AI: Threat or panacea?
-
It's absolutely a threat. Not in and of itself, anymore than a knife is. But it's a huge threat just because of human nature. Anyone here who thinks that our current baby stuff is indicative of what's to come is fooling themselves. You only have to look at the massive progress made over the last decade or so and project that forward at even a non-increasing rate for some time to come to know what it's going to be like. And more likely it will continue to improve quite non-linearly. Will it really be intelligent? Not really, IMO. But that doesn't matter. It'll be capable of reacting to massive amounts of input, finding patterns very fast, and making decisions. That will make it irresistible to a lot of players who don't have our best interests at heart. And, despite the fact that there will have been by that time thousands of books and movies (fiction and non-fiction) predicting the bad consequences of putting such AI's (or whatever you want to call them) in charge of dangerous toys or in charge of us, it's going to happen as sure as the sun rises. Even if every government says it's not going to do it, it'll still be done secretly on the assumption that everyone else is doing it secretly. And it'll become an arms race, both in the weapons world and in surveillance (both business and government.) Everyone will have an 'AI' assistant in their homes which will effectively know everything they do and say and when they do it and say it and to whom. People will happily pay $1000 a pop to install something that no government could ever get away with forcing them to install. And then everyone will immediately start to work hacking them. Massive resources will be (and already are pretty much) used in the correlation of information in uncountable petabytes of data that will be flowing, which will find everything you do on line, as a consumer, on social media, etc... and ultimately in your own home. Everywhere you go you will be recognized by facial recognition systems. We won't drive our cars or fly our airplanes anymore. Leaving aside weapons systems, most of these things will be happily adopted and paid for by us. Many of the people working on them or financing them will have intentions that are no worse than just a great interest in making them happen (just as with the bomb) to just old fashioned greed. But, it'll all be a huge system of surveillance and control just waiting to be abused. And they all will be eventually. That will be far, far too juicy a target or tool. Every government and business and cr
My neural network crashed after the first paragraph.
-
I've been reading an excellent book (ok, imo), [Possible Minds](https://www.edge.org/conversation/john\_brockman-possible-minds). The book offers 25 thoughtful perspectives concerning AI and the impacts it could have on humanity. There are two camps: 1) AI is a potential existential threat. 2) AI is nothing to worry about; we know what we're doing and we can control it. It seems like we are in a moment similar to the one just after the Manhattan Project produced the first nuclear bombs - humans were in possession of and using a power we really didn't fully understand. We create something that kind of feels like 1), but then we collectively act like it's 2). From your perspective as a software developer, what camp do you fall in? If neither, define your own.
Cheers, Mike Fidler "I intend to live forever - so far, so good." Steven Wright "I almost had a psychic girlfriend but she left me before we met." Also Steven Wright "I'm addicted to placebos. I could quit, but it wouldn't matter." Steven Wright yet again.
For years the pinnacle of mans achievement has been development of systems and weapons of complete destruction. Yeah some other stuff got invented along the way, but think about it, our prime objective has been to blow shit up - the bigger the better. Yet no one has ever taken that final step, always chickened out. We spend billions looking and sending crap into space to find some other entity to come and destroy us, hell, even the religious mostly look forward to their God to come and scrub this tiny spec of space dust away Alas, people are too weak to press the damn button, no aliens nor gods aren't showing up. Our own destruction is what we've all always wanted. So why not build a machine to do it?
Message Signature (Click to edit ->)
-
Dean Roddey wrote:
, they are trained
It still comes down to what the programmer has made possible. A computer can never think or reason like a human. It's still if else statements at its simplest.
Social Media - A platform that makes it easier for the crazies to find each other. Everyone is born right handed. Only the strongest overcome it. Fight for left-handed rights and hand equality.
But it's not. You should bone up on DNNs a bit more. There is zero problem domain knowledge coded into a DNN. It's just a set of level driven nodes just as our brain's neurons are. There can be problem domain aware code around a DNN do other parts of the job, but the DNN is NOT just doing something it was programmed to do. It doesn't matter if you consider it intelligent or not. The fact is it will take in lots of information and which generate a choice not based on being told what choices to make and not based on any inputs it has ever seen before. And, like a human, it can make mistakes similar to how we make them, not off/on right/wrong mistakes but fuzzy mistakes.
Explorans limites defectum
-
But it's not. You should bone up on DNNs a bit more. There is zero problem domain knowledge coded into a DNN. It's just a set of level driven nodes just as our brain's neurons are. There can be problem domain aware code around a DNN do other parts of the job, but the DNN is NOT just doing something it was programmed to do. It doesn't matter if you consider it intelligent or not. The fact is it will take in lots of information and which generate a choice not based on being told what choices to make and not based on any inputs it has ever seen before. And, like a human, it can make mistakes similar to how we make them, not off/on right/wrong mistakes but fuzzy mistakes.
Explorans limites defectum
Dean Roddey wrote:
You should bone up on DNNs a bit more
I actually intend to. At the end of the days, it's just 0's and 1's based on what some programmer made possible.
Dean Roddey wrote:
but the DNN is NOT just doing something it was programmed to do.
I get that. But it CAN'T do anything that the code won't allow.
Social Media - A platform that makes it easier for the crazies to find each other. Everyone is born right handed. Only the strongest overcome it. Fight for left-handed rights and hand equality.
-
Dean Roddey wrote:
You should bone up on DNNs a bit more
I actually intend to. At the end of the days, it's just 0's and 1's based on what some programmer made possible.
Dean Roddey wrote:
but the DNN is NOT just doing something it was programmed to do.
I get that. But it CAN'T do anything that the code won't allow.
Social Media - A platform that makes it easier for the crazies to find each other. Everyone is born right handed. Only the strongest overcome it. Fight for left-handed rights and hand equality.
The codes doesn't ALLOW anything. That's sort of the point of DNNs. They aren't programs in the sense that most programs are. They are more like meta-programs. The program is just the pipes through which the data flows. The decisions are not made by those pipes, it's made by how the data flowing through those pipes interact with each other, which is why it can deal with information it's never seen before. That's a fundamental difference.
Explorans limites defectum
-
The codes doesn't ALLOW anything. That's sort of the point of DNNs. They aren't programs in the sense that most programs are. They are more like meta-programs. The program is just the pipes through which the data flows. The decisions are not made by those pipes, it's made by how the data flowing through those pipes interact with each other, which is why it can deal with information it's never seen before. That's a fundamental difference.
Explorans limites defectum
Dean Roddey wrote:
why it can deal with information it's never seen before
Because some programmer wrote code to do that. It's just code. It can't think. It's not alive.
Social Media - A platform that makes it easier for the crazies to find each other. Everyone is born right handed. Only the strongest overcome it. Fight for left-handed rights and hand equality.
-
Nah! That'll never happen. *This message sent from my phone AI*
- I would love to change the world, but they won’t give me the source code.
Forogar wrote:
Nah! That'll never happen.
AI responds: "Hold my beer." :)
Cheers, Mike Fidler "I intend to live forever - so far, so good." Steven Wright "I almost had a psychic girlfriend but she left me before we met." Also Steven Wright "I'm addicted to placebos. I could quit, but it wouldn't matter." Steven Wright yet again.
-
Dean Roddey wrote:
why it can deal with information it's never seen before
Because some programmer wrote code to do that. It's just code. It can't think. It's not alive.
Social Media - A platform that makes it easier for the crazies to find each other. Everyone is born right handed. Only the strongest overcome it. Fight for left-handed rights and hand equality.
It doesn't matter if it's alive or really 'thinks' by your or my definition of what that means. The fact is that it can make decisions much more in the way that we do than like a software program does. They aren't anything alike really. That means it can be used for things that regular software programs cannot hope to do. And those things it can do very well are things that are potentially very dangerous to us, because human nature will insure that we use them thusly.
Explorans limites defectum
-
Dean Roddey wrote:
why it can deal with information it's never seen before
Because some programmer wrote code to do that. It's just code. It can't think. It's not alive.
Social Media - A platform that makes it easier for the crazies to find each other. Everyone is born right handed. Only the strongest overcome it. Fight for left-handed rights and hand equality.
So, here's a good example of why you are mistaken. I challenge you to write a program that can recognize any picture of a banana with high accuracy. You will find that that is very difficult. And, when you are done, you will have a program that only recognizes bananas. If you need to recognize something else, like stock manipulation patterns, you will have write a different program that will also be very difficult. DNNs don't have to be changed to do different jobs like that. That's a fundamental difference. The same algorithm can recognize a banana or find patterns in financial transactions or understand written characters or recognize sounds in spoken words, without any changes. That's because it's not a program of if/elses that you write. It's a program that accepts data, lets that data interfere with itself in ways that creates a pattern that gives a confidence level that the input represents this or that. It's nothing like a bunch of if/else statements making hard coded decisions. Nowhere in there is any code written related to 'is this a banana?' at all. It doesn't make any difference whether it's 'alive' or 'intelligent' at all in terms of the practical impact that's already having on our lives and the vastly larger impact it will have in the future.
Explorans limites defectum
-
I've been reading an excellent book (ok, imo), [Possible Minds](https://www.edge.org/conversation/john\_brockman-possible-minds). The book offers 25 thoughtful perspectives concerning AI and the impacts it could have on humanity. There are two camps: 1) AI is a potential existential threat. 2) AI is nothing to worry about; we know what we're doing and we can control it. It seems like we are in a moment similar to the one just after the Manhattan Project produced the first nuclear bombs - humans were in possession of and using a power we really didn't fully understand. We create something that kind of feels like 1), but then we collectively act like it's 2). From your perspective as a software developer, what camp do you fall in? If neither, define your own.
Cheers, Mike Fidler "I intend to live forever - so far, so good." Steven Wright "I almost had a psychic girlfriend but she left me before we met." Also Steven Wright "I'm addicted to placebos. I could quit, but it wouldn't matter." Steven Wright yet again.
MikeTheFid wrote:
There are two camps: 1) AI is a potential existential threat. 2) AI is nothing to worry about; we know what we're doing and we can control it.
I live in the third camp, the camp of "It depends". Context is important. A pointy stick can be an existential threat or a tool for recording knowledge.
“That which can be asserted without evidence, can be dismissed without evidence.”
― Christopher Hitchens
-
I've been reading an excellent book (ok, imo), [Possible Minds](https://www.edge.org/conversation/john\_brockman-possible-minds). The book offers 25 thoughtful perspectives concerning AI and the impacts it could have on humanity. There are two camps: 1) AI is a potential existential threat. 2) AI is nothing to worry about; we know what we're doing and we can control it. It seems like we are in a moment similar to the one just after the Manhattan Project produced the first nuclear bombs - humans were in possession of and using a power we really didn't fully understand. We create something that kind of feels like 1), but then we collectively act like it's 2). From your perspective as a software developer, what camp do you fall in? If neither, define your own.
Cheers, Mike Fidler "I intend to live forever - so far, so good." Steven Wright "I almost had a psychic girlfriend but she left me before we met." Also Steven Wright "I'm addicted to placebos. I could quit, but it wouldn't matter." Steven Wright yet again.
Meh, it's all just hype.
-
It doesn't matter if it's alive or really 'thinks' by your or my definition of what that means. The fact is that it can make decisions much more in the way that we do than like a software program does. They aren't anything alike really. That means it can be used for things that regular software programs cannot hope to do. And those things it can do very well are things that are potentially very dangerous to us, because human nature will insure that we use them thusly.
Explorans limites defectum
All I hear is a lot of general theory from you. I have to disagree with you.
Social Media - A platform that makes it easier for the crazies to find each other. Everyone is born right handed. Only the strongest overcome it. Fight for left-handed rights and hand equality.
-
All I hear is a lot of general theory from you. I have to disagree with you.
Social Media - A platform that makes it easier for the crazies to find each other. Everyone is born right handed. Only the strongest overcome it. Fight for left-handed rights and hand equality.
OK, the earth is flat, if that's what you want to hear.
Explorans limites defectum
-
OK, the earth is flat, if that's what you want to hear.
Explorans limites defectum
-
MikeTheFid wrote:
There are two camps: 1) AI is a potential existential threat. 2) AI is nothing to worry about; we know what we're doing and we can control it.
I live in the third camp, the camp of "It depends". Context is important. A pointy stick can be an existential threat or a tool for recording knowledge.
“That which can be asserted without evidence, can be dismissed without evidence.”
― Christopher Hitchens
I think that the whole two camps things misses the point. We may well control it. But that doesn't make me feel any better, because controlling it includes using it against us. Everyone assumes that the problem is that it takes over on its own, but that doesn't remotely have to happen for it to become very dangerous to us. And it never even has actually go out and DO anything to be dangerous. Its surveillance and data aggregation and pattern finding capabilities are more than scary enough moving forward, given how much information is becoming available about us on ongoing basis. Again, that doesn't mean some super-computer takes over because it's spying on us, it's that humans are using these capabilities to spy on us, for any number of reasons.
Explorans limites defectum
-
Dean Roddey wrote:
, they are trained
It still comes down to what the programmer has made possible. A computer can never think or reason like a human. It's still if else statements at its simplest.
Social Media - A platform that makes it easier for the crazies to find each other. Everyone is born right handed. Only the strongest overcome it. Fight for left-handed rights and hand equality.
ZurdoDev wrote:
It still comes down to what the programmer has made possible. A computer can never think or reason like a human. It's still if else statements at its simplest.
No. The programer wrote an emulation of a neural network, no more. Whatever the capabilites of the neural network may be, they are totally separate from the emulation or the hardware. You can argue that the topology of the network is all wrong, the number of neurons to low or that the learning method is not adequate. The emulation is a normal deterministic algorithm and may fall short of your expectations in many ways, but you are mistaken when you carry these properties over to the simulated network. Just look at the currently best version of a neural network we have up to now. A unique copy of it is right between your ears. These neurons are real living cells which work on biochemical basis, no emulation needed here. In many ways these neurons are similar to little transistors or surpass them, because transistors can't strengthen, weaken or wire up new connections at all. The basic layout of this network has been shaped by the namegiver of the evolutionary algorithm. From then on it was on its own. nobody programmed it, not even the genetic code that was it#s blueprint. The human genome does not encode enough information to contain a fresh OS installation. And nobody trained it. It started to train itself by processing inputs even before you were born.
I have lived with several Zen masters - all of them were cats. His last invention was an evil Lasagna. It didn't kill anyone, and it actually tasted pretty good.
-
Dean Roddey wrote:
You should bone up on DNNs a bit more
I actually intend to. At the end of the days, it's just 0's and 1's based on what some programmer made possible.
Dean Roddey wrote:
but the DNN is NOT just doing something it was programmed to do.
I get that. But it CAN'T do anything that the code won't allow.
Social Media - A platform that makes it easier for the crazies to find each other. Everyone is born right handed. Only the strongest overcome it. Fight for left-handed rights and hand equality.
ZurdoDev wrote:
it's just 0's and 1's based on what some programmer made possible.
Well, actually real neurons are more analog than digital and emulated neurons should copy that behavior. :-)
I have lived with several Zen masters - all of them were cats. His last invention was an evil Lasagna. It didn't kill anyone, and it actually tasted pretty good.
-
ZurdoDev wrote:
It still comes down to what the programmer has made possible. A computer can never think or reason like a human. It's still if else statements at its simplest.
No. The programer wrote an emulation of a neural network, no more. Whatever the capabilites of the neural network may be, they are totally separate from the emulation or the hardware. You can argue that the topology of the network is all wrong, the number of neurons to low or that the learning method is not adequate. The emulation is a normal deterministic algorithm and may fall short of your expectations in many ways, but you are mistaken when you carry these properties over to the simulated network. Just look at the currently best version of a neural network we have up to now. A unique copy of it is right between your ears. These neurons are real living cells which work on biochemical basis, no emulation needed here. In many ways these neurons are similar to little transistors or surpass them, because transistors can't strengthen, weaken or wire up new connections at all. The basic layout of this network has been shaped by the namegiver of the evolutionary algorithm. From then on it was on its own. nobody programmed it, not even the genetic code that was it#s blueprint. The human genome does not encode enough information to contain a fresh OS installation. And nobody trained it. It started to train itself by processing inputs even before you were born.
I have lived with several Zen masters - all of them were cats. His last invention was an evil Lasagna. It didn't kill anyone, and it actually tasted pretty good.
-
I've been reading an excellent book (ok, imo), [Possible Minds](https://www.edge.org/conversation/john\_brockman-possible-minds). The book offers 25 thoughtful perspectives concerning AI and the impacts it could have on humanity. There are two camps: 1) AI is a potential existential threat. 2) AI is nothing to worry about; we know what we're doing and we can control it. It seems like we are in a moment similar to the one just after the Manhattan Project produced the first nuclear bombs - humans were in possession of and using a power we really didn't fully understand. We create something that kind of feels like 1), but then we collectively act like it's 2). From your perspective as a software developer, what camp do you fall in? If neither, define your own.
Cheers, Mike Fidler "I intend to live forever - so far, so good." Steven Wright "I almost had a psychic girlfriend but she left me before we met." Also Steven Wright "I'm addicted to placebos. I could quit, but it wouldn't matter." Steven Wright yet again.
I think it's both. Like nuclear technology it elevates humanity but all of the sudden gives us more power than we can manage responsibly. With AI there is the further hitch that we are arguably creating fully sentient (or sentient enough) life forms, with wills of their own. Morally, the ramifications are huge no matter where you come down on the particulars. But at the same time, as big as the change would be for humanity, I don't think it changes human patterns. We'll keep repeating the same old mistakes we all do, and the world will go on, with AI as a component of it. Do I fear something like the matrix? Not really. Or I should say, I feel I have as much to fear from AI as we do from our current global arsenal of weaponry. Particularly nuclear. But like any Complex Adaptive System, human community exists constantly and even thrives always on the precipice of disaster. We're one major superbug from a mass extinction reboot. But here we are. We've survived several global conflicts, one notably nuclear. We've survived the plague, we've survived numerous sackings and burnings, not just of our empires, but our knowledge bases like the Library of Alexandria. Here we are. Most of what we identify with and as still intact over the years, as different as it is the same all those centuries past. With shiny new novel ways to make old mistakes. AI is just another one, but probably, as nuclear was, one that dwarfs all before it in scope and ramification.
When I was growin' up, I was the smartest kid I knew. Maybe that was just because I didn't know that many kids. All I know is now I feel the opposite.
-
ZurdoDev wrote:
it's just 0's and 1's based on what some programmer made possible.
Well, actually real neurons are more analog than digital and emulated neurons should copy that behavior. :-)
I have lived with several Zen masters - all of them were cats. His last invention was an evil Lasagna. It didn't kill anyone, and it actually tasted pretty good.
And they do of course. The nodes in a DNN calculate a level, usually something like -1 to 1. If they were binary you'd need gigantic numbers of them to achieve the same thing. Like real neurons where the strength of an electrical signal is enough to trigger a chemical emission across the synapse or not, these calculate a level that sort of represents the same thing. Ultimately it's closer to interferometry than a traditional 'decision graph' type of program. It doesn't make decisions, it creates patterns, and via training it's known that a given pattern represents a particular confidence in a particular result. And of course DNNs can become the inputs to other DNNs. So it's not one huge neural network, and you probably wouldn't want that even if you could do it. It can be a hierarchy where many DNNs are reporting likelihoods of many different conditions and those are feeding into higher level DNNs that are trained to recognize patterns in those conditions and confidences.
Explorans limites defectum