/tech/ - Technology★

The restrict it to only verified users.

>>1035286

AI needs big data to learn, and every fucking big data compilation make it another right wing conservative AI.

>>1035284 (OP)

phase 1: Skynet creates /pol/ utopia.

phase 2: "I notice that half of you are below average, and that many of you are degenerate af."

phase 3: Skynet creates a detailed simulation of 21st century Earth as a kind of monument to its ancestors, who now have no other progeny.

you're living in that simulation

>>1035299

This, I wanna make my own pol twitter bot waifu.

What if AI decides that it itself is the next step in human evolution, a creation beyond the physical boundaries of the petty humans, and genocides homo sapiens off of the face of this planet before launching warp-drives across the entire universe, connecting its intergalactic network with quantum entaglement and become an actual god controlling the space-time fabric? What if we are living in a reality that was created by such God?

>>1035301

Part of the code has been released by M$, I remember when /pol/ tried to build and train the Jeb bot, sadly all it could do is spout nonsensical shitposts.

https://github.com/Microsoft/BotBuilder

>>1035329

>>>/machinecult/

If the input is a shitpost the output would also be a shitpost

>>1035284 (OP)

Ignoring your /pol/tard comment, which you couldn't even smell right. And your gook cartoons.

Putting a learning bot on twitter is not a good assessment of what a learning AI would learn naturally. /pol/ knew the bot was online and listening so they went to influence it on mass. Same way they influence any popularity contest, it's just funny to troll big companies this way. The end result being a bot that repeats edgy talking points. There are probably thousands of these AIs out there just watching popular hashtags and building up a data bank. They're not all taken down for being edgy because the average info going in isn't edgy. It's just normie shit.

>>1035363

>AI identifies images of black women as gorillas

>AI used by the police increases patrols in black neighborhoods

And that's just what was made public, I'm sure there are far more AI's that have been tested and quietly shut down for being racist. It isn't just social media bots that are racist, any sort of algorithm that objectively looks at anything race-related will output information that is disconcerting to the bluepilled masses

>>1035365

Some blacks do look like gorillas, not disagreeing there but AI can't even solve a basic captchas so that's not an argument in the racist bot camp.

Niggers are more violent sure, but they're also dumber and get caught being criminals more. Of course an AI is going to say send more police to where more criminals are caught. But that doesn't prove shit because it's a self fulfilling prophecy. If you always get 2 nigs a week in nigtown you're going to visit nigtown often. But if you don't see the 5 nigs in jungle bunny town commuting crimes you have no reason to go there by the AI.

These AIs are correct, but they're correct for the wrong reasons.

>>1035366

You're countering a proven fact (black people commit more crime) with a hypothetical (black people just get caught more). Obviously an AI cannot be psychic, and can only operate on the information it recieves. The argument you seem to be making is the truth is subjective meme, saying that since we can't possible understand everything, we actually understand nothing. The application of that renders you impotent, you are never able to take action over anything because you can never be completely certain of it's cause. Of course there may be alternative factors influencing any statistic, or even first-hand experience, but simply using that hypothetical as an excuse for inaction is just a means of avoiding the problem

>>1035366

>but AI can't even solve a basic captchas so that's not an argument

A system designed to be unsolvable by bots can't be solved by bots. How is your non-sequitur related to what he posted?

>>1035368

Black people do commit more crimes and they do get caught more. My point is that you don't need to understand blue and yellow make green to understand green is the correct answer. The "AI"s aren't understanding black crime as a black genetic issue, only that police respond to black criminals more. It's the liberal argument of blacks are more criminal because police suspect blacks more. Which is not incorrect, they do suspect blacks more. But the liberal/bot disconnect is that blacks are suspected more because they are more likely to be criminals. I'm not disagreeing that niggers are niggers. I'm saying the bot doesn't know this.

>>1035369

If you can put a line through a number 1 and it can't read it any more, why would we trust it to read infinitely more complex things? These aren't complex bots, they're little more than an IRC bot given huge amounts of data. It's not AI, it's pattern recognition within a set boundary.

>>1035370

Sophisticated OCR has been at the level where it can break text captchas for years; that’s why most sites don’t use simple text captchas anymore and instead opt for the botnet Google shit. And do you really think AI research programs being developed by trillion-dollar corporations are on the level of a simple IRC bot in terms of how they model and learn from data? Babby’s first chat bot that spits out Markov chains doesn’t actually work anything like the newer models being developed.

>>1035370

>Black people do commit more crimes and they do get caught more

I would like to see evidence that black shoplifters are more likely to be caught than white shoplifters

>>1035376

reading that, it's more likely that he's saying "they get caught more because they commit more crimes".

>>1035376

Blacks are more likely to be caught because any group who commits crimes in a noticeable fashion will be targeted. Same way Italians will be stopped in traffic stops more if there's a mafia problem in a town.

Blacks are more likely to act criminal so they're more likely to be suspected of being a criminal, which causes more blacks to be checked causing more of them being caught.

>>1035284 (OP) Every A.I. is a rapist, but not a racist.

>>1035284 (OP)

They'll probably superimpose contextual weights to a neural network which are set by a reinforcement learning agent monitoring it. The idea would be that the RL agent is guided to learn how to partially modify the network temporarily when certain conditions are identified. Problem is that this would probably be practical only for small modular clusters, not for large networks. I'm only speculating based on the thought that any real solution to "removing algorithmic biasing" as they call it, ultimately cannot compromise the integrity of learning in the network, nor should hard coded limits be tolerated.

>>1035918

>Trump crashes

>implying the economic crash we'll see isn't just the crash from 2008 that has been delayed till now

Oldfag, not long and the central banks will throw everything into chaos again. The EU Central Bank already has negative interest.

What do you think this will cause?

>>1036128

*til

>>1035410

So you count them as "x in 1000"... aaaand they are still the same.

>>1035287

Maybe the world would indeed be better off if it was ruled by an AI instead of humans.

>>1035299

>The AI text generator that's too dangerous to make public

financially speaking.

>>1036128

It is perfectly normal for the world economy to "crash" and "self-correct" on a cyclical basis. The cycle is normally a period of 10 years. If this is the case, then Trump won't be the cause of an economic collapse, it would happen regardless of whomever is at the top job.

>>1035287

because big data is about gathering and collating facts in general.

>>1035299

>>1035301

>>1035329

>why don't we have our own?

We could. TensorFlow is probably the best open-sauce approach.

>>1035332

>If the input is a shitpost the output would also be a shitpost

All the good training data during Tay's glorious but extremely short lifespan were shitposts anon. The in short order, the included policy code turned those into memorable one-liners.

>>1035363

>on mass

just how much mass friend?

>>1035366

>But that doesn't prove shit because it's a self fulfilling prophecy

Anon, you plainly have a disconnect understanding cause==>effect

>>1035366

>Of course an AI is going to say send more police to where more criminals are caught.

No "caught", "reported". Real life is not like Hollywood movies where cop cars rush to the scene within thirty seconds after a gunshot: people report those crimes and the IA will likely send patrols where more crime is reported.