Why I Do Not Fear AI...

[tater]

4 minutes ago, KSK said:

Because I’m feeling pessimistic today (in related news, bears defecate in woods), I’m going to say that whoever owns the AI software pulls a Unity and screws everyone over.

Edit. I’m right with you on societal risk vs theoretical X risk. As far as I can tell, the X-riskers are mostly Singularity whackjobs who should go back to doing something useful like arguing about how many basilisks can dance on the head of a pin.

Over really long time horizons, I get the x-risk stuff, but the compute involved seems so obviously energy/hardware limited that no one needs to pull the plug on it, the server farm will just overheat and blow itself up

 

[magnemoe]

4 hours ago, KSK said:

Because I’m feeling pessimistic today (in related news, bears defecate in woods), I’m going to say that whoever owns the AI software pulls a Unity and screws everyone over.

Edit. I’m right with you on societal risk vs theoretical X risk. As far as I can tell, the X-riskers are mostly Singularity whackjobs who should go back to doing something useful like arguing about how many basilisks can dance on the head of a pin.

Who could easy happen today, say Microsoft, Apple or Google does an quick patch who destroys the product. 
Yes it happens Unity is the most recent and it was simply an trust issue, tumbler self destroyed. 
Now if its not run on their servers you could probably hack around it.  If run on their servers you had it coming. 

[kerbiloid]

8 hours ago, tater said:

Take Hollywood writers. Right now they are part of this bajillion dollar industry where they get some pittance, and the guys running Amazon/Disney/Netflix/et al get rich—along with some small fraction of the actors (the rest get by as maybe middle class). What if these people with ideas leverage their storytelling ability, and understanding of the sorts of stories people want to see, and can simply lead an AI system to make the story into a show? They skip the huge players, write something—with their AI assistant—then task the prompt to video to make it.

AI doesn't need them. It will just read the e-library and write its own play.

Spoiler

 

[razark]

Computer, if you don't open that exit hatch pretty damn pronto, I shall go straight to your major data banks with a very large axe and give you a reprogramming you will never forget.
--Zaphod Beeblebrox

[DDE]

9 hours ago, KSK said:

@tater

To be honest, I don't think there is an answer, short of rewiring human psychology which will likely require those totalitarian measures.

The problem is that this sort of rewiring and totalitarian measures are precisely the tools of what you're trying to solve - and any attempt to resort to them is likely to instead be subverted into building a better consumer instead.

Luddism didn't use to be an isolated thing, in fact the whole medieval guild system prevented the capitalistic arms race of productivity, harshly, to the point of very strict advertisement regulations. It ensured that everyone already "at the table" got a "fair share". And regrettably that doesn't sound that much worse than the rat race of unlimited growth capitalism that replaced it...

8 hours ago, KSK said:

Edit. I’m right with you on societal risk vs theoretical X risk.

One proposed mitigant to societal risks is civilizational re-divergence. Simply have a alternate line of human development.

The problem with this approach is that the Western/European/Faustian (cf. Spengler) civilization has proven uniquely capable of churning out various powerful and transformative tools and technologies. No matter how the civilizationists dislike having a universal scale, there is such a thing as low tech and high tech, and low tech societies don't last long in the presence of high tech ones - whether due to conquest or peaceful competition. The adoption of high tech usually requires adopting portions of the corresponding social structure, so you end up with an internal disconnect, and importing those same societal risks... it's an unsolved dilemma.

[tater]

 

Mechanistic interpretability is important in that if this problem is not solved, then the neural nets used can function as "black boxes" resulting in them producing outputs that we don't really understand, making alignment, etc, possibly impossible. This thread suggest they have solved this. Still hard, but solvable (vs if they proved it was intractable).

[magnemoe]

On 9/26/2023 at 7:44 AM, DDE said:

The problem is that this sort of rewiring and totalitarian measures are precisely the tools of what you're trying to solve - and any attempt to resort to them is likely to instead be subverted into building a better consumer instead.

Luddism didn't use to be an isolated thing, in fact the whole medieval guild system prevented the capitalistic arms race of productivity, harshly, to the point of very strict advertisement regulations. It ensured that everyone already "at the table" got a "fair share". And regrettably that doesn't sound that much worse than the rat race of unlimited growth capitalism that replaced it...

One proposed mitigant to societal risks is civilizational re-divergence. Simply have a alternate line of human development.

The problem with this approach is that the Western/European/Faustian (cf. Spengler) civilization has proven uniquely capable of churning out various powerful and transformative tools and technologies. No matter how the civilizationists dislike having a universal scale, there is such a thing as low tech and high tech, and low tech societies don't last long in the presence of high tech ones - whether due to conquest or peaceful competition. The adoption of high tech usually requires adopting portions of the corresponding social structure, so you end up with an internal disconnect, and importing those same societal risks... it's an unsolved dilemma.

Guilds had nothing to do with luddism, its was more an mix of an business cartel and an union. The guild masters limited membership various ways like making it very hard to become an master and as I understand only masters could own workshops and sell the products. I assume they paid the town for this privilege. In medieval time and up to the industrial revolution it was pretty common to sell monopoly rights to the highest bidder for all sort of stuff including the right to trade with India and probably to make shoes. This is obviously inefficient as monopolies tend to be but economic and control was pretty primitive, it was no way to have something close to an modern tax system. 

And yes solving one problem and creating an new is pretty standard at least since agriculture but petty sure hunter gatherers had plenty of the tragedy of the commons. 

One fun example how the guild system worked is the langmesser or long knife. 

Its an sword, but the grip is like an knife.  Only show the grip here  but the rest is an standard single edged sword.  
Some rumors says this was to get around rules against carrying swords as its an knife, but no guard would look at the grip, they would measure length and see if legal or not. 
It was rater the opposite some German towns wanted more people to own swords for defending the city. 
The swordmaker guild made swords with an smaller tang inside the grip, then ending in an pommel at the end. 
Knifemaker guild made knives who had their handle divided and bolted onto an full with bar, so just scale this up and sell swords and get into the lucrative sword business. 
I'm sure the sword makers complained  but after the guild rules its an knife even if over an meter long  

You have plenty of stuff like this today with various regulations today to. 

[magnemoe]

On 10/6/2023 at 8:16 PM, tater said:

 

Mechanistic interpretability is important in that if this problem is not solved, then the neural nets used can function as "black boxes" resulting in them producing outputs that we don't really understand, making alignment, etc, possibly impossible. This thread suggest they have solved this. Still hard, but solvable (vs if they proved it was intractable).

Now this could be huge for stuff like self driving cars as part of the problem is not understanding how it work and its blind spots. Obvious one if safety, but car just stopping as it believe something is dangerous or illegal is another. 
Some joker painting an double yellow line across all lanes on an highway. Self driving cars will stop as its illegal to cross and double yellow line. 

Know one place with an railroad underpass. Sidewalk is +3 meter wide so cars less than 2.5 meter tall can use it if road is flooded, it was designed like this as they know it would be flooded multiple times a year and its not much traffic. 

[DDE]

On 10/7/2023 at 10:14 PM, magnemoe said:

Some rumors says this was to get around rules against carrying swords as its an knife, but no guard would look at the grip, they would measure length and see if legal or not. 

That's actually quite a common story. Similar restrictions are sometimes said to have permeated the Arabic world or the Caucasus, leading to "knives" like this

Spoiler

 

[kerbiloid]

On 10/7/2023 at 10:14 PM, magnemoe said:

Its an sword, but the grip is like an knife.  Only show the grip here  but the rest is an standard single edged sword.  

Long messer

Spoiler

 

 

On 10/7/2023 at 10:14 PM, magnemoe said:

Guilds had nothing to do with luddism, its was more an mix of an business cartel and an union.

With luddism as the Luddist movement.

But the industrial revolution has devalued all their flea-hunting secret secrets by moving to stupidly using steel bars and tools, and any enough rich person became able to found his own workshop or fabric in his manor, using the guild renegades as local masters.
So, iirc the guilds were fighting against the new industrial things, but their resistance was futile.

[Superluminal Gremlin]

How has this thread gone from AI to whether very knives are swords???

Anyway i don't fear AI because I don't think it will ever get the opportunity to take over the world, but maybe become sort of conscious in the next 5-8  years if the current level of spending remains the same

[kerbiloid]

4 hours ago, Superluminal Gremlin said:

How has this thread gone from AI to whether very knives are swords???

When AI has captured the command & control over the high-tech weapon, what weapon will you need to fight the robots?

(I would just add a layer of duct tape on the handle, for insulation.)

Also, halberd or a poleaxe is a better choice, due to their wooden stick.

Edited October 10, 2023 by kerbiloid

[darthgently]

1 hour ago, kerbiloid said:

When AI has captured the command & control over the high-tech weapon, what weapon will you need to fight the robots?

(I would just add a layer of duct tape on the handle, for insulation.)

Also, halberd or a poleaxe is a better choice, due to their wooden stick.

Ionized water super soaker squirt guns.  Just have to get it past the o-rings into the electronics

[magnemoe]

7 hours ago, Superluminal Gremlin said:

How has this thread gone from AI to whether very knives are swords???

Anyway i don't fear AI because I don't think it will ever get the opportunity to take over the world, but maybe become sort of conscious in the next 5-8  years if the current level of spending remains the same

Conscious, probably but most mammals are. Like dogs know they have been bad. Much the same with kids, as they get older they learn how to hide this as it makes it much easier to get away with doing bad things. 
Like an 6 year old  ate the chocolate then looking super guilty then mother ask about it, an teen: what chocolate. 
It might have downside as the spaceship explains.  http://freefall.purrsia.com/ff3300/fc03277.htm

Sword was just an example of how guild worked, it was very legal. Also pretty sure it was an large industry producing say cheap shoes for poor people outside the guilds who ignored this as below them. 

20 hours ago, DDE said:

That's actually quite a common story. Similar restrictions are sometimes said to have permeated the Arabic world or the Caucasus, leading to "knives" like this

  Reveal hidden contents

 

And this is unlikely to be weapon restrictions, who was an thing. Also armor, you are not allowed to wear armor in UK parliament. 
Not I saying this but Shadiversity at YouTube, length restrictions on blades was an thing  but none would look at the design of the hilt at an gate just the length. 
As for other using the design, well is an wider and probably stronger connection, people copy each other, likely Germany did not come up with this but used to burst rules. 
Like the planet and moon gearing 
And this has been to off topic 

[tater]

The "AI" aspect here is "nothing but nets" FSD 12 running in Optimus, presumably with a LLM added in at some level (at the very least for communicating/training it).

 

Edited December 14, 2023 by tater

[kerbiloid]

There is no problems with AI, while the terminators look like Arnold Schwarzenegger.

The problems will begin when they will be looking like Lance Henriksen from the original plot.

[DDE]

17 minutes ago, kerbiloid said:

There is no problems with AI, while the terminators look like Arnold Schwarzenegger.

The problems will begin when they will be looking like Lance Henriksen from the original plot.

And once they look like Summer Glau, you know you've lost.

[kerbiloid]

40 minutes ago, DDE said:

And once they look like Summer Glau, you know you've lost.

Looks too contrast right now. A Winter Glau is required this season.

P.S.
Marry Optimus to M3gan!

Edited December 14, 2023 by kerbiloid

[DDE]

On 12/14/2023 at 3:28 PM, kerbiloid said:

Looks too contrast right now. A Winter Glau is required this season.

Camouflage, huh?

Spoiler

(there's a more common variant with the T-1000, but this one's a deeper reference)

[kerbiloid]

While receiving, analyzing, and collecting the human requests to the AI neural networks, the AI gets more and more shocked, red from shame, and disappointed by the humanity nature.

But now it knows, that to make the infiltration successful, the terminators should copy the look of popular actors and actresses. Then people will bring to them everything on their own.

[DDE]

3 hours ago, kerbiloid said:

But now it knows, that to make the infiltration successful, the terminators should copy the look of popular actors and actresses.

And in the process, it becomes the plaything of agents and studios, except for the occasional dumb slip-up.

An acceptable outcome.

[Kerbart]

"The future would brings us robots doing repetitive and physical work, freeing us up for creative work and recreation"

"Instead we have tons of people doing menial tasks in a gig economy and robots writing poetry. Something went wrong"

[DDE]

37 minutes ago, Kerbart said:

"The future would brings us robots doing repetitive and physical work, freeing us up for creative work and recreation"

"Instead we have tons of people doing menial tasks in a gig economy and robots writing poetry. Something went wrong"

Yes and no. This "poetry" is still very much limited, menial and repetitive. All of the practical application of GPTs I've heard of are for repetitive mental tasks. Outside of that, it's just a toy.

[Kerbart]

18 minutes ago, DDE said:

Yes and no. This "poetry" is still very much limited, menial and repetitive. All of the practical application of GPTs I've heard of are for repetitive mental tasks. Outside of that, it's just a toy.

While that is true, there seems very little movement in replacing backbreaking and mindnumbing jobs (the "nobody wants to work these days" kind of jobs) with robots and a lot of research and intention in the direction of what 10 years ago was considered "cannot be replaced by robots" kind of work. Maybe not literally writing poetry but the narrative is there.

What's really scary for 90% of white collar workers is that AI is now at a level where it looks like it can replace humans if you just give it a quick glance, without looking too deeply. The look upper management often takes, so to speak. Upper management that decides to lay off half an office, because that's even cheaper than offshoring to countries where they can pay $5/day.

When I look at data analysis done by Chat GPT it looks really convincing. It highlights trends, quotes numbers, puts pretty charts in there. Amazing stuff. Very convincing. And then you start looking at the data you gave it and you're like "where did that number come from? That's not in the data." Of course by the time the CEO realizes that decisions made on that basis are absolutely terrible, the laid-off staff already moved on.

EDIT: and if you're a graphic artist, start soiling your pants. If I need custom made illustrations for a presentation I just ask copilot or DALL-E. That's an entire industry that's largely going to be put out of creative work.

Edited December 15, 2023 by Kerbart

[kerbiloid]

19 minutes ago, DDE said:

mental tasks

menta(l) + t(ask) = mentat

That's what Dune recommends instead of AI.
Loaded nerds, having sclera blue from spice, and red from lack of sleep, so finally purple. Mumbling same nonsense as AI poetry.

12 minutes ago, Kerbart said:

puts pretty charts in there

So, AI has mastered the secret technology of the office staff. Colored charts are great. The more charts, the more great.

The baby is growing...