Artificial Intelligence and the Technological Singularity

khorum

Pronouns: zie/zhem/zer
19,342
58,572
95d 45m
Babysitter robots should be the next big thing. Makes too much sense.

In the meantime, the big hit at CES is a Baking Robot.

 
  • Like
Reactions: chthonic-anemos

Cybsled

Bronze Baronet of the Realm
8,631
3,531
14d 8h 6m
Interesting, but as pointed out, we don’t fully understand other cascade effects. For instance, removal of this gene might impact memory suppression. While that can enhance learning, it could also make it more difficult to cope with trauma. It’s a really hard thing to account for. This is a perfect example of genes not being a 1:1 ratio of gene to changes on the body.

I can also understand the “Brave New World” fears that could arise.

I do think it is something with amazing applications potentially, but considering the life spans of people, you’re not going to get a full data set for a very long time.
 
  • Like
Reactions: chthonic-anemos

Ukerric

Bearded Ape
<Bronze Donator>
4,459
3,475
27d 28m
Quake Arena CTF is now officially dead:


Quick summaries: Set of specialized neural nets taught themselves to play CTF in 2-player Quake III procedural arena mode. Humans consistently lost until they slowed the Bots to human-level reflexes where they (the humans, that is) would win about 30% of the time. The only part where humans excelled over the bots: Long-range sniping, due to our massive visual cortex.
 

Whidon

Trakanon Raider
1,415
2,063
6d 14h 35m
Obviously, it's not realistic but I would love to know how a group of humans would fair if after playing 200k matches of Quake 3 in a row vs equally skilled opponents. Or sc2 ect..

it would still be fascinating to see how the AI compared to humans assuming both are equally "trained". My guess is the humans would destroy the ai in such a scenario but I doubt we will see it as it's more of a curiosity than something that would actually help AI research.
 
  • Like
Reactions: chthonic-anemos

Ukerric

Bearded Ape
<Bronze Donator>
4,459
3,475
27d 28m
Obviously, it's not realistic but I would love to know how a group of humans would fair if after playing 200k matches of Quake 3 in a row vs equally skilled opponents. Or sc2 ect..
In that case, you should be picking newborns that aren't exposed to any other stimuli. Because that's what Deepmind is: a virgin (tiny) brain without an idea of what it's seeing or doing at first.
 
  • Like
Reactions: chthonic-anemos

Aldarion

Vyemm Raider
1,657
3,339
13d 13h 13m
China’s CRISPR twins might have had their brains inadvertently enhanced

Not sure if it's relevant to this thread, but I'm deeply interested in CRISPR gene editing and shake my head at the people who want to block it because of moral reasons.
If youre just shaking your head, you're not thinking hard enough about it.

1. Current technology for CRISPR editing requires the deliberate destruction of very large numbers of human embryos. People can differ on the balance of rights between a child in utero and her mother, and thus arrive at different decisions; but deliberate farming and destruction of humans is uncomplicated evil.

2. Once we get past that, you have to take consent into account. We consider it unethical to modify a child's body without its consent (female genital mutilation, or male circumcision for those who value consistent reasoning). But turning the child into a GMO, with all the unforeseen consequences that entails, thats OK? We've known for a decade or more that epistasis is ubiquitous. And we now know that CRISPR editing has huge number of off-target effects, in addition to epistatic effects of the target site.

editing your kid's genes to improve them is as ethical as infant lobotomy, and as good an idea as sticking a fork in your kid's laptop to repair it.

3. CRISPR editing creates a new sentient species with advantages over humans. That never ends well for humans, and at best ends in genocide for one side or the other. If you don't accept this point you havent thought it through. CRISPR edited humans are not going to mate with mudbloods. This is inevitable and inarguable. The existing races are maintained by assortative mating, even though interracial mating has no biological negative consequences for the offspring (arguably it has advantages). Once there is a race with actual, proven, genetic advantages over mudbloods, the races will never interbreed again.

The day the Chinese edited human embryos with CRISPR was the birth of a new species, and one of our species will go the way of the Neandertal.

The Bene Gesserit were right, and humans should end the filthy Tlielaxu while we still can. But because a bunch of shortsighted idiots who dont even understand what epistasis is, think theyre qualified to have an opinion on this, and itd be neat to give little Timmy blue eyes and make him not need glasses.

And thats exactly how you get the filthy Tlielaxu.
 

TrollfaceDeux

Pronouns: zie/zhem/zer
<Bronze Donator>
19,575
3,726
42d 18h 36m
Quake Arena CTF is now officially dead:


Quick summaries: Set of specialized neural nets taught themselves to play CTF in 2-player Quake III procedural arena mode. Humans consistently lost until they slowed the Bots to human-level reflexes where they (the humans, that is) would win about 30% of the time. The only part where humans excelled over the bots: Long-range sniping, due to our massive visual cortex.
Skynet will fuck us so bad ....
 

iannis

Chairman Meow
25,253
18,389
35d 23h 30m
In that case, you should be picking newborns that aren't exposed to any other stimuli. Because that's what Deepmind is: a virgin (tiny) brain without an idea of what it's seeing or doing at first.
I wonder if that even exists biologically.

Look at infant animals, the term "instinct" may not be precise in any scientific sense but it does describe something real. Cats do teach each other how to hunt through observation, i've watched them do it. But that young cat is predisposed to effective learning. If you do not have a teacher cat but you do have two untrained cats, they'll train each other through play. But there's some predisposition there being expressed. Insects probably have some property of mind which we would have to call at least a rudimentary conciousness, and while bees do teach each other intracies of their dances they're also hatched knowing them in some raw form.

It may be, and it's just conjecture, that a virgin mind is really only possible digitally. If you use meat, meat comes with inbuilt biases. Silicon might too, but if the biases are so abstract that we can't recognize them then they can probably be dismissed until we begin to recognize them.
 

wormie

Silver Baronet of the Realm
6,606
9,105
69d 5h 47m
Dont compare this stuff to humans. Humans learn to do many different things in a span of a few years. These programs spend hundreds of years worth of time to learn to do a single task and cant do anything other than this one task. They are nowhere near what a human can do.
 

Lambourne

Blackwing Lair Raider
1,114
2,012
24d 3h 23m
True, but once it's learned something, it can be saved and copied forever. AI only needs to learn something once. If a human is as good at Quake that's great but once that human is gone, so is that skill. Basically Quake is a now a solved issue forever. If you extrapolate this to other issues, it's hard to see how this won't be changing the world in drastic ways in the next few decades.
 

wormie

Silver Baronet of the Realm
6,606
9,105
69d 5h 47m
Quake is solved but the program that plays Quake cant do anything else. And there is nothing as of yet that allows this singular ability to play Quake to be generalized. Every single one of these bots that play Go or chess or Starcraft or Dota or Quake have been more or less the same exact thing just tweaked and applied to a different problem. Its been 20 years and there has been no improvement in generalization and I doubt there can be given what these programs actually are. Comparing them to human ability to learn the world is silly, they are nowhere close.
 
  • Like
Reactions: chthonic-anemos