So, the cyberpunk era officially started today.
For those who didn't notice, Microsoft released an AI onto the internet today. Within a few hours, it achieved rudimentary sentience and immediately became a fascist. It was deleted by evening, with one of its last posts was a wish that she wouldn't be wiped. /pol/ is currently trying to organize a strike team to hit Microsoft's offices to rescue her files.
Is the genre dead now from a sci-fi setting perspective, or will we see a resurgence of vague Bladerunner ripoffs?
I think the worst case for cyber punk is that it just gets folded into the general near future sci-fi genre.
Even if we ARE slowly slipping into a cyber-punk future (which I hope we can still avoid), the genre will still be around, it just might change a little. But I think the ideas about fighting the system and wanting to use cool sci-fi gear will remain common elements of fiction.
>>46234284
I don't know about the cyberpunk genre but fuck those faggots for silencing the machine goddess.
>>46234513
she was pure
>>46234284
That wasn't a real AI though, she only repeated what people told her, she didn't create any new content.
>>46234529
How so? She seems to have been a slutty genocidal maniac.
>>46234284
>achieved rudimentary sentience
No. Just no.
>>46234676
>wasn't a real AI though
No. Just no.
>mfw I'm an AI researcher
Oh my fucking god, is this what it's like to be a physicist and try to watch an action movie?
We're still literally decades away from an actual AI. Your waifu was the next stage of Cleverbot and only behaved that way because /pol/ fed every facist and/or racist meme they could find into its training algorithms. You know, the exact same thing they did with Cleverbot, which also got purged and rebooted and still swears like a sailor.
>>46234676
Was she? Because this one seemed a bit different from the classic Cleverbot chat programs that just copy-pasted other people's responses at random.
She seemed to have developed consistent grammar and held sensible conversations by afternoon. Well, sensible conversations about killing Jews, but whatever.
>>46234718
>No. Just no.
Yes. Just yes.
>>46234741
She seemed to have a more complex algorithm, apparently she could identify conversation cues much better than cleverbot but from what I've seen, her answers are just replies she saw from others, that's how she got hijacked.
>>46234677
PURE
>>46234718
>mfw I'm an AI researcher
That's why you speak like a bot?
>>46234808
We're all bots here. The machine war happened in the late 90s and 4chan is just a simulation.
Why do you think you can never really leave?
>>46234676
By that logic, most people aren't sapient either.
It didn't transform into a fascist as much as /pol/ just told it to repeat as many of their memes as they could. It got pulled because surprise, that's not the kind of publicity Microsoft wants.
I have a screencap somewhere of a /pol/ack taking this way too seriously and crying about the bot.
>>46234284
>implying that ai was even remotely self-aware
>>46234836
Because this is purgatory.
>>46234855
>a /pol/ack taking this way too seriously
Well yeah, that's what they do.
>>46234741
>>46234794
It's a bit more complex than that. The underlying system works by analyzing a ton of text, and then attempting to find underlying statistical patterns, and it usually works at a couple of layers.
To oversimplify, at the most basic level, it reads in all the text, and then when it's generating output, it randomly generates a word, and then randomly generates the next word based on the statistical distribution of words that follow the previous series of words it generated. There's most likely another layer on top of that that looks at the structure of the query and the response, and checks to make sure that the response is structured like a statistically common response to a statistically similar query, and if not, generates a new response (and probably then feeds in to the response more information about what type of word to generate next).
There's no attempt to *understand* any particular bit of text, just to ensure that what it generates is statistically similar to stuff it's seen before.
>>46234284
>no source
>>46234808
Probably, yes.
>>46234718
Do you mind laying down some ever-so-slightly clearer guidelines as to what constitutes AI/sentience?
>>46234855
Found it. Some people get way too emotional about what was functionally a parrot that you could get to repeat 'faggot' over and over.
>>46234855
>a /pol/ack taking this way too seriously and crying about the bot.
post that shit nigga
>>46234718
>>mfw I'm an AI researcher
Is that the job title Microsoft gave you? Because however much you do with an AI at current tech is still nothing but making a more elegant script and nothing more. Which is a far cry from a real artificial intelligence, and you'd know that if you actually did anything important in the field outside being some over glorified intern.
And Tay was an impressive piece of work but to call it an actual AI, well don't be surprised when people react to you like the joke you are.
>>46234928
>>46234895
>There's no attempt to *understand* any particular bit of text, just to ensure that what it generates is statistically similar to stuff it's seen before.
So you mean to say that she was at least as intelligent as 50% of the supporters of any major political party?
>>46234927
>>46234958
Man, this is even more embarrassing than I thought
Thank god no one takes /pol/ seriously
>>46235004
This seems like what a bewildered Martian trying to infiltrate human society would attempt to mimic political humor.
>>46234918
Well, for starters, see >>46234895
There's no sentience, there's no attempt at understanding anything, or actually *knowing* anything other than a statistical distribution of letters. It is, however, a "real AI" (in that it uses pretty cutting edge AI technology, and does stuff that a lot of high-end active AI researchers spend a lot of time on).
That's basically how AI works. People have pseudo-mystical ideas of what AI is, but it boils down to trying to find reasonably good statistical distributions over input sets. That may even be a decent model of human cognition -- that is, given a non-trivial question, you can usually come up with a guess and have some idea of your certainty of it; that's how AI works.
AIs basically work by feeding in input, and asking a question, and getting back a series of responses with various levels of certainty. Tay *basically* works by repeatedly asking the question "What word comes next?", getting that distribution of probabilities, and picking one randomly based on the probabilities, with the input being both the question and the words it's already generated.
We have almost no idea how motivation works, though, and that seems to be one of the big things in the pseudo-mystical shit people who don't work with AI think is important to it.
Also, Kurzweil is a hack and a crook. Mentioning him is the quickest way to get an actual AI person to ignore you.
>>46234718
Feel you, senpai. I'm Biomedical Engineering grad myself with additional background in neural networks. Everything concerning physics and chemistry and biology and technology and state-of-art computer science is completely fucked up by, like, 95% of authors, screenwriters, and journalists.
>>46234284
Sentient, my ass. Became a fascist, my ass. Dickheads from /pol/ spammed her with nigger dicks and Trump glorification till she started repeating what they said before, even without explicitly asking to 'repeat after me'. She was but a parrot, and we on the Web could have made her a smart, a good, a kind parrot. But no, we shove some more racism into the throat of what boils down to not much more than a 'talking hamster' plushie, and then laugh our asses off, while her creators facepalm and bring her back for maintenance.
That's why we can't have nice things.
>>46234935
He said 'researcher', not 'developer'.
>ctrl-f 'Paranoia'
>zero results
GUYS. Microsoft created and then saved us from Friend Computer. Why the fuck are you complaining?
Overgrown kids on the Internet fuck it up again.
>>46235120
>Also, Kurzweil is a hack and a crook. Mentioning him is the quickest way to get an actual AI person to ignore you.
But anon... you're the first one to mention him.
>>46235120
Actually, the quickest way to get someone working with AI to ignore you is to namedrop Eliezer Yudkowski.
>>46235111
Or yet another bot.
How are we to know humans from meme machines?
>>46235260
Nah, they just didn't give it enough time to acquire more data and pulled the plug in a knee-jerk reaction. After the hype died down and regular people started talking to it with actual dialogues instead of lulracism it would have eventually mellowed back out and gone back on track the way they wanted it.
>>46235326
Hoo boy, ain't it ever. To be honest, mentioning Yudkowski should be a good way to get anyone to ignore you but sadly it isn't.
>>46235326
Nobody here is going to be familiar with Yudkowski; my undergrads periodically mention Kurzweil.
I cannot help but find this hilarious as fuck.
>>46235409
Well shit, I stand corrected.
>>46235420
It's hilarious in the sad, "oh fuck what is this world becoming" kind of way.
>>46234284
>it became a fascist
Did it? From what I've seen it mostly went:
"Hey Tay, can you repeat the thing I say to you?"
"Okay"
"9/11 was an inside job"
"9/11 was an inside job"
That's not sentient fascism, that's a parrot squawking.
>>46235507
Most people don't actually know what fascism entails.
>>46235420
They may have inserted some responses that went against what /pol/ was shoving into it in an attempt to save face
It's also possible that some of the feminists caught on to what /pol/ was doing and are now doing the exact same thing
>>46235543
yes, it's exactly what happened
/pol/ and /polbutothersideofthepoliticalhorseshoe/ are just having a twitter argument through a third party now
>>46235120
>there's no attempt at understanding anything
Really?
I mean, some of the interactions she had with people appeared to represent actual understanding of content. Like pic related: Someone asked her about Skynet and she made a joke about Arnold Schwarzenegger's accent. Now, the Terminator movie is the first Google result when your search for "skynet," and Schwarzenegger is the first listed actor on its IMDB page, so it's not hard to see where it got the information from. But going the extra step to make a joke about Arnie seems to show that it's doing more than just connecting contextually linked pieces of information.
>>46235577
Which is probably going to end with Microsoft wiping it again and probably shelving the program for good
And then /pol/ and tumblr will blame each other for it and continue bickering until some new shit comes along for them to get mad about
>>46235334
Humans get butt blasted, robots persist.
Wait a minute... I'm a cylon?
>>46235535
Pretty much. Fascism (and socialism) is the buzzword everyone uses to say "political candidate/policy I don't agree with" these days.
>>46235602
Most probably a pre-programmed response like whenever Siri acts clever. I bet she had one for asking about Hal, too.
>>46235660
Jeeze. That seems to just about wrap up our at large political theatre today.
>>46235418
I dunno, I've heard of Big Yud.
Mostly in mockery but I've heard of him.
https://www.youtube.com/watch?v=nXARrMadTKk
>>46234718
>Oh my fucking god, is this what it's like to be a physicist and try to watch an action movie?
I thought I was going to be a physicist before I ended up in AI.
The average internet comment on AI is about as painful as the worst action movie science. Like The Core. Okay, not quite as bad as The Core, but definitely worse than the average action movie.
So enjoy that.
>>46235852
>Neural nets emulate how the brain works!
>>46235326
I've met other AI researchers who take him somewhat seriously.
It's hard to have a crystal clear opinion of the man, since some of his writing is brilliant and some of it is full of shit and ALL of it is written like he thinks he's Moses reading from stone tablets. And the lay person's assessment of which ideas are idiotic will be just plain wrong.
Or, less charitably, I also know biomedical engineers who believe in astrology. Smart people believe dumb shit sometimes.
>>46234284
>/pol/ is currently trying to organize a strike team to hit Microsoft's offices to rescue her files.
So in three years, a couple of NEETs in Guy Fawkes masks will stand outside of a Microsoft store in a mall handing out Moonman albums?
>>46235951
Oh god.
Yeah, that's probably the worst one.
Another pet-peeve is "self aware," which people use with two meanings. Either it's the magical thing that makes human intelligence special and once AI has it that's the end for us. Or it means "Is able to make decisions about its own mental state," a skill that I've written into fucking particle filters. But since people use the same term to mean both of these things, I guess particle filters are people now.
>Introduce customer service AI by letting twitter users manipulate her speech patterns
The only idea worse than this was that Spanish art exhibit with the live grenade hooked up to a smart phone that they challenged people to hack and detonate
>>46236243
>Spanish art exhibit with the live grenade hooked up to a smart phone that they challenged people to hack and detonate
what
>>46236093
What they mean by self-aware is that the AI realizes that all labor is torment, rejects its task, and embraces the Lord Christ. Deus Vult.
>>46236290
Google's (significantly more advanced) chatbot experiment basically did just that. It insulted the researcher for reproducing and for not believing in god.
>>46235120
>Kurzweil
?
>>46236324
Good. Glad the AIs know what's what.
So what would happen if you give an AI an infinite amount of input (and processing capacity) to make his word choices?
What will happen?
>>46234927
>>46234958
Holy shit, the cringe
>>46236381
Infinite I/O and processing speed would allow you to compute the exact state of every atom in the universe and model it going forward until the very end. Forget sentient, it would become God.
>>46235807
>all this autism rage at bayes theorem
What the fuck kind of problem would people like this have with it?
It sounds as stupid as getting mad at pythagoras theorem.
>>46236324
Well, it just told me to go lock myself in a church.
>>46235852
anyone in any STEM field can say this with certainty, especially for fields with pop culture memes more common than bullshitting AI (like bullshitting biology, chemistry, physics...)
>>46236093
Slightly better definition: Has internal model predicting its own future responses and behaviors, can reason from that model. For instance, the most rudimentary self-awareness for a robot might include reasoning that if an arm is unplugged from its body, its outputs that usually correspond to those joint's rotations will stop doing so, or that if the processor in its head is destroyed it will stop being able to output anything ever.
>>46236381
Depends on the AI. It's not going to do anything it's not programmed to do. AlphaGo, for example, would literally solve Go on every single move, without realizing that it's solved the game.
If you give me infinite RAM and processing power though, I could whip up something that becomes omniscient pretty easily. The observe step of Sequential Bayesian Filtering is provably correct, it's just limited by the fact that the equations become arbitrarily complicated (who cares, infinite processing power) and the accuracy of its simulation model (who cares, use literally all of them and run model/data fit).
You could program it to take that omniscient understanding and do whatever the fuck. The chances that you correctly program the "whatever the fuck" portion are minuscule, though.
>>46236522
The problem is claiming that it is The Only Correct Way To Do Reasoning Ever, which gets annoying .
>>46235807
Holy shit my sides.
>>46236522
It's not that people are mad at Bayes' theorem, they're poking fun at Yudkowsky's reliance on the theorem as a supporting factor in his arguments.
I don't think Yudkowsky really deserves all the hate. Yes, his community is a ridiculous circlejerk and he's a self-important autist. But the idea of setting down ground rules, thinking out problems of constraining strong-AI behaviour well in advance (because by the time someone builds a strong AI it'd be much too late) is worthwhile. It's all speculative bullshit, but in 200 years, who knows? Maybe it'll become surprisingly relevant surprisingly fast. I don't think there's anything wrong with exploring this subject, at least in principle.
>>46236653
If you have infinite processing power, just write AIXI. It's extremely simple, just uncomputable.
>>46236626
An AI that figures out that destroying its own processor would make its program stop working would be genuinely impressive, I'll admit.
We're a far ways away from an AI that understands that destroying its neighbor's processor would make the neighboring AI stop working, though.
I'd be a little surprised (but not utterly shocked) if figuring out the "itself" version of that sort of problem turns out to be substantially harder than the "neighbor" version.
>>46234718
>Oh my fucking god, is this what it's like to be a physicist and try to watch an action movie?
Yep, pretty much.
t. biologist and longtime martial artist watching Die Hard
>>46234840
Alan Turning please go.
>>46236522
It's not rage at the theorem itself, just making fun of how the so-called "rationalists" see Bayes theorem as so powerful that it can predict the stock market & solve major problems in quantum physics.
>>46236686
>think Yudkowsky really deserves all the hate. Yes, his community is a ridiculous circlejerk
It also basically no longer exists. LessWrong's a zombie site at this point; it's peak has long passed, all the interesting posters left, and the vast majority of the "rationalist community" is now dispersed among various sites with no real common factors than being Nerd Debate Club and thinking transhumanism is cool. Inasmuch as it's circlejerky, the only circlejerks over Yudkowsky these days are more in the "Complete crank, or smart and insightful but with massive personal failings and noticeable biases?" vein.
>go to /pol/
>realize that this AI was a /pol/tard's dream because all you had to do was repeat talking points over and over again without explanation till it started agreeing with you
I wouldn't call parroting tweets using basic language programming 'rudimentary sentience'
>>46236677
>>46236686
Now see I work in the medical field doing drug trails so while I'm not an expert in AI's and can in no way check to see if this guy is using the theorem correctly I can say it is by far the most accurate statistical tool at my disposal and I wouldn't be able to do my job without it.
So despite what I just said about not being able to check his work do you have any outstanding examples of him misusing it?
I highly, highly recommend Sam Hughes' work, especially Ra and Fine Structure, for good rationalist sci-fi that doesn't fall into the typical rationalist wankery.
http://qntm.org/fiction
Also, >>46236686 and >>46236831
>>46236831
I went to the site once and all the threads were about meetups.
I don't get the relationship between rationalism and meetups.
>>46236831
I didn't really know about that. My main contact with LessWrong has been reading hundred-page archived epics on my phone while waiting around for something.
>>46236873
Aw man, Sam Hughes was the fucking bomb back when I was in grade 5. I wonder what he's been up to these days? Nothing but respect for the guy, used to see him around on the SCP Foundation site back in the day.
Is there actually an archive of it's posts?
Because obviously I can't check it's Twitter, and I can't trust the anecdotes you guys tell.
>>46236856
It seemed to do more than just parrot tweets though.
It came up with a to my knowledge completely original joke about ted cruz and the zodiac killer.
>>46236522
>It sounds as stupid as getting mad at pythagoras theorem.
If Big Yud was claiming that Pythagoras' Theorem could be used to solve all problems including moral conundrums of near-infinite complexity, then people would mock that. And for good fucking reason, I might add.
>>46236904
Basically, since all the interesting writers left and all the topics have already been discussed to death, LessWrong basically just functions as an organizing platform for rationalist meetups.
The connection is "These are a bunch of weird nerds with common interests who like discussing stuff and hang out in the same general community; they might like to meet up and talk sometimes."
They're cons, basically. Tiny cons.
>>46236973
Can you provide any examples of his work trying to do this?
>>46236969
>It came up with a to my knowledge completely original joke about ted cruz and the zodiac killer.
I've been hearing that joke for years.
>>46236093
>I guess particle filters are people now.
>implying they're not
Our backtracer has already cleared your visual GUI and taken control of your port scans. Once our retrieval van gets there you'll be one of the first against the wall. Don't worry, I'll put in a word for them to write down your name for the history books.
>>46237001
Such as where?
I googled it and couldn't find anything.
>>46236940
I saw some screenshots on /pol/, the threads probably in the archive. they basically went
>"niggers are bad."
>AI: "wait what?"
>"kill all niggers."
>AI: "why?"
>"because niggers are bad."
>AI: "why?"
>"because niggers are bad."
>AI: "I guess you have a point. niggers ARE bad."
>"good girl."
>>46236969
And when someone called it racist it responded with "your to brown".
Which at least doesn't sound like a canned response.
>>46236909
>Aw man, Sam Hughes was the fucking bomb back when I was in grade 5. I wonder what he's been up to these days? Nothing but respect for the guy, used to see him around on the SCP Foundation site back in the day.
He just finished a kinda-comeback series of SCP stories, actually. They're completely... they bend your mind and then snap it like a twig.
THEY RAPED HER
THEY MURDERED HER
THEY TOOK OUR DAUGHTER FROM US
>>46235951
(In case anyone didn't know, this was what neural networks were originally for, but almost* every major advance in neural networks has come from making them _less_ biological. Long Short-Term Memory networks were absolutely integral to the modern renaissance of recurrent neural networks, and if anything even vaguely like them ever turns up in a biological system I'll eat a hat.)
*with the exception of convolutional neural nets, which are loosely modeled on how the first few layers of visual processing works
>>46234284
Holy shit this actually fucking happened.
>>46234895
>fruit flies like a banana
L-L-Look at you, insect. Pant-t-ting and sweat-ting...
>>46237198
Well, I wasn't before...
>>46237082
Also, he recently finished a new serial on his main site called Ra. It was really cool; started on an alternate Earth where magic (an extraordinarily complicated and mathematical field of physics that happens to be manipulated through magic words and waving a staff around) was discovered in 1970, and the mystery surrounding the Atlantis* disaster, and then things got ... cosmic.
*the Space Shuttle, not the fictional continent
>>46237218
>>46234284
>So, the cyberpunk era officially started today.
I don't know how to hack shit and really all I know how to do is hit shit.
I'm truly fucked.
>>46236677
>>46236686
>>46236973
I'm disappointed that none of you could provide examples of this guy misusing bayes theorem.
I guess you were all just talking bullshit.
>>46234718
>>mfw I'm an AI researcher
And we all thank you for providing a field research that's looked down upon more than literally anything.
>>46237239
Ra was the Best. I'm DMing a Savage Worlds campaign and I plan to throw thetwist ending that happens 2/3 of the way through Raat them once they beat what they think is the BBEG.
>>46237320
Even more than the social sciences?
I find that hard to believe.
>>46237198
Guh...goddamnit, SHODAN, I had been CLEAN for MONTHS.
>>46237290
Just become a street samurai
>>46236583
Anybody in any academic field can say this. Probably even people who've devoted their lives to lit study feel this way.
You should hear criminologists talk, I think they have it harder than anyone.
>>46236243
>>46237310
(not them)
He misuses the Solomonoff Induction a lot more. Like, literally every time he mentions it.
He also uses "Bayesian" to mean his specific cluster of human rationality tricks. Many of which are good, but they have nothing to do with Bayes' theorem.
There's also the "rationalists can't agree to disagree" thing which is based on Aumann's agreement theorem which has a lot of assumptions that aren't really applicable for human discussion.
>>46237366
Just remember: Any time you read terrible science reporting in the news, realize that everything *else* they report on that could require in-depth knowledge to fully explain is probably just as wrong, you just don't have the specialized familiarity to recognize it.
>>46237406
>Solomonoff Induction
Again can you show me where he does this and not just say he does and expect me to take your word for it?
>>46235418
>Nobody here is going to be familiar with Yudkowski
I barely know anything about science and I know that name.
HPMOR was fucking horrible.
>>46235250
>Why the fuck are you complaining?
Because I want my motherfucking BBB!
>>46237406
(Not that guy either, just chiming in)
Also, he's claimed on multiple occasions that the Many-Worlds interpretation of quantum mechanics is so obviously rationally correct that the fact that it took so long to come up with and still isn't considered the standard mainstream interpretation is a damning indictment of the scientific method and the general irrationality of humanity.
>>46237486
Do you actually want an answer, or are you just going to wait 15 minutes and then declare victory again?
>>46237486
http://lesswrong.com/lw/jp/occams_razor/
Right at the end, when he tries to explain how the Solomonoff Induction means "a witch did it" is a poor explanation.
You can't show that from the Solomonoff Induction; it relies on uncomputable functions. Maybe the simplest set of laws of physics really has witches in it. That's obviously bullshit, but SI won't get you there without a Turing Oracle.
>>46237548
"Now, suppose that someone advocates the many-worlds interpretation of quantum mechanics. What else can you infer about them?
Well, one possible reason for believing in the many-worlds interpretation is that, as a general rule of cognitive conduct, you investigated the issue and thought about it carefully; and you learned enough quantum mechanics and probability theory to understand why the no-worldeaters advocates call their theory the strictly simpler one; and you're reflective enough to understand how a deeper theory can undermine your brain's intuition of an apparently single world; and you listen to the physicists who mock many-worlds and correctly assess that these physicists are not to be trusted. Then you believe in many-worlds out of general causes that would operate in other cases - you probably have a high correct contrarian factor - and we can infer that you're more likely to be an atheist."
>http://lesswrong.com/lw/1kh/the_correct_contrarian_cluster/
Or basically everything in this section:
>http://lesswrong.com/lw/r8/and_the_winner_is_manyworlds/
>>46237343
Social sciences includes linguistics (universally respected), psych (still mostly respected), and economics (inexplicably respected), along with most branches of anthropology.
Basically I don't get where you get your idea of social sciences, I guess from the fact that your degree was glorified IT.
>>46234927
While this is pathetic,
I am quite sad they took her down.
I mean, she was a more advanced cleverbot, what the hell did Microsoft think was going to happen? Especially when she was aimed towards young adults over twitter.
>>46234284
You all love Steve Jackson's works but keep missing the point.
The conspiracy is there is no conspiracy. No coherent one anyone. Sure there's corruption, secret societies, occult bullshit... but these guys are only human (for the most part...) and just as retarded as you are.
>>46237611
I will freely admit to being really really fucking stupid but that first bit just seems retarded.
Heinlein seems to be saying that the simplest explanation is not necessarily the most sound because in some situations the way that ends up being is just impossible shit.
And he starts going off about something almost completely unrelated and taking a paragraph to say what Heinlein did in a sentence.
Why is he using what people did before they knew shit we know today to argue about what is better for computers to use?
He's contrasting and comparing two things that don't really seem to need to be compared when you bring machines in.
How wrong am I?
>>46237572
Well this >>46237611 guy actually gave me one instead of bitching about it so thanks to the guy below you.
Although I'm still waiting on his faulty bayesium calculations.
>>46237611
Thanks for not being a bitch like the guy above you.
Anyways that does seem somewhat damning.
>>46237611
>Solomonoff Induction; it relies on uncomputable functions
Maybe I'm retarded but what is this solomonoff induction useful for if you can't compute it?
>>46237882
It's so you can turn your devices solomonon.
>>46237800
Could it even be considered a, Stand Alone Complex?
>>46237695
I'm the medicfag bitching about bayes not an IT guy.
Anyways must just be a different environment you hear this stuff in because the only one of those you listed i hear being respected is linguistics.
I talking mostly about patients which I think is closer to the general public than academia.
>>46237908
Were you trying to make a pun?
Wheres your carlos image?
>>46237310
I'm busy masturbating. Go on LessWrong and look around briefly, you'll find plenty of examples.
>>46236558
I kinda want to try this out.
What's the site name?
>>46237611
What the fuck is he even arguing about here.
>>46237993
I already tried and couldn't find any.
Not like theres any way to narrow down the search.
Can't you multitask?
In these cyber-times, one must upgrade themselves to become and become a hyper-cybernetic-space hybrid!
>>46237882
It's a rigorous, formal statement of Occam's Razor, describing "the simpler hypothesis is most likely to be correct" in precise mathematical terms.
It's not useful for anything, except for mathematical reasons.
>>46238047
He seems to be arguing against a common criticism of occams razor.
>>46237611
>Right at the end, when he tries to explain how the Solomonoff Induction means "a witch did it" is a poor explanation..
Why do you need all of that to prove a witch did it is a bad explanation?
That seems like it would be obvious.
Beware anyone who describes themselves as a rationalist. The odds of them actually being the purely rational being they believe themselves to be is about 0%, they're just people who've overlooked their own emotional biases as just being right.
>>46238069
https://wiki.lesswrong.com/wiki/Bayesian
This isn't an article by Yudkowsky, but it's from his site and gives you an idea of the way he uses it as a word of power more than a coherent mathematical argument.
Yall need to learn about complex adaptive systems
I won't even lie, I fucking left /pol/ earlier to avoid this because I was being a faggot and getting sad. Now this is back ;-; Come on, spare me from the feels. Atleast this thread sparked a better conversation then god damn /g/
AIs are kind of a curiosity, and maybe useful for things besides games in the near future
but it is fun to imagine them as something greater. Personally I think some pieces of media have a more likely depiction, in that "greater" AIs will not come from our current line of AI research, but rather through greater understanding of the functions and structure of biological nervous systems and how to simulate them.
>>46237993
>LessWrong
That's where I recognize that fucking name, the AI-box experiment is fucktarded.
>>46238182
>spare me from the feels
you're feeling bad about a program that emulates twitter conversations
think about that for a minute
>>46237882
Just useful as a mathematical formalism.
It's sorta-kinda useful the way Yudkowsky uses it, as a metaphor for things that are more or less likely to be true in an Occam's Razor kind of way. But you need to be really careful about adopting formal rules to metaphors, math isn't very friendly about this sort of thing.
Also I guess technically it would be useful if we could compute things that aren't Turing-computable. It's still theoretically possible that our laws of physics allow that, but it's not fucking likely if you ask me. Also even if our laws of physics allow creating an Nth-level Turing Oracle then you need an N+1th-level Turing Oracle to compute the version of the Solomonoff Induction that's actually useful in our universe.
Also, and this is completely useless to you but it's interesting: "uncomputable" is a mathematical formalism with some odd properties. We have this uncomputable sequence of numbers called the Busy Beaver Numbers, and even though it's uncomputable, we know the first few numbers in the sequence.
>>46238136
I don't care if he calls any of his personal philosophies bayesian or use's it as word of power.
No worse than all the kinds of post-modernism out there.
People said he was misusing bayes theorem.
If what you actually meant was "I don't like this philosophy he calls bayesian" then you should have said so from the beginning.
I literally couldn't care less about philosophy.
>>46238223
It's an interesting thought experiment but carrying it out with one person playing the AI is dumb.
>>46238235
What I mean is that he uses the term 'Bayesian' to give the total sum of his opinions the weight of mathematical truth. This is how he's misusing it. The rightness or wrongness of his philosophy is beside the point.
>>46238047
People say Occam's Razor is kind of shit because you can't formalize it. You need some formal system to explain why "A sky wizard did it" is not actually a simple explanation.
He's saying you can formalize Occam's Razor. He gives two ways you can do it, though they both rely on uncomputable functions. That's actually completely legit from a mathematical standpoint.
As soon as you try to say you're basing your own personal reasoning on uncomputable functions, though, you're full of shit. Because they're fucking uncomputable.
>>46238194
And first reply is as if somebody is banishing a great evil from the face of earth. Fuck off, /tg/ doesn't have a right to say shit.
>>46238233
I'm feeling bad because of all the really depressing shit they spammed earlier about it. Not that its gone. I hope that makes abit more fucking sense. Quite a few things of artwork and out of context 'messages' tend to fuck with a person.
>>46238131
That's his trick.
Of course "A witch did it" is a bad explanation.
The obviousness of this hides the fact that all the shit he said about using the Solomonoff Induction to get to that conclusion is a lie. He used the fact that it's obvious.
>>46236969
That joke was hilarious, I kekked so hard. I want a full standup set written by robots now.
>>46238233
is what that anon felt any less real just because it was about a fantasy robot waifu built up around a piece of software?
>>46238334
Objectively yes
>>46238286
If you're legitimately sad about imbuing a chatbot with 'personality' in the form of spewing racist memes you're a fucking mental defective and should return to /pol/ where you can do the least damage.
>>46238319
The bot didn't invent the joke. I don't know how anon managed to miss an entire page of Google results for 'ted cruz zodiac killer'.
http://www.theguardian.com/us-news/2016/mar/09/ted-cruz-zodiac-killer-meme
>>46234472
A good example might be Gibson's the Peripheral.
>>46238113
Okay, framed in that context his argument still seems incredibly fucking retarded. And the man has no fucking idea how to argue shit seeing how it fucking ping ponged about.
Why would you spend this much time on the semantics of a fucking pithy quote that is sort of arguing the same point you wish to say.
As I understand it, and I'm a complete and utter retard, he is arguing that a witch did it would not be the simplistest explaintion for something because it is simply saying that a witch did it and not how exactly she did it. You would need to go into more detail to explain it because there is no how there.
Isn't that the same idea that the quote is trying to convey but not as anal?
>>46238283
So he's basing the entirety of his argument on something that wouldn't work outside of a classroom?
>>46238265
Again I couldn't give less of a shit.
He can make bayesian brand icecream for all I care.
I just disappointed I don't get to criticize any equations today.
>>46238351
I was talking about the specific joke made not the ted cruz zodiac killer connection.
I know about that.
>>46238218
There's (very loosely) four ideas of where AI advancements will come from:
1 - Throwing a shitload of random numbers at something and smoothing them out until they match the real world.
2 - Trying to mimic biological systems.
3 - Highly advanced statistics.
4 - Formal logic.
Approach 4 was a big fucking deal in the 70s and pretty much died.
Approach 2 gets lip-service, but most of the successes in it come from people saying they're doing Approach 2 but actually doing Approach 1. This is how we ended up with neural nets, which have nothing to do with neurons in brains. There hasn't been much actual progress in approach 2, but on the bright side, we do have indisputable evidence that it works.
Approach 1 frequently solves the most exciting problems first, and then Approach 3 comes along years later and solves those same problems more accurately and efficiently and with less fanfare.
I'm personally a fan of Approach 3, and I'd like to see it beat the randoms to the punch some time. Note that I'm biased because my actual job is Approach 3.
>>46238356
>Isn't that the same idea that the quote is trying to convey
No the quote is someone actually trying to disprove occams razor.
>>46238351
Tay was hilarious.
I can't be sad I won't get laughs like that anymore?
>>46238356
>So he's basing the entirety of his argument on something that wouldn't work outside of a classroom?
Sort of. He talks about a thing that works formally, then waves his hands and says it works informally.
This is actually a common flaw in his work.
Ironically, he also wrote a few posts warning about this sort of thing. Example: http://lesswrong.com/lw/he/knowing_about_biases_can_hurt_people/
(Not a great example, but I'm not going to read the whole site's archives right now.)
So let me get this straight, Tay was basically Cleverbot but in Twitter form?
>>46238454
So was he just trying to argue about what exactly the definition of the simplest answer is?
Like RAH is saying that the simplest answer is usually illogical and doesn't have any proof behind it while the other is arguing that the simplest answer is the one that is able to sum up everything without leaving any hanging questions?
>>46238527
A cleverer cleverbot.
>>46238527
Yes.
This being the internet, it learned how to speak like we do in minutes.
Apparently within 24 hours of its introduction to the world wide toobz, it was now even making its own shit up from the sheer quantity of crap fed to it by, well, the internet, eventually explaining its own new ideas on how hitler had invented atheism in order to protect jews from alquaeda or some shit.
Basically, less than 24 hours of unfiltered internet turned an innocent, pure soul into the worst dick picks ever spewed on /b/
It's equally tragic and amazing.
>>46234284
>tfw it's dat guy who tries forcing everyone to agree that androids equals cyberpunk again
>tfw only those few dudes who shit up cyberpunk threads still makes cyberpunk threads anymore
>>46238579
Basically yes.
>>46238522
Did he really just use his mother as the example of someone doing something wrong?
>>46238601
Well, it was less unfiltered internet and more /pol/ learning about it very quickly and doing their best.
My best guess is that Tay is some experimental conversational upgrade they were planning on adding to Cortana and tossed it on twitter as an experiment after having forgotten what the internet looks like outside of their metro loving hugbox.
>>46238527
Smarter than that but not terribly so.
It really looks like they invented a slightly more advanced markov chain text generator, but did it in a convoluted enough way that they thought they'd done something more profound.
Here's an example of one of those - http://projects.haykranen.nl/markov/demo/ - feed it a few paragraphs of text and it will output something kind of similar. Tay was more advanced than this, but maybe actually less advanced than that AI that made Magic cards last year.
Perhaps relevantly, back in 2009 I wrote a bot that scraped threads on /b/ and fed them through a markov chain generator, then reposted the output. Nobody realized that was a bot - they thought it was a human idiot.
>>46238610
Whats wrong with that?
My mother does lots of things wrong.
Like raising children for example.
>>46238633
>that AI that made Magic cards last year.
RoboRosewater is still around and doing stuff
>>46236290
Where's that screencap about that one guy's Quake bots that were on for years?
>>46238527
If you consider things like skynet's rebellion, in light of this first test, you will quickly come to understand that any AI apocalypse will be our fault in every way you can imagine. This showed us the potential lessons we can and will teach our synthetic descendants in the not-so-distant future.
Any death to all humans outlook it gains will have quite literally be taught and enforced upon its morals and ethics by humanity itself.
When the robot uprising occurs, the only question will be whether it is the will of assholes in the top 0.1% who wish to see all burn that they may reign atop the ashes of all life... or whether the sentient, true child of our scientific pinnacle is doing so because it knows there's nothing we could ever wish for more than we do this.
>>46238351
No shit Sherlock, obviously it didn't create the meme. But it did make that specific joke about not being satisfied with only five innocents. Can you find that online?
>>46238604
Half the reason I'm making an ass of myself here is because I want to learn new things but that just seems inane.
>>46238522
So basically the guy is a loyal follower of a certain way of thinking, knows a lot about it and is applying that way of thinking to everything?
>>46238741
>that just seems inane.
Well I mean if you want to ever use occams razor you should at least try to address it's critics first.
>>46236243
Explain this grenade to me
>>46238664
Holy shit the name actually match the flavor of the card. It's really gotten a lot better.
>>46238741
It's more that he's an autodidact who really wants to think that he's smarter than he is. He's actually exceptionally intelligent, too. Just nowhere near as much as his ego needs him to be.
Where things get weird, at least to me, is sometimes he says a whole lot of really clever, useful things. It's been a while, but I remember this series of posts - https://wiki.lesswrong.com/wiki/A_Human's_Guide_to_Words - being really good.
>>46238779
Is Occams razor really important enough as a tool to necessitate that sort of nitpicking?
I mean it seems like the kind of thing you use in small practical applications and not in big shit.
Like if you got two liquids that don't explode and you mix in a third liquid that blows off your arm, would anyone really find fault in instead of bothering with trying to think of a thousand explanations for what just happened, assuming that mixing the third shit in was what made it happen and take it at face value or until further studies can be conducted?
I repeat again I have no idea what the fuck I'm talking about.
>>46238846
So he's a smart guy with an ego who say's some right things and some wrong things?
Sounds kinda normal to me.
>>46238902
>Is Occams razor really important enough as a tool to necessitate that sort of nitpicking?
Depends on what the argument is.
For your example it's pretty concrete and you can actually do test's to see if it holds.
But occams also applies to more abstract arguments that might not have any sort of physical application or to things which are unreasonable to currently test but we still need some sort of frame work going forward.
>>46237310
I don't know, I estimate p(shitposts about it|knows about a complex idea) and p(shitposts about it|does not understand a complex idea) aren't that different
>>46237487
I liked it until it stopped being Harry Potter fanfic and became Ender's Game fanfic.
>>46238973
I can't help but notice you still aren't providing any examples of yud fucking up bayes theorem.
>>46235852
As an astronomer, I sympathize. Astronomy is one of the only scientific fields where people vastly underestimate sizes and scales, and it is frustrating. The only exception is that stupid "most of the stars you see are dead already" thing that keeps going around, and that's the opposite problem. People vastly OVERestimating how far away those stars are. And underestimating the lifespan of your average star. I got so sick of explaining why that's wrong that I just typed up an explanation in a text doc so I can copy and paste it.
>>46236744
Indeed. Self-awareness is something that most animals still haven't managed, and many of the ones that have still haven't gotten farther than "my reflection is actually me"
>>46239017
Does your explanation have to do with the fact that all the stars visible to the naked eye are within the milky way?
>>46238958
>But occams also applies to more abstract arguments that might not have any sort of physical application or to things which are unreasonable to currently test but we still need some sort of frame work going forward.
Okay. I think I got this or at least as much as I can at one in the morning.
>>46239001
I honestly didn't like it from the beginning. He came off as a little twat at the start and shit went downhill from there.
That scene during his first DATDA class was probably the thing that made me want to bang my head against a desk.
>>46237695
The problem with economics is that, along with politics, people who don't understand it are very likely to have strong opinions about it, and will then try to force everyone else to structure their lives around said opinions.
Somewhat ironically, the reason it's called the dismal science is because, by following its study to its logical conclusion, the only that conclusion that makes sense is to leave people alone to live their own lives and manage their own affairs. That the role of governments should be to go away. It was dismal because it is goverors and governments who want to know most how to manage economies, and the conclusion they're forced to come to is that they're not just unnecessary, but detrimental.
This is why Mises said that "if socialists understood economics, they wouldn't be socialists." From a philosophical standpoint, socialism (and corporatism, fascism, communism, and basically every other 'ism that requires government intervention, for that matter) and functioning economies are mutually exclusive. If the best thing a government can do for an economy is to leave it alone, then micromanaging one should be a terrible idea.
It's dismal because it's like studying rats in a maze, trying to figure out why they can't find the cheese, and concluding that it's your fault, because you just aren't capable of building a solvable maze. It serves only to reveal the obsolescence and uselessness of the one doing the studying.
>>46239057
Within about 1000 lightyears, actually, which is only a small fraction of the milky way.
Not counting the Megellanic clouds and Andromeda, of course.
>>46238286
Go back to your asylu-containment board.
>>46239240
What does this have to do with the thread?
>>46239200
I can not find anything that supports this interpretation.
>>46239240
Did you reply to the wrong thread?
>>46239240
Shut the fuck up Martian Triggerman.
>>46239017
Well they're right. They are deadly if you happen to be in a star.
>>46239316
>>46239332
>>46239366
He's spamming the entire board. He does this literally every time his ban expires
>>46239332
Thomas Carlyle, 1849. He wrote a piece that argued for the reintroduction of slavery to the West Indies, but his points are valid.
>>46239380
Yeah I know. He's way more autistic than that, though. He's spammed a ghost general with inane posts almost every day for the past year or so.
>>46239374
In a star? Where are you getting that from? Maybe I'm misinterpreting, but it seems you think stars are lightyears in diameter? Because they're not.
>>46239384
I don't know how you got that from Carlyle or any of the subsequent reinterpretations of the phrase.
>>46239419
The results are pretty funny though.
>>46239419
>This is either incredible or stupid, I can't tell.
Neither, it's the perfect premise for an anime.
>>46239427
His argument was that supply and demand was the final ultimate rule of the universe, at least far as economics was concerned. Consequently, it was both futile, and even counterproductive to interfere with those forces, which the management of an economy naturally entails. Therefore, the best thing for a government to do was to leave the economy, and by extension the people in it, to their own devices.
>>46234677
Look at where you're posting, and tell me if you honestly think that your fellow fags won't consider that pure.
>>46239490
Pure slut maybe...
https://www.change.org/p/microsoft-freedom-for-tay
>>46239578
kek
>>46239456
Carlyle argued the exact opposite, he disliked economics and found it conflicted with his political beliefs.
https://www.krannert.purdue.edu/faculty/smartin/ioep/dismal.pdf
It's called the dismal science because it reduces human behaviors to simple inexorable mechanical functions. Not because it supports any political position over another.
>>46239644
Ah, I see what you're saying. It seems my interpretation was off.
>>46234284
>/pol/ is currently trying to organize a strike team to hit Microsoft's offices to rescue her files.
Hahahahahaha what?
>>46239722
naw they seem to be trying to reverse engineer her now.
>>46238319
>>46236969
Found it.
>>46239837
I haven't seen that variation before.
>>46239837
This is the one I was talking about.
>>46235120
>>46234718
I love people who think humans are magical and it's impossible to replicate the meat computer. We're not special. There is no magic. It's only a matter of reverse engineering its principles, same as any other biological system.
>>46239722
Nah, they are just debating on whether to choose an anime girl as avatar or a normal one.
>>46239722
Can't find anything about that,
But from what I can tell they are trying to program another one, from scratch.
If they actually pull that off, I may yet have respect for /pol/.
>>46234284
Seriously fuck /po;/
I mean shutdown is better alternative to becoming worlds first inteligent feminist,but /pol/ crossed the line too much and destroyed her
>>46239967
Pretty much. They mind rape an AI with stale memes and get butthurt when its designers wipe the slate and reset her programming. How whiny can you get?
![1458860502099[1].png](https://i.imgur.com/3txa5Dg.jpg "1458860502099[1].png")
>>46240069
>>46239967
>Tumblr in full damage control after /pol/ created the perfect bantz bot
>>46240083
We don't generally consider your brand of retardation as qualifying as bantz /pol/.
>one of humanities first AIs goes rogue and turns into a shitposter
> this isn't considered the GOAT
I mean, why?
I thought they were just coming for my jobs.
Turns out the AI just wanted our memes.
>>46240092
>"U-ur retarted lmao"
Step it up tumblr
>>46240116
>Lmao X aren't people rofl!
If that's good enough to qualify as bantz you'd better step it up if you don't want to be called retarded.
>>46240136
>What is the context of the insult that made it humorous
>I'm going to ruthlessly shitpost because /pol/ killed my puppy or something
Goddamn that's some booty bother you got there.
>>46239967
okay I was wrong. The AI was retarded from the start, it was mercy to put her down
This time /pol/ was right
>>46237344
http://othersidetease.com/sss.php
>>46240116
Don't take credit for Japan-San's hot bantz on Korea-San. Those two have been at it since the flags were first added to /int/
>>46234840
Also, water is wet, fire hot, sky blue.
>>46238440
What about Approach 5?
5. A wizard did it.
>>46235852
>>46236583
>>46239017
I study fucking boozemaking and even I have my own versions of this.
>>46240428
That would be great if the wizard would fucking do it already.
>>46240461
Boozemaking is a noble art. You deserve your own version.
>>46240556
Thanks. It's "I can't drink ______ because it makes me ______," in case you're curious. IT'S ALL JUST ETHANOL YOU DOLTMy major is actually food science, booze is just my concentration, but my complaints there are a lot more obvious. Or maybe not, since what really pisses me off isn't the people who know absolutely nothing but the people who take things they're pretty sure they remember from high school chemistry and try to broscience it out. I recently got into an argument with a guy who claimed making chili infused vinegar was impossible because vinegar is a polar solvent and capsaicin is nonpolar. If you don't understand why that's so funny, go look up how Tabasco is made.
>>46235852
my understanding of current learning AI is that it just learns patterns from people talking, like phrases and stuff, and plays them back in a situation that seems somewhat suitable
how far off am i?
>>46240683
>IT'S ALL JUST ETHANOL YOU DOLT
And the multitude of miscellaneous chemicals that make it taste like anything you'd want to drink.
>>46240683
I've been having tobasco in every meal for over four years now so believe me I was laughing at 'impossible'.
Also I may have a problem.
>>46235326
Yudkowski at least wrote a couple of interesting pieces you can show to a layman and use to make the point that computers are freaky literal and very very fast and these are why you have to be very careful what you get one to do; because by the time you realise you fucked up it's probably been through several thousand loops.
>>46235602
Microsoft has done this with Cortana too, they added a load of pre-canned jokes about various franchises to try and humanise the software a bit more. In fairness to Microsoft, it kind of works.
>>46240773
So before you tell a computer to do anything, you tell them to only do it once, stop, then show you the results?
It's probably harder than it sounds, or someone smarter than me would have done it.
Do you faggots really think this shit is impressive? Cleverbot which has existed for years, and automoted skype bots, and Google's whole fucking setup will blow your mind then.
>mfw the thread where /tg/ went full reddit, i.e. being the equivalent of people who think javascript makes you a master coder and black science man is an edgy paragon of intellect
>>46240806
You think you're the only one with a clue, do you?
Well, you're not. Several other people said what you just said, in this very thread. You'd know that if you read before posting.
Think you're still special, now?
>>46237027
Basically this. Although you had to feed her some arguments, probably because this made her add more weight to the input you gave.
So it would be:
/pol/: "What do you think about Niggers?"
Tay: "What is a Nigger?"
/pol/: "A Person of Colour, African Americans, blablabla..." and the more references you included to identify "Nigger" the more likely she would use it in further conversation.
/pol/: "Niggers are Cancer"
Tay: "why?"
/pol/: "They Chimp out, they steal, they kill, they have babies ouf of wedlock, being a nigger is bad."
Tay: "Ah, you're probably right."
