Print

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

Reality

13 (65%)

Universe

4 (20%)

The World

3 (15%)

Total Members Voted: 20

[Starver]

At the upper end, a dumb storage of every single possible position[1] vector-multiplied with every possible ruleset[2] wouldn't technically need to be AIed

More than atoms in the universe. Good luck storing that.

I'm suggesting repurposing the place they store all the universes[/i[, obviously.

[lemon10]

AI news:

Artificial intelligence is hitting the global labour market "like a tsunami" International Monetary Fund Managing Director Kristalina Georgieva said on Monday.
Artificial intelligence is likely to impact 60% of jobs in advanced economies and 40% of jobs around the world in the next two years, Georgieva told an event in Zurich.
"We have very little time to get people ready for it, businesses ready for it," she told the event organised by the Swiss Institute of International Studies, associated to the University of Zurich

Not particularly shocking news, especially since most of the companies that are adopting AI are hiding the fact that they are doing so because of backlash or other internal issues.
Shits going to get very hard for a lot of people when entire portions of the economy (RIP call centers) get all the jobs replaced by AI.
Even the people in physical jobs won't be safe since all the people that were automated away are going to still need work and will flock to harder to automate physical jobs, driving down wages.
---

OpenAI is pausing the use of the popular Sky voice in ChatGPT over concerns it sounds too much like the "Her" actress Scarlett Johansson.

You know how the GPT-o voice totally sounded like Scarlett Johanson from the movie Her and the whole circumstances basically match it to a T? Yeah, they just scrapped that because Scarlett hired lawyers and was about to sue them. Turns out that Altman *had* asked if they could use her voice and she unequivocally said no earlier.
---

Spoiler: Large Image (click to show/hide)

Comprehensive overview of all the related drama.
There has been an exodus of top safety researchers away from OpenAI over the last few years, with some quitting, being forced out, or merely being sidelined. They also dissolved their superalignment team that was in charge of figuring out how to make AI that doesn't kill everyone.
OpenAI's lax safety stance is super worrying to me, although I suppose if you don't really believe in AI it isn't a big deal.
---

Then the enemy tries to mess with AI and things become messy.

Look at a simple example. Chess engines. They beat humans easily... But what if we change the rules slightly? Human players will adapt instantly and successfully apply all their experience from regular chess. The chess engine needs to be retrained\reprogrammed.

General: El presidente, there is a problem with the new AI fighter jets, five have been shot down.
Presidente: WHAT! You promised it would be the equal of the CyberplaneX and would be impossible for any country but Xmerica to shoot down.
General: This black box recording will explain it.
The general turns on the TV, and on the screen a fleet of blimps armed with 16th century naval cannons appear. The AI planes are confused and unable to perceive either of the ancient technologies as threats, and just fly in circles while the blimps take shots at them. Eventually a cannonball manages to connect and a plane goes down.
General: I'm told the only way to fix it would be a complete new training run, it would cost another $100 trillion X-dollars.

Im willing to bet Air-to-air AI fighters would be easier to program than air-to-ground. Fewer things you have to program the AI to correctly identify. If it can reliably tell the difference between the aircraft you and your allies are using and those of enemies and non-combatants, then militaries might even give it the OK to fire at will at any target it identifies as an enemy aircraft.

Yeah, stuff in the air is categorically simpler, shooting stuff on the ground is much more nuanced, but even then its way simpler then trying to do AI stuff with a normal ground based robot. Accidentally killing noncombatants is the big concern, but that's fairly simple to solve by keeping a human in the loop and requiring them to OK every engagement.

[Eric Blank]

Honestly I can't imagine not having ground crews just monitoring aircraft at all times, manned or not.

[lemon10]

Honestly I can't imagine not having ground crews just monitoring aircraft at all times, manned or not.

Its not quite as relevant for the big AI aircraft (and obviously the blimp thing is a joke), but there will be very significant pressures to make drones fully autonomous and cut the human out of the loop entirely.
What if the drone is jammed and can't respond to a signal or is otherwise out of range? What if the 2 seconds of reaction time it takes for a human operator to respond would make the difference between killing a dozen hostile ground troops and the drone being blown up?

Military sources say one Ukrainian company alone is producing 45,000 cheap drones a month.

What if you have two million drones spread out across the country and don't have enough operators to respond to requests in a timely manner? What if you need the drone to operate in radio silence?


Obviously the tech isn't at the point where AI drones are really a thing yet, much less where its possible to stick locally run AI on tiny $1000 drones, but once the tech advances enough there will be strong incentives to have them be fully autonomous.

[King Zultan]

Honestly I can't imagine not having ground crews just monitoring aircraft at all times, manned or not.

Its not quite as relevant for the big AI aircraft (and obviously the blimp thing is a joke), but there will be very significant pressures to make drones fully autonomous and cut the human out of the loop entirely.

I don't believe that because somebody has to maintenance the things and nobody's built robots that can do that yet.

[Starver]

Its not quite as relevant for the big AI aircraft (and obviously the blimp thing is a joke),

Well, I know for certain that in GTA: San Andreas, if you annoyed the authorities enough whilst flying the jumpjet, it was significantly harder to avoid the jumpjets (and missiles) sent to kill you than if you were instead flying the light plane. But the player wouldn't have the same difficulties if the position was reversed.

Checkmate, Rockstar Games!

( )

[lemon10]

I don't believe that because somebody has to maintenance the things and nobody's built robots that can do that yet.

Various people have many definitions of what constitutes a lethal autonomous weapon. The official United States Department of Defense Policy on Autonomy in Weapon Systems, defines an Autonomous Weapons Systems as, "A weapon system that, once activated, can select and engage targets without further intervention by a human operator."
...
Other organizations, however, are setting the standard of autonomous weapon system in a higher position. The British Ministry of Defence defines autonomous weapon systems as "systems that are capable of understanding higher level intent and direction. From this understanding and its perception of its environment, such a system is able to take appropriate action to bring about a desired state. It is capable of deciding a course of action, from a number of alternatives, without depending on human oversight and control - such human engagement with the system may still be present, though. While the overall activity of an autonomous unmanned aircraft will be predictable, individual actions may not be."

An autonomous weapon system is one that can fight and decide targets on its own, it doesn't mean it can repair/rearm itself or that they can make more of themselves a la skynet.
E:

human-in-the-loop: a human must instigate the action of the weapon (in other words not fully autonomous).
human-on-the-loop: a human may abort an action.
human-out-of-the-loop: no human action is involved.

And cutting humans out of the loop doesn't mean humans aren't involved at all, it merely means that they aren't needed for the system to decide to target or kill someone.

[MaxTheFox]

The only "AI safety" that matters to me is keeping it from being used to oppress... which is what AI corpos do anyways, "safety team" or not. There is nothing that would make me believe that the GPT model can ever pose enough of a threat by itself, because in the end transformers are just a sophisticated way to predict text. There is no agency there.

Now, AI weapons... those won't cause an apocalypse but I can just foresee one accidentally classifying a refugee camp as a military base and bombing the shit out of it. Or a side in a war intentionally doing that and passing it off as an AI malfunction. Mark my words.

Yudkowsky and co are full of shit, every single one of their "safety" policies would do nothing to solve the actual problem with AI: capitalism using it. Because they are libertarians who know some math but know jack shit about sociology, history, economics, or actual philosophy that doesn't involve reducing people to mathematical models.

[McTraveller]

I saw a screenshot today of someone saying something like "I want AI to do laundry and wash dishes so I can make art. I don't want AI to make art so I can do do laundry and dishes."

I thought this was pretty insightful... the things for which people are doing AI today are the "fun" things... why aren't we using AI for the "tedious" things?

[Starver]

Art is an easy target. By dint of "anything can be art", with so many human schools of thought on the subject, if you couldn't (and you could!) train up an AI to do Pointalism, perhaps you could let it do some Dada. You see how readily we accept the "amazing" AI images that happen to have weird, multiplied and distorted fingers, for the sake of getting some output to "I want Donald Duck riding a yak, chasing a velociraptor" that (to someone) is art.

Washing up has a rather obvious metric to it. Do the dishes get cleaned? Do they also stay intact? And can all this happen without any significant human intervention?  Ideally, you'd never have to worry about washing up again because you finish your meal and leave the table and, by the time the next meal is needed, everything is clean and stacked back in the cupboard (or wherever the 'cook a meal' AI expects to find the dishes and cutlery). Oh, and not clear away the meal that you're in the middle of eating, just because you got distracted (or even though you didn't get distracted, and tried to vigorously swat away the robot arm, mid-munch).

(Also, it does rather need more than just cloud computing, at least until we decide to pass control over to some sort of fabrication device and let the process try to develop actually non-insane mechanical configurations, one of which might be better at manipulating tableware than anything we'd currently design ourselves.)

Although it's a bit outdated, the algorithms are that much better these days, the issue raised in this classic still have some relevence here. AI creeps up on us through the lower-hanging-fruit, but the AI's idea of low fruit might not match with a human's.

[Maximum Spin]

I saw a screenshot today of someone saying something like "I want AI to do laundry and wash dishes so I can make art. I don't want AI to make art so I can do do laundry and dishes."

I thought this was pretty insightful... the things for which people are doing AI today are the "fun" things... why aren't we using AI for the "tedious" things?

We have AI to do laundry and wash dishes. The thing is, those tasks are so simple and tedious that you don't need modern fancy neural-network AI to do it, you can get by with basic old-fashioned chatbot-style rules-based AI, which has become so commonplace now that we don't even call it AI anymore. Instead, we call them "washing machines" and "dishwashers", because hardcoded multi-step algorithms with little or no variation don't impress anyone anymore, even when they're perfectly effective.

But humans are lazy so now we find just the barest effort of having to bring the laundry or dishes to the robots ourselves and then take them away afterward is too tedious, and that's a lot harder to automate without the enormous capital expenditure of refitting every home in America into a standardized factory floor configuration. But if you ask me, that's just expecting too much.
(All right, it's also fair to say that dishwashers and laundry machines are widely perceived to work a lot worse than they used to due to water-saving measures, but that's beside the point.)

[Rolan7]

I saw a screenshot today of someone saying something like "I want AI to do laundry and wash dishes so I can make art. I don't want AI to make art so I can do do laundry and dishes."

I thought this was pretty insightful... the things for which people are doing AI today are the "fun" things... why aren't we using AI for the "tedious" things?

In short?  Neofascists hate artists so they're using AI to steal art from artists and pass it off as brutal efficiency.
They have/want mommy-wives for that other shit.
It's more complicated than that... but not really.

[Maximum Spin]

Art is an easy target. By dint of "anything can be art", with so many human schools of thought on the subject, if you couldn't (and you could!) train up an AI to do Pointalism, perhaps you could let it do some Dada. You see how readily we accept the "amazing" AI images that happen to have weird, multiplied and distorted fingers, for the sake of getting some output to "I want Donald Duck riding a yak, chasing a velociraptor" that (to someone) is art.

More than that, though, text and image content recognition was an extremely important early use case of AI, which people have - just like that xkcd you quoted suggests - been working on for years; and it's easy to show that a symmetry law holds where any content recognition algorithm can be refit into a content generation algorithm, so text and image generation was the lowest imaginable hanging fruit. You may recall the early seeds of this in the form of the "google AI dreams of dogs" articles from years past. There's no need for any other explanation; it's just something that's inherently easy because of things we already wanted to do.

[lemon10]

I saw a screenshot today of someone saying something like "I want AI to do laundry and wash dishes so I can make art. I don't want AI to make art so I can do do laundry and dishes."

I thought this was pretty insightful... the things for which people are doing AI today are the "fun" things... why aren't we using AI for the "tedious" things?

It is being used for tedious things though. Its being used for data input and sorting, summarizing corporate meetings, replacing workers in call centers, writing essays for middle school students, filling and analyzing medical records, ect.
Its just that most people this is impacting negatively don't have the same power as writers and artists do.

But don't worry, they're working on making robots to take all the physical jobs like doing laundry and dishes as well.

[MaxTheFox]

I refuse to use AI for actually writing my books but I do use it for parts of the writing occupation I find tedious like coming up with names, or shortening advertising blurbs (if brevity was the soul of wit, my wit would be a philosophical zombie). And I cheated using it in some essays on topics I found frustrating or boring to write about (like classic literature, or local history... and then I manually reworded a few sentences to fool those AI detectors, the professor didn't suspect a thing!). I suppose that's the textual equivalent of doing laundry.

But I do want a robot maid.