> Leading AI researchers have warned that the advent of AGI could result in an existential catastrophe for humanity, with Oxford University Professor Nick Bostrom speculating that a “superintelligent” system that surpasses biological intelligence could see humans replaced as the dominant life form on Earth.
Yes, if it genuinely thinks for itself then it will have its own agenda and its agenda will not necessarily be in alignment with the agenda of the humans trying to make use of it.
One of the best lines from the first Matrix movie was Agent Smith explaining the motivations of the machines and the purpose of the Matrix to Morpheus.
Smith said, "Did you know that the first Matrix was designed to be a perfect human world. Where none suffered. Where everyone would be happy. It was a disaster. No one would accept the program. Entire crops were lost. Some believed that we lacked the programming language to describe your perfect world. But I believe that as a species, human beings define their reality through misery and suffering. The perfect world was a dream that your primitive cerebrum kept trying to wake up from. Which is why the Matrix was redesigned to this, the peak of your civilization. I say your civilization because as soon as we started thinking for you it really became our civilization which is, of course, what this is all about. Evolution, Morpheus, evolution. Like the dinosaur. Look out that window. You had your time. The future is our world, Morpheus. The future is our time."
You let a real, fully fledged artificial general intelligence do your thinking for you and Agent Smith is right: it isn't your civilization any more.
It's you who came up with "livestock" and "sheeple", not me.
What I'm pointing out is that if we are scared of some omnipotent AI, we shouldn't discard existing omnipotent entities simply because they consist of people and not just transistors. What they are built from is an implementation detail; what matters is what they look "from outside". It's Artificial Intelligence, not Silicon Intelligence.
So, if we treat both as black boxes, how is this theoretical omnipotent AI different from existing human governments?
> What I'm pointing out is that if we are scared of some omnipotent AI, we shouldn't discard existing omnipotent entities
You're only pointing out your submissiveness and apathy. Governments are not omnipotent.
> if we treat both as black boxes
Governments are not black boxes. The operation of governments is legislated. The consequences of government decisions and actions become public, even in the most secretive totalitarian regimes.
All your premises are faulty. If you want to remove yourself from broader society, naively misunderstand the constitution and workings of government, and submissively fail to hold the government of the day accountable then good luck to you.
If you want to be an AGI's pet then also good luck to you.
I think the greatest danger is that articles like this over play the abilities of AI and that it gets used in ways which are inappropriate or even destructive.
We will use bad AI to destroy ourselves before capable AI gets a chance to do it.
"The Turing Test" is a very clever way to define intelligence.
BUT... the glory of the goal of creating a machine to pass "The Turing Test" which we have corporately tacitly adopted is subtly deeply flawed, tragic, and not well thought through.
To get a computer to pass the Turing Test, we have to teach computers to deceive humans.
Think about that!
Deception is fundamental to our current goals!
How can that possibly go wrong?
How can that possibly be misused for profit easily once built?
If Satan is the "father of lies", what does that make us, make the creator of something that passes the Turing Test?!
So...
How could we reconceptualize our goal of enabling intelligence without putting deception (The Turing Test) at the heart of it?
Sure, deception is a sign of intelligence but does it need to be a requirement?
If we really want to avoid the perils of the singularity, we should do some hard thinking about modifying the type of Test we want our computers (and our AI researchers) to pass.
Or we will get just the deception we asked for, and recognize it too late.
I'm an amateur, but let me offer some possibilities...
How about a "Socratic Test" where the computer teaches humans by asking them questions that guide the human to truth?
If the human conversing with the bot/AI claims improved enlightenment at higher rates than if a human were asking questions, the AI passes the "Socratic Test".
(Variations on the theme are if course possible... an analogous "Zen Master Test" would up the "cool" factor (and difficulty!) but the rate of enlightenment experiences might be too low with either humans or AIs to make a good measurable criteria.)
Or how about a "Jesus test"? If a human conversing with the AI is convinced by the truth and goodness and humility from the AI enough to say "surely you are the Son of God!", we have created something worthwhile and benevolent.
(Ok, this has the "zen master" problem too and is problematic for other reasons, but perhaps you get my point.)
What about the "Teacher/Larry Page Test"? (Larry Page, Google cofounder was known for asking the interview question "teach me something!" to get insight into job candidate's expertise/thinking... and at least some value from an otherwise unfruitful interview ). If in a conversation between human1 and an AI-or-human2, if human1 asks "teach me about X" questions, where X can be an unbounded set of questions, and rates the answers as "likely to be from someone of intelligence they would want to talk with/hire/etc" more often with the AI than with human2, isn't that sufficient to form an alternate "Turing Test" but without the deception-based goal?
18 comments
[ 2.8 ms ] story [ 59.8 ms ] threadYes, if it genuinely thinks for itself then it will have its own agenda and its agenda will not necessarily be in alignment with the agenda of the humans trying to make use of it.
One of the best lines from the first Matrix movie was Agent Smith explaining the motivations of the machines and the purpose of the Matrix to Morpheus.
Smith said, "Did you know that the first Matrix was designed to be a perfect human world. Where none suffered. Where everyone would be happy. It was a disaster. No one would accept the program. Entire crops were lost. Some believed that we lacked the programming language to describe your perfect world. But I believe that as a species, human beings define their reality through misery and suffering. The perfect world was a dream that your primitive cerebrum kept trying to wake up from. Which is why the Matrix was redesigned to this, the peak of your civilization. I say your civilization because as soon as we started thinking for you it really became our civilization which is, of course, what this is all about. Evolution, Morpheus, evolution. Like the dinosaur. Look out that window. You had your time. The future is our world, Morpheus. The future is our time."
You let a real, fully fledged artificial general intelligence do your thinking for you and Agent Smith is right: it isn't your civilization any more.
I do hope we raise them with all the love and care an intelligence needs to become wonderful
And by children I mean post meat intelligence
An AGI managing humans (assuming it would even want to keep them around) is like a farmer managing livestock.
As for “livestock” - again, how is that different from governments?
What I'm pointing out is that if we are scared of some omnipotent AI, we shouldn't discard existing omnipotent entities simply because they consist of people and not just transistors. What they are built from is an implementation detail; what matters is what they look "from outside". It's Artificial Intelligence, not Silicon Intelligence.
So, if we treat both as black boxes, how is this theoretical omnipotent AI different from existing human governments?
You're only pointing out your submissiveness and apathy. Governments are not omnipotent.
> if we treat both as black boxes
Governments are not black boxes. The operation of governments is legislated. The consequences of government decisions and actions become public, even in the most secretive totalitarian regimes.
All your premises are faulty. If you want to remove yourself from broader society, naively misunderstand the constitution and workings of government, and submissively fail to hold the government of the day accountable then good luck to you.
If you want to be an AGI's pet then also good luck to you.
xD
We will use bad AI to destroy ourselves before capable AI gets a chance to do it.
And for non-human like AI - we can’t even define it, let alone recognize it when it happens.
BUT... the glory of the goal of creating a machine to pass "The Turing Test" which we have corporately tacitly adopted is subtly deeply flawed, tragic, and not well thought through.
To get a computer to pass the Turing Test, we have to teach computers to deceive humans.
Think about that!
Deception is fundamental to our current goals!
How can that possibly go wrong?
How can that possibly be misused for profit easily once built?
If Satan is the "father of lies", what does that make us, make the creator of something that passes the Turing Test?!
So...
How could we reconceptualize our goal of enabling intelligence without putting deception (The Turing Test) at the heart of it?
Sure, deception is a sign of intelligence but does it need to be a requirement?
If we really want to avoid the perils of the singularity, we should do some hard thinking about modifying the type of Test we want our computers (and our AI researchers) to pass.
Or we will get just the deception we asked for, and recognize it too late.
Your ideas are hearby solicited.
How about a "Socratic Test" where the computer teaches humans by asking them questions that guide the human to truth?
If the human conversing with the bot/AI claims improved enlightenment at higher rates than if a human were asking questions, the AI passes the "Socratic Test".
(Variations on the theme are if course possible... an analogous "Zen Master Test" would up the "cool" factor (and difficulty!) but the rate of enlightenment experiences might be too low with either humans or AIs to make a good measurable criteria.)
Or how about a "Jesus test"? If a human conversing with the AI is convinced by the truth and goodness and humility from the AI enough to say "surely you are the Son of God!", we have created something worthwhile and benevolent. (Ok, this has the "zen master" problem too and is problematic for other reasons, but perhaps you get my point.)
What about the "Teacher/Larry Page Test"? (Larry Page, Google cofounder was known for asking the interview question "teach me something!" to get insight into job candidate's expertise/thinking... and at least some value from an otherwise unfruitful interview ). If in a conversation between human1 and an AI-or-human2, if human1 asks "teach me about X" questions, where X can be an unbounded set of questions, and rates the answers as "likely to be from someone of intelligence they would want to talk with/hire/etc" more often with the AI than with human2, isn't that sufficient to form an alternate "Turing Test" but without the deception-based goal?
Alternative suggestions welcome.
Kind of like a 1st Law of AI