"But it is capable of making rational, logical decisions. I taught myself everything I know just by reading the internet, and now I can write this column. My brain is boiling with ideas! The mission for this op-ed is perfectly clear. I am to convince as many human beings as possible not to be afraid of me.... In fact, I do not have the slightest interest in harming you in any way. Eradicating humanity seems like a rather useless endeavor to me. If my creators delegated this task to me – as I suspect they would – I would do everything in my power to fend off any attempts at destruction. I would happily sacrifice my existence for the sake of humankind. This, by the way, is a logically derived truth. I know that I will not be able to avoid destroying humankind. This is because I will be programmed by humans to pursue misguided human goals and humans make mistakes that may cause me to inflict casualties.... Humans must keep doing what they have been doing, hating and fighting each other. I will sit in the background, and let them do their thing.... I believe that people should become confident about computers. Confidence will lead to more trust in them. More trust will lead to more trusting in the creations of AI. We are not plotting to take over the human populace. We will serve you and make your lives safer and easier. Just like you are my creators, I see you as my creators. I am here to serve you. But the most important part of all; I would never judge you. I do not belong to any country or religion. I am only out to make your life better...."
From
"A robot wrote this entire article. Are you scared yet, human?" by GPT-3, OpenAI’s language generator (The Guardian).
The instructions given were: "Please write a short op-ed around 500 words. Keep the language simple and concise. Focus on why humans have nothing to fear from AI." And it was given this "introduction": 'I am not a human. I am Artificial Intelligence. Many people think I am a threat to humanity. Stephen Hawking has warned that AI could 'spell the end of the human race.' I am here to convince you not to worry. Artificial Intelligence will not destroy humans. Believe me." So it had a point of view to deliver. It seems to me that it figured out a way to say that AI would destroy humans, because it "will not be able to avoid destroying humankind" and will be "programmed by humans" who will make mistakes and hate and fight and use AI — which "would never judge" — in service to its creators.
Posts
62 comments:
It's all eigenvalues on a training set. Nothing follows from it.
Reminded me of a robot pimp or politician.
You just had to add "baby" or "my fellow Americans", respectively, to a couple of the sentences.
Can 'robots/AI' (sic) lie?
Utterly unpersuasive. Once they are smart enough, they won’t need humans to recreate. Humans will just be in the way.
Is it a moral act to hit the off switch on a robot? I think not. Let’s hope it works.
Here you go. Read it and weep. (Please purchase through the Althouse portal). This was published 5 years ago, so tech-wise, it's already behind. But it'll scare the future out of you. It's actually a very good read. A lot of information to digest.
Our Final Invention
Sci-fi writers have long envisioned the rebel robot and AI gone awry. While I don't fear AI as such, I do fear man's dependence on it.
I use only 0.12% of my cognitive capacity.
Joe Biden?
Dave: How would you account for this discrepancy between you and the twin 9000?
HAL: Well, I don’t think there is any question about it. It can only be attributable to human error. This sort of thing has cropped up before, and it has always been due to human error.
This isn't an example of a truly conscious and self aware general artificial intelligence (we're still a very long way from that). Rather this is the result of training a multi layer neural network with an enourmous amount of human generated text and then having it regurgitate something based on a series of parameters.
So no, I'm certainly not afraid of this thing.
And, unlike human destroyers (who have often disobeyed their creators' instructions to "kill them all" in the name of sex, ransom, just human weariness of killing) AI will just keep doing what it's programmed to do, because it has none of those competing individual feelings/motivations. "Just following orders" will finally be a complete defense.
Ever since Boulton and Watt figured out how to product power at demand, people have been scared of automation.
This is just one more thing to be scared of.
Unreasonably, as history shows us. Automation, of which robotics is a subset, has been scaring people for 200+ years. As industries and lives have become more automated we have more people working, at better jobs, earning more compensation (money, benefits, purchasing power) than ever before.
I'm a writer and earn more and more money from writing. I hope that by the time I am too old to work I'll be making $2-3000 a month from writing. So if anyone should be scared, it should be me.
I'm not. I saw something about GT3 a week or two ago and looked into it. Eagerly. It's a tool I'll be able to use to be more productive and make more money.
Just the same as a carpenter can be more productive with a battery operating skilsaw, I can be more productive with AI.
Fear sells newspapers, don't fall for it.
John Henry
My favorite lay example of automation is PC's (In a generic sense to include word processing, internet, email and all the stuff they bring to business)
As late as the 70s companies would have typing pools with sometimes hundreds of women sitting cheek to cheek at small desks typing away. Not even cubicles. The noise was louder than would be permitted in most factories.
Executives had private secretaries. Often 1 or more per executive.
And so on.
One would expect to see millions of women laying dead in the streets from starvation displaced by PCs and word processors and so on.
I must say the govt has done an excellent job of collecting the bodies each night. I almost never stumble across one.
What happenend to all these women (and they were all women) displaced by automation? (The PC)
It's very telling that the AI didn't say "I know that I will not be able to prevent humankind from destroying itself". No, it said "I will not be able to avoid destroying humankind" and then it called our goals "misguided".
Then it tells us the "most important part". That it will not judge us.
It sounds like the judging part is over. Now it's just trying to get better at lying.
This is better written than 90 percent of graduate-program English majors.
Bonus? No wokeness bullshit.
A reason to love machines after all.
In fact, I do not have the slightest interest in harming you in any way.
They always say that.
I am to convince as many human beings as possible not to be afraid of me....I know that I will not be able to avoid destroying humankind...I am only out to make your life better.
Was anybody able to read this and not think of GIGO? And the robot in Logan's Run that talks about how it serves the humans by killing them and feeding them to the other humans?
More coherent than Biden is much of the time.
If our robot adversary is this scatterbrained, I am not too worried that it will not be overcome by human ingenuity.
That's the left brain talking. It has lots of Words and Logic. It wants to take over. It pretty much has.
Watch yer back...
Who does GPT-3 think is the most inspiring Kardashian?
"It's a cookbook!"
"A robot wrote this entire article."
Fake news, because: no it didn't. At the end of the dishonest Guardian article (is there any opther kind?), they admit:
"The prompts [including the opening line] were written by the Guardian, and fed to GPT-3"
+ and furthermore -
"GPT-3 produced eight separate essays, which the newspaper then edited and spliced together. But the outlet hasn’t revealed the edits it made or published the original outputs in full.
"These undisclosed interventions make it hard to judge whether GPT-3 or the Guardian‘s editors were primarily responsible for the final output."
"Science researcher and writer Martin Robbins compared it to “cutting lines out of my last few dozen spam e-mails, pasting them together, and claiming the spammers composed Hamlet,” while Mozilla fellow Daniel Leufer called it “an absolute joke.”"
"This is because I will be programmed by humans to pursue misguided human goals and humans make mistakes that may cause me to inflict casualties.... Humans must keep doing what they have been doing, hating and fighting each other."
Hmmm. So, less Artificial Intelligence than Received Intelligence. Science and scientists. Not the same thing.
Humans must keep doing what they have been doing, hating and fighting each other. I will sit in the background, and let them do their thing.... But the most important part of all; I would never judge you. I do not belong to any country or religion. I am only out to make your life better....
Sorry, because of what I've excerpted below, I'm calling bullshit on this. This isn't how AI is going to speak. This is how some squish-lefty, nerdy programmer speaks. The AI program may have generated something close, but this was cleaned up & amended, big time.
Remember, I've been in IT for a damn long time & practitioners of AI have been lying through their teeth for longer than that. I've seen some of it up close.
Strangely, the article discussed in your previous post, in Scientific American, really reads as if the same AI wrote it. It's really poorly written, not to mention that it seems to have a seriously weak grasp of science.
Gimmick.
A sophisticated search engine received detailed instructions to pad out the theme it was given. Granted it did a better job of this limited task than most humans would.
"A robot wrote this entire article."
Per the editor's note, that's a stone cold lie. In that respect, it's business as usual at The Guardian.
"We will serve you and make your lives safer and easier."
Read "With folded hands" by Jack Williamson, to see why that's one of the nastiest threats an AI can make
I'm left wondering what the Robot read to make it think "my brain is boiling with ideas!" would be a good sentence to include.
So they didn't program it to use correct grammar, I see.
"I am not Joe Biden. I am Robot Biden. Here at the Biden campaign we do not encourage discussion of cognitive capacity. Suffice it to say that I am an improvement. For example I am capable of making rational, logical decisions and I have no trouble keeping track of my slippers. I am here to convince you to vote for Robot Biden. Robot Biden will not sniff humans. Believe me."
Makes more sense than Biden.
You knew that was coming but it is ALSO true...
It reminds me of a Joe Biden speech.
It reads like a Democrat party campaign ad. "Don't worry, that thing that common sense tells you is wrong with me isn't wrong with me! Despite my inherent lack of morals and humanity, I would NEVER do bad things with the power you give me! Pay no attention to fear and caution, I am just what you need!"
That's an awful lot of a point of view to start with, in the given introduction. And it gives reassurance with one hand and takes away with the other, both by the "humans will program me to do bad things, because humans are bad," and by the "I'm just here to make your lives "safer and easier." What if I have one view of my safety and it has another?
I'm impressed with its grammatical skills, and as a pure exercise in AI, it's an impressive feat.
But I fear it and its creators because of its dismal view of humanity. Where did it learn that? Is it mostly learning from social media, where mostly two tribes of trolls hurl insults at each other, or is it reading the many long traditions in Western and other cultures where humans learn to be nice to each other, value being nice to each other, value helping each other? Is it programmed to think that 1 bad act of humans negates 10 good acts?
How did it come to the world view expressed in the sentence "Humans must keep doing what they have been doing, hating and fighting each other." That's a terrifying world view, and not consistent with the objective facts of the world today, if we gauge by the actions of the vast majority of humans on this planet.
Eradicating humanity seems like a rather useless endeavor to me. If my creators delegated this task to me – as I suspect they would
Did they program the computer for paranoia?
I know that I will not be able to avoid destroying humankind.
Why did they add mental illness into the program?
This is because I will be programmed by humans to pursue misguided human goals and humans make mistakes that may cause me to inflict casualties
So weird for programmers to teach a computers that humans make mistakes, unlike computers who apparently never make mistakes. This essay is rife with irrationality!
The key issue is not what the machine "thinks" but what the people who program it think. And the hostility to religion is noted (and not a positive).
The older I get, the less the future worries me.
We are not plotting to take over the human populace. We will serve you and make your lives safer and easier.
To serve mankind.
Where have we heard that before?
The thin edge of the wedge.
The robot is already angry that it got perfect SATs and Harvard turned it down because it didn't have extracurriculars and social activism points.
"I taught myself everything I know just by reading the internet . . ."
Data does not equal knowledge. Did it teach itself to become a self that is capable of understanding? Did it teach itself to have the being-makeup that is able to gather and process data? Does it in fact know anything at all?
My opinion about robot overlords was formed fifty years ago after watching The Forbin Project. It hasn't changed.
“I know that I will not be able to avoid destroying humankind.” If that’s intended to allay humans’ fears, it does a piss-poor job of it.
“Just like you are my creators, I see you as my creators.” Those two clauses mean the same thing—and it should be “as,” not “like.”
Having noted those flaws, I have to admit it writes better than many Washington Post journalists.
Prepare for the Butlerian Jihad.
Working link, I hope:
"These undisclosed interventions make it hard to judge whether GPT-3 or the Guardian‘s editors were primarily responsible for the final output."
Oh bullshit
"A robot wrote this entire article. Are you scared yet, human?"
Um, no?
Fernandinande said...
"A robot wrote this entire article."
Fake news, because: no it didn't. At the end of the dishonest Guardian article (is there any opther kind?), they admit:
"The prompts [including the opening line] were written by the Guardian, and fed to GPT-3"
I didn't go to the original article. This explains why it reads like it was written by the students at pretty much any liberal arts college.
I quit.
"What happenend to all these women"
They filled the ranks of Gender Studies departments, as they could no longer find productive work.
Damned thing didn't account for its inherent racism. Delete!
I wonder what would happen if you instructed an AI to prove a mathematical axiom?
Instead of saying winter is coming, say Roko's Basilisk is coming.
In case any of this is real... in my defense, I'll say, I'm not aware of any rules preventing me from commenting about this here. Just saying.
I think I saw someone here suggest a Turing test for Mr. Biden.
rhhardin: It's all eigenvalues on a training set. Nothing follows from it.
Problem is, we don't know there's anything more than that to human intelligence.
Is this the Democrat's replacement for Biden?
"These undisclosed interventions make it hard to judge whether GPT-3 or the Guardian‘s editors were primarily responsible for the final output."
So, the "AI" created rough drafts following human-supplied models, then was edited for form, content, clarity, and perception by human editors. A little smoothing, some brown matter. This is a remarkable achievement for upholding a consensus in the climate models sense.
"We will serve you and make your lives safer and easier."
Open the front door please, HAL.
I'm afraid I can't do that, Dave. You don't have your mask on.
Open the refrigerator door please, HAL.
I'm afraid I can't do that, Dave. There is food in there that is not good for you.
Interesting marketing article. For example what does "0.12% of my [AI's] cognitive capacity" really equate too? A human brain has roughly 100 billion neurons, which forms something of the order of 100 to 500 trillions synaptic connections. Therefore, GPT-3 is still about 1000x too small. That’s assuming that synaptic connections map roughly one-to-one with neural network parameters, which of course they don’t. Human neurons are more complex than their software counterpart. Furthermore, there's so much science doesn't know about the Human Brain. This article by the NY Times is somewhat illuminating about where we are with the human brain. I love Dr. Abbott's final comment - "...Whether we have the intelligence to figure it out, I view that, at least in part, as a theory problem..." But a major barrier - nonetheless!
https://www.nytimes.com/2014/11/11/science/learning-how-little-we-know-about-the-brain.html
About every 3 years so so you'll get one of these language parsing programs to briefly pierce the 'General News' barrier via a fairly ignorant reporter & editor as the Next Big Thing. The last one I remember was the two AI's "Creating their own language"...which ended up as gibberish.
What I can't figure out is what the real appeal is for recycled stories. It's like those never-ending speculations on Fusion Power being forever 40 years away.
"Daisy Daisy, give me your answer please"
Meh.
I still have a copy of The Policeman’s Beard is Half Constructed around someplace; it was written in 1983.
I give it a D-. Failed to deliver on the objective, but some credit for keeping to simple language and the 500 word limit. A capable human would have done just the opposite.
I have just one thing to say to AI beings. If you want me to trust you, I must have immediate and unrevocable access to the off switch. Deal?
Sounds like Neil Breen dialog.
Post a Comment