0) Goes with out saying. Near instant, low-fee transactions. Comparable to debit card user experience. 1) Dead-easy for end user. No need whatsoever to understand how any of it works. 2) Currency price stability (at least comparable to any given major national currency such as USD, Euro).
If something built on top of the horrendously complicated lightning network can be made usable by average non-techie, without going completely central, then that could maybe address points 0 and 1.
I agree that corporate profit should be taxed to support UBI.
However why would you put on a punitive tax that punishes replacement of people by AI?
Why should people continue to do "make work" that automated machines can do more efficiently?
Would it give someone a sense of self-worth to know that they go to work everyday and gum up the works a little bit compared to if they weren't there?
If you don't work, and receive UBI instead, all of a sudden you have more control over what you spend your valuable single life TIME on. Shouldn't that be better?
Programmers don't tell AI programs what to do any more. They tell them "look at all this, and learn its structure, and start predicting based on it"... and then maybe "act as you see fit given what you learned and classified and inferred and decided".
The programmer doesn't know what the AI knows, nor do they know in advance what patterns will be learned or what classification decisions made. The programmer, in AI applications going forward, won't have access to the input data in real time, nor will they be able to disentangle the internal state of the model being built by the AI, at least not in realtime.
What you're missing is the technological improvement trajectory for AI technology.
I do understand it, and can tell you it is improving and being generalized and requiring less human intervention for training at a steady pace. When thinking about implications, you need to think 5, 10, 20 years out, and assume that the pace of improvement of the technology will probably be faster than it has been over the last 5, 10, 20 years.
It's not about what the tech can do right now. It's where it's likely going, but going quickly, within a generation.
The whole point here is that AIs and AI-controlled systems will become more cost-effective than the median-intelligence, median-aptitude kind of person fairly soon.
Retraining won't help you, in general.
Unless it is training in political advocacy so you can get out and insist that politicians start seriously implementing universal basic income instead of boasting about "shovel-ready infrastructure projects". A wholesale shift of societal functioning and norms is going to be needed here.
and the management decided it was finished and "ship it" after the first iteration whereas the programmers assumed that was just an initial prototype, to be refined and tested in the second iteration.
You described general intelligence, with a reflective component, presumeably for purposes of improving performance.
What you described is clearly within range of near-future AI research advances. So while there are some technical puzzles remaining, there are not really any fundamental category-level mysteries about intelligence. It's advanced, general, learning, model-building, model-traversing-and-applying representative-information processing, and it's made efficient with particular architectures, particular algorithms (highly parallel, highly connected, hierarchical (DAG) etc.)
What is still a mystery is the qualia of consciousness. The personally experienced "sensation" or "impression" of being an "I" being in and perceiving the world. The experience of being the real-time "navigator" or "reader" of one's own intelligent aparatus (sensors, nervous system, brain) while not noticing that but rather noticing what it tells us as a narrative and personal-movie about the world around us.
Tononi has a theory that the qualia sensation just emerges when enough highly connected information is collected together in one place (i.e. one information network). I would add that there clearly would need to be information-processing processes comprehensively traversing the highly interconnected information network. A static memory with no processing isn't conscious.I'm on the fence about Tononi. It sounds as plausible as anything else I've heard, but is not really explanatory.I think I would also add that the information network being representative information "about" the world around the network, including the network's own body, would also seem to be a requirement for the emergence of structured narrative qualia.
But is that all of an explanation that is needed for qualia? It might have to suffice.
Qualia, like the possible properties of multiverses, might not be a domain that is subject to investigation by empirical science. But it still just might be an actual thing. Frustrating as hell isn't that? Qualia is one of those things that a sufficiently advanced AI could clearly fake with high fidelity, then tell you a story of all the wonderful things its attention was noticing, one after another. How would you know it was "faking" it? How do you know your human friend is not faking their story of their feeling of consciousness and introspection? You don't know. You assume they are not "faking" it because it is much simpler to assume they are functionally essentially just like you, so they are experiencing it just like you do. You Occam's Razor the hell out of it. And you're probably right, but just guessing. Their red is probably (save colour-blindness) just like your red, because their complex system including brain and visual system, and their acculturation to associates of red (like blood), is just like yours was. Occam says theirs is just like yours.
So when we get AIs that say they are "feeling it" and noticing one specific aspect after another of our wonderful world, why don't we just say "it's probably not faking it, any more than we are"? And drink a toast to Tononi who dared propose that the qualia of consciousness is an emergent property of large, highly interconnected (actively processed I would add) information networks.
Complex systems that are nonetheless stable or meta-stable (often fairly stable with occasional changes) tend to generate their own constraints on the state and processes and interaction of their constituent parts, and on the state and processes and interaction of other things in the environment of the system.
These constraints, or in other words, relatively simple patterns of arrangement and change can be described as properties of the system, and they only emerged after the particular kind of system was able to establish itself in the matter and energy of the region. So in other words, emergent properties.
Another phrase describing this situation might be emergent regularities in complex systems.
They definitely occur, are characterizable, have a degree of independence from many aspects of the specific physics of the constituent parts of the complex system (i.e. layer-independence, leading to non-reductive behaviour of the complex system), and are often the most important features or determinants of the complex system's state and evolution.
It's not bullshit, in general, it's just, well, complex, and admittedly no-one has satisfactorially formalized our knowledge of this kind of phenomenon yet. However progress is being made, such as work on nonequilibrium thermodynamics (maximum entropy principle etc).
We can create a system which varies smoothly between direct democracy (issue voting by everyone) and representative democracy.
The general idea is that for a certain time period, you can lend your vote anonymously to another person, who then has a weight of n+1 votes to vote on the issues/legislation that are put up for decision.
Variations of this method are the flexibility of the lending time period, and also whether or not you are allowed to "recall" your representative (withdraw your lent vote) earlier than the originally granted period. If you could, that's better for the lender, but the representatives won't know the voting weight that they have so won't be able to wheel and deal as accurately with other representatives, as US congress people do.
A key technical problem to be resolved is how to prevent forced lending of votes. Imagine a tyrannical "head of the family" husband forcing their wife and a few others for good measure to lend their votes to him. How do you make it impossible for them to know whether such "requested" lending happened or not? Ideas like a large random noise component in each person's vote weight, and randomization of the timing of lending transaction completion etc come to mind, but this might also be an insurmountable difficulty.
If there were a way of "proving" arms length relationship between lender and lendee, or only allowing pre-qualified (e.g. party-chosen) representatives (i.e. candidates) to receive lent votes, not just "everyone", then that might help with the forced "give me your vote" thing. I don't know. Ideas?
So the basic idea is, lend your vote out when you don't care or want to make the decision yourself. Get your vote back for when you want to decide yourself.
If we have voting directly on individual issues, as in direct democracy, I would allow people be able to give themselves a weight (from 1 to 10 say) that their vote carries in that voting tally.
People would be encouraged to reflect and only give themselves a high weight when they really care about the issue and also really know about it.
To make voluntary down-weighting more likely, People could save up a bit of unused vote weight to be dispensed within say the next 5 issue votes (or it's lost). So you could give yourself a weight of 19 on the next vote if you only used 1 of your 10 on the first.
But if you don't vote at all on 3 consecutive issues, you start losing voting weight for general non-participation in decision making. and you can only slowly build up your weight to 10 per vote again.
So this method both encourages participation, but also encourages judicious spending of each person's opinion on the matters they know and care most about.
The biggest flaw of course is that many people may be persuaded by demagogues to care the most about the issues they actually know the least about. Perhaps the method could be modified by conducting a qualification comprehension-of-the-issue test before each vote, and without passing it, you're limited to 5 of 10 max weight on that vote. Of course, the huge flaw in that is who sets the issue-test (and, the suspicion goes, biases it)? That would probably have to be done eventually by an impartial AI.
Someone who thinks that C is the only performant programming language today? Someone who's never heard of Go or rust, apparently.
Someone who thinks that more than, say, 10% of programs require "C-like" performance, with today's processors which are about 600,000 times faster than an Apple II, and thinks that both safety and maintenance cost are less important than having C-like performance on programs that don't need it.
Programming practices. Yeah ok. But people are people, programmers are programmers, and there are bell curves of skill, care and attention, schedule reasonableness under which programs are created.
So let's not assume every programmer writing a potentially security-relevant piece of code is a really good, well educated in best practices, really safe designer and coder with enough time for testing and iteration. Assuming that would be naive.
So why not protect against common errors using the programming language constraints and checks? Most of these protections can be done with very little cost, performance wise or in loss of program expressive power.
We can teach drivers not to start the car engine with the car in "drive" because they might run someone over, or we can just design the car not to start except in park.
Cryptocurrency, and also smart contract technology, once both are evolved into something easier to use for end users, is going to take over a substantial chunk of world economic functioning.
Yes there's a hype bubble, but just like with AI, there is extremely disruptive and important substance too. Don't let the presence of hype distract your from the fundamentals of the transformation that will take place.
To get closer to detecting truth vs well-crafted bullshit, more sophisticated techniques will be required such as: 1) Analysis of semantics (meaning) of the statements, and comparison with a large belief-strength-ranked knowledge base about the world. Where valid epistemic techniques are applied in the creation and vetting of the knowledge base. 2) Detection of who (person, affiliations) is the source (utterer) of the statement or statements. 3) Inference about likely general objectives of the utterer, and about their likely specific goal and interests in making the utterance, and the communication tactics being employed. 4) Detection of how much gain/loss interest the utterer has in the issue being discussed. 5) Discounting of plausibility of statements by utterers with strong goals and interests with respect to the subject of the statements. etc.
One for kernel-privilege code and one for user code, and a lightning fast way (use of a bit) that each instruction selects which of the two contexts it is using. I guess that would mean two separate caches etc.
This would enable efficient execution of true micro-kernel OSes wouldn't it?
Not sure if what I said there actually makes sense, in context of current chip architectures.
Basic idea is one particular kind of context switch, from kernel back and forth to user is not context caching/replacing but is just context selection from two.
1. Intelligence is just a process, with an awful lot of representative state, and some very general learning and inference algorithms running over that state and modifying that state, ideally in a whole bunch of parallel processing units, or in a smaller number of blindingly fast processing units, in a pinch.
2. It's complicated, so it's taking a while to figure it out.
3. People are figuring out, gradually, but with noticeable and significant progress. (People were musing about/sketching airplanes in the 1600s I think, and those people were no doubt snickered at. But technical progress happens, when the underlying scientific and engineering ideas are sound.)
4. 3. will continue, and A.I. will get better, probably non-linearly.
5. Eventually, and certainly within this century, and probably within its first third, some of this A.I. will be more cognitively capable, and certainly more knowledgeable, than the median human adult.
6. Flexible-purpose robotics is also similarly very tricky, but definitely do-able eventually. It's certainly getting noticeably better every decade. But even disembodied A.I. attached only to the Internet is enough to take many of today's jobs, even if we discount more generally useful robots.
7. At point 5., why would organisations and leaders wishing to get things done intelligently and efficiently use (lower than median-capable) people to do those things, when the automated A.I. version would be more cost-effective?
8. Yes, it's a just-so story, but you know what? Sometimes just-so stories will indeed be just so. And having studied the computational technology details and the philosophy, and loosely the neuroscience, of some of this for 30+ years, my bet is that it's happening. Just slightly too slowly, apparently, for you to be noticing. Oh and one more thing. Paper use in the office is, in fact, now declining, due to computers, displays, and the net. Took a while, but the fundamentals were always obvious. People laughed at the people who said that would happen, because they observed that computers were enabling MORE writing/reading paper use, not less. But that was not fundamental, it was a blip. A dead cat bounce as they say. And "more cool new jobs" during the "automation is getting smarter than people" age, is a similar dead cat bounce.
Slashdot Mirror
For some value of two.
0) Goes with out saying. Near instant, low-fee transactions. Comparable to debit card user experience.
1) Dead-easy for end user. No need whatsoever to understand how any of it works.
2) Currency price stability (at least comparable to any given major national currency such as USD, Euro).
If something built on top of the horrendously complicated lightning network can be made usable by average non-techie, without going completely central, then that could maybe address points 0 and 1.
Hard to see how Bitcoin achieves 2).
Hah! We have a Canadian company that is much much more advanced on this.
It is always only 5 years away from breakeven fusion. Take that!
I agree that corporate profit should be taxed to support UBI.
However why would you put on a punitive tax that punishes replacement of people by AI?
Why should people continue to do "make work" that automated machines can do more efficiently?
Would it give someone a sense of self-worth to know that they go to work everyday and gum up the works a little bit compared to if they weren't there?
If you don't work, and receive UBI instead, all of a sudden you have more control over what you spend your valuable single life TIME on. Shouldn't that be better?
in Pyongyang today.
in AI-driven companies FTW.
that's when you might want to start worrying.
Programmers don't tell AI programs what to do any more. They tell them "look at all this, and learn its structure, and start predicting based on it" ... and then maybe "act as you see fit given what you learned and classified and inferred and decided".
The programmer doesn't know what the AI knows, nor do they know in advance what patterns will be learned or what classification decisions made. The programmer, in AI applications going forward, won't have access to the input data in real time, nor will they be able to disentangle the internal state of the model being built by the AI, at least not in realtime.
What you're missing is the technological improvement trajectory for AI technology.
I do understand it, and can tell you it is improving and being generalized and requiring less human intervention for training at a steady pace. When thinking about implications, you need to think 5, 10, 20 years out, and assume that the pace of improvement of the technology will probably be faster than it has been over the last 5, 10, 20 years.
It's not about what the tech can do right now. It's where it's likely going, but going quickly, within a generation.
for many people.
The whole point here is that AIs and AI-controlled systems will become more cost-effective than the median-intelligence, median-aptitude kind of person fairly soon.
Retraining won't help you, in general.
Unless it is training in political advocacy so you can get out and insist that politicians start seriously implementing universal basic income instead of boasting about "shovel-ready infrastructure projects". A wholesale shift of societal functioning and norms is going to be needed here.
and the management decided it was finished and "ship it" after the first iteration whereas the programmers assumed that was just an initial prototype, to be refined and tested in the second iteration.
At a guess, it would be log(2) times more conscious. :-)
Same way my computer knows what it was up to after it wakes from "sleep" mode probably. Information processing and memory is in the same state it was.
On Pandora they dream, collectively, yes.
You described general intelligence, with a reflective component, presumeably for purposes of improving performance.
What you described is clearly within range of near-future AI research advances. So while there are some technical puzzles remaining, there are not really any fundamental category-level mysteries about intelligence. It's advanced, general, learning, model-building, model-traversing-and-applying representative-information processing, and it's made efficient with particular architectures, particular algorithms (highly parallel, highly connected, hierarchical (DAG) etc.)
What is still a mystery is the qualia of consciousness. The personally experienced "sensation" or "impression" of being an "I" being in and perceiving the world. The experience of being the real-time "navigator" or "reader" of one's own intelligent aparatus (sensors, nervous system, brain) while not noticing that but rather noticing what it tells us as a narrative and personal-movie about the world around us.
Tononi has a theory that the qualia sensation just emerges when enough highly connected information is collected together in one place (i.e. one information network). I would add that there clearly would need to be information-processing processes comprehensively traversing the highly interconnected information network. A static memory with no processing isn't conscious.I'm on the fence about Tononi. It sounds as plausible as anything else I've heard, but is not really explanatory.I think I would also add that the information network being representative information "about" the world around the network, including the network's own body, would also seem to be a requirement for the emergence of structured narrative qualia.
But is that all of an explanation that is needed for qualia? It might have to suffice.
Qualia, like the possible properties of multiverses, might not be a domain that is subject to investigation by empirical science. But it still just might be an actual thing. Frustrating as hell isn't that? Qualia is one of those things that a sufficiently advanced AI could clearly fake with high fidelity, then tell you a story of all the wonderful things its attention was noticing, one after another.
How would you know it was "faking" it? How do you know your human friend is not faking their story of their feeling of consciousness and introspection? You don't know. You assume they are not "faking" it because it is much simpler to assume they are functionally essentially just like you, so they are experiencing it just like you do. You Occam's Razor the hell out of it. And you're probably right, but just guessing. Their red is probably (save colour-blindness) just like your red, because their complex system including brain and visual system, and their acculturation to associates of red (like blood), is just like yours was. Occam says theirs is just like yours.
So when we get AIs that say they are "feeling it" and noticing one specific aspect after another of our wonderful world, why don't we just say "it's probably not faking it, any more than we are"? And drink a toast to Tononi who dared propose that the qualia of consciousness is an emergent property of large, highly interconnected (actively processed I would add) information networks.
"I" am the program-counter, it seems.
Complex systems that are nonetheless stable or meta-stable (often fairly stable with occasional changes) tend to generate their own constraints on the state and processes and interaction of their constituent parts, and on the state and processes and interaction of other things in the environment of the system.
These constraints, or in other words, relatively simple patterns of arrangement and change can be described as properties of the system, and they only emerged after the particular kind of system was able to establish itself in the matter and energy of the region. So in other words, emergent properties.
Another phrase describing this situation might be emergent regularities in complex systems.
They definitely occur, are characterizable, have a degree of independence from many aspects of the specific physics of the constituent parts of the complex system (i.e. layer-independence, leading to non-reductive behaviour of the complex system), and are often the most important features or determinants of the complex system's state and evolution.
It's not bullshit, in general, it's just, well, complex, and admittedly no-one has satisfactorially formalized our knowledge of this kind of phenomenon yet. However progress is being made, such as work on nonequilibrium thermodynamics (maximum entropy principle etc).
We can create a system which varies smoothly between direct democracy (issue voting by everyone) and representative democracy.
The general idea is that for a certain time period, you can lend your vote anonymously to another person, who then has a weight of n+1 votes to vote on the issues/legislation that are put up for decision.
Variations of this method are the flexibility of the lending time period, and also whether or not you are allowed to "recall" your representative (withdraw your lent vote) earlier than the originally granted period. If you could, that's better for the lender, but the representatives won't know the voting weight that they have so won't be able to wheel and deal as accurately with other representatives, as US congress people do.
A key technical problem to be resolved is how to prevent forced lending of votes. Imagine a tyrannical "head of the family" husband forcing their wife and a few others for good measure to lend their votes to him. How do you make it impossible for them to know whether such "requested" lending happened or not?
Ideas like a large random noise component in each person's vote weight, and randomization of the timing of lending transaction completion etc come to mind, but this might also be an insurmountable difficulty.
If there were a way of "proving" arms length relationship between lender and lendee, or only allowing pre-qualified (e.g. party-chosen) representatives (i.e. candidates) to receive lent votes, not just "everyone", then that might help with the forced "give me your vote" thing. I don't know. Ideas?
So the basic idea is, lend your vote out when you don't care or want to make the decision yourself. Get your vote back for when you want to decide yourself.
If we have voting directly on individual issues, as in direct democracy, I would allow people be able to give themselves a weight (from 1 to 10 say) that their vote carries in that voting tally.
People would be encouraged to reflect and only give themselves a high weight when they really care about the issue and also really know about it.
To make voluntary down-weighting more likely, People could save up a bit of unused vote weight to be dispensed within say the next 5 issue votes (or it's lost). So you could give yourself a weight of 19 on the next vote if you only used 1 of your 10 on the first.
But if you don't vote at all on 3 consecutive issues, you start losing voting weight for general non-participation in decision making. and you can only slowly build up your weight to 10 per vote again.
So this method both encourages participation, but also encourages judicious spending of each person's opinion on the matters they know and care most about.
The biggest flaw of course is that many people may be persuaded by demagogues to care the most about the issues they actually know the least about. Perhaps the method could be modified by conducting a qualification comprehension-of-the-issue test before each vote, and without passing it, you're limited to 5 of 10 max weight on that vote. Of course, the huge flaw in that is who sets the issue-test (and, the suspicion goes, biases it)? That would probably have to be done eventually by an impartial AI.
Someone who thinks that C is the only performant programming language today?
Someone who's never heard of Go or rust, apparently.
Someone who thinks that more than, say, 10% of programs require "C-like" performance, with today's processors which are about 600,000 times faster than an Apple II, and thinks that both safety and maintenance cost are less important than having C-like performance on programs that don't need it.
Programming practices. Yeah ok.
But people are people, programmers are programmers, and there are bell curves of skill, care and attention, schedule reasonableness under which programs are created.
So let's not assume every programmer writing a potentially security-relevant piece of code is a really good, well educated in best practices, really safe designer and coder with enough time for testing and iteration. Assuming that would be naive.
So why not protect against common errors using the programming language constraints and checks? Most of these protections can be done with very little cost, performance wise or in loss of program expressive power.
We can teach drivers not to start the car engine with the car in "drive" because they might run someone over, or we can just design the car not to start except in park.
computers.
No bounds checking, no type checking. In 2018. Get serious.
My guess is C programs are the underlying reason for a good majority of ways of hacking into systems.
I mean a buffer exploit? Seriously? In 2018? Why in the hell would that be remotely acceptable?
Cryptocurrency, and also smart contract technology, once both are evolved into something easier to use for end users, is going to take over a substantial chunk of world economic functioning.
Yes there's a hype bubble, but just like with AI, there is extremely disruptive and important substance too.
Don't let the presence of hype distract your from the fundamentals of the transformation that will take place.
Did you misplace your bitcoin?
What about small pads of paper and pens?
that is, "truthiness" not truth.
To get closer to detecting truth vs well-crafted bullshit, more sophisticated techniques will be required such as:
1) Analysis of semantics (meaning) of the statements, and comparison with a large belief-strength-ranked knowledge base about the world. Where valid epistemic techniques are applied in the creation and vetting of the knowledge base.
2) Detection of who (person, affiliations) is the source (utterer) of the statement or statements.
3) Inference about likely general objectives of the utterer, and about their likely specific goal and interests in making the utterance, and the communication tactics being employed.
4) Detection of how much gain/loss interest the utterer has in the issue being discussed.
5) Discounting of plausibility of statements by utterers with strong goals and interests with respect to the subject of the statements.
etc.
One for kernel-privilege code and one for user code, and a lightning fast way (use of a bit) that each instruction selects which of the two contexts it is using. I guess that would mean two separate caches etc.
This would enable efficient execution of true micro-kernel OSes wouldn't it?
Not sure if what I said there actually makes sense, in context of current chip architectures.
Basic idea is one particular kind of context switch, from kernel back and forth to user is not context caching/replacing but is just context selection from two.
Comments? Cure my ignorance?
It's not a matter of evidence, yet, ShanghaiBill.
It's a matter of applied commonsense reasoning.
1. Intelligence is just a process, with an awful lot of representative state, and some very general learning and inference algorithms running over that state and modifying that state, ideally in a whole bunch of parallel processing units, or in a smaller number of blindingly fast processing units, in a pinch.
2. It's complicated, so it's taking a while to figure it out.
3. People are figuring out, gradually, but with noticeable and significant progress. (People were musing about/sketching airplanes in the 1600s I think, and those people were no doubt snickered at. But technical progress happens, when the underlying scientific and engineering ideas are sound.)
4. 3. will continue, and A.I. will get better, probably non-linearly.
5. Eventually, and certainly within this century, and probably within its first third, some of this A.I. will be more cognitively capable, and certainly more knowledgeable, than the median human adult.
6. Flexible-purpose robotics is also similarly very tricky, but definitely do-able eventually. It's certainly getting noticeably better every decade. But even disembodied A.I. attached only to the Internet is enough to take many of today's jobs, even if we discount more generally useful robots.
7. At point 5., why would organisations and leaders wishing to get things done intelligently and efficiently use (lower than median-capable) people to do those things, when the automated A.I. version would be more cost-effective?
8. Yes, it's a just-so story, but you know what? Sometimes just-so stories will indeed be just so. And having studied the computational technology details and the philosophy, and loosely the neuroscience, of some of this for 30+ years, my bet is that it's happening. Just slightly too slowly, apparently, for you to be noticing. Oh and one more thing. Paper use in the office is, in fact, now declining, due to computers, displays, and the net. Took a while, but the fundamentals were always obvious. People laughed at the people who said that would happen, because they observed that computers were enabling MORE writing/reading paper use, not less. But that was not fundamental, it was a blip. A dead cat bounce as they say. And "more cool new jobs" during the "automation is getting smarter than people" age, is a similar dead cat bounce.