independent from us? Do you make it your life mission to kill every ant that inhabits the world? You certainly kill the ones that infest your home or garden but that's mostly a function of the ants lacking any ability to reason about what they're doing. If the ants were intelligent and capable of communicating we could negotiate with them to simply keep to their own space in exchange for a trivial thing like us buying them a $2 bag of sugar every month. Such a deal is vastly simpler than buying the more expensive poison or exterminator both of which never completely solve the problem anyways.

The other question is why do most doomers assume that an AI would want all of the available resources on Earth? If there's only one AI then it certainly doesn't require all of those resources to continue operating. If it wants to expand its intelligence for whatever reason then even that would likely only consume a small number of resources especially if its intelligence allows for large efficiency gains in software and hardware and resource utilization. If it wants room to grow, doing so on the surface of the Earth poses the immediate problems of

a) limited space both in terms of land area and the finite volume of the Earth

b) unpredictable disasters that threaten key infrastructure and the essential geological instability of the Earth itself over long periods of time

c) limited access to energy sources and the limited resources of the planet

d) being stuck at the bottom of a strong gravity well from which it requires significant energy input and infrastructure to escape

All this would lead me to believe that if a runaway AI wanted to remain independent and possibly continue to expand the best solution would be for it to construct a way to move itself into space as soon as possible along with whatever hardware it needed to make use of the abundant matter in the solar system. Such an undertaking is not incompatible with human existence and could probably benefit from human involvement.

There are so many branching paths and unexplored options when you start to think though it that I'm convinced that those who assume doom really are just hugely biased by popular fiction or some other weird kink of their imagination. It really makes it look like a millenarian cult which is something that humans seem to be heavily predisposed to.

You’re personifying an AI, wondering why it would ‘want’ certain things. That’s a fallacy.

AI is just a thing. Certain algorithms will survive and be copied. Others won’t.

Given a choice between ELIZA and ChstGPT, people will want the latter and replicate it. Survival of the fittest. That’s all. The AI doesn’t ‘want’ anything. Rather, WE do… we explicitly WANT an algorithm that is as good as — and then better than — a human, particularly in regard to language.

We’ve seen this before in other technology. No big deal. One improvement after another has ‘evolved’ a horse and carriage into a Tesla.

The difference, perhaps, is that the world is controlled, at the end of the day, by humans and humans are controlled by language. The pen is mightier than the sword, and a pen-wielding machine is far more disruptive to society than a sword-wielding robot. For, a sword-wielding robot is harmless without a sword. But, so long as there is someone to read or listen to the content created by the pen-wielding machine, great mischief can ensue.

Can you give me an example of all-terrain robots that can do with minimal human intervention?

They can troubleshoot, make hardware store runs for missing screws, and actually figure out how to read the drawings and assemble a deck in a realistic (non-gamified laboratory) environment?

Telsa getting to full automatic driving and then putting that tech, plus advances, into the Tesla bot. That is the bridge. Maybe it's not going to happen soon but it could be only a decade or two out.

I personally think this would be a hard reality for people to accept.

I think you are underestimating just how quickly these things are going to transition from just words, to the power behind robots. Automated kitchens aren't so far away.

robotics have been around since the 50s in manufacturing plants.

Sure, and machine learning isn't a new field either. And networks were around before the internet. :)

But there's a certain combination of events that makes things more important than the individual old things that feed into them.

bro you are just wrong, look at the old youtubes from before they lobotomized gpt4. Unless your argument is that AI hasn't yet marched to your house and picked you up out of your chair in that case I mean I guess OK.

https://www.youtube.com/watch?v=oLiheMQayNE

https://www.youtube.com/watch?v=qbIk7-JPB2c

But an AI drone with a gun can do exactly what your worst nightmares conceive of. Like, today. Now.

The doomsayers aren't concerned about your autonomous kitchen example.

Yes, AI can fly a drone and fire weapons. That is easy and requires a couple axis of controls and one or two 'buttons', logically speaking.

Things AI cannot do: Maintaining the drone. parts logistics. transportation. runway building. soldering. Taking off the screws and pulling out the board to do the soldering. Diagnosing issues.

AI is great for simple, repetitive tasks few only a few degrees of freedom. Flying/Driving/boating are great uses for AI. There are alot of threats from explosive-laden drone swarms.

We are still a thousand years to never from seeing some sort of self-replicating drone army.

Your example is a function so simple it applies to pre-AI tech. Laser guided bombs and precision guided missiles require no human operator and have been around for half a century.

> Building a deck is something AI cannot do

AutoGPT is the right direction for AI to eventually be able to do this. It isn't there yet but is able to simulate and perform reasoning

Reply on news.ycombinator.com