If I provide a more competitive price or product than you, you will not make a sale, and you will see that as a 'loss' to your business. You don't have a right to profit, and *should* lose to the more competitive supplier.
Traders increase liquidity and decrease market spread. These are good things.
I disagree. I think the answer is to *allow* placing orders that you might not *want* to be filled.
Cancelling orders is normal even for manual trading. Normally the price moves against you and you have to place another order at a worse price for yourself. Sometimes you are at the front of the queue, but you can see your side dropping off, so you cancel to place a better price for you (but this happens less often).
Given that you almost always must be able to cancel your orders, and given that your orders *may* get taken even if your don't *intend* for them to be taken, let's just drop the shenanigans and allow people (and bots) to place and cancel orders as fast as they possibly can given that we now have the technology, and that they are always at risk of having their orders matched.
Then smart uni students can compete with the HFT bots and take their money when the uni students provide more efficient pricing than the HFT bots (by definition, that's how the market is supposed to work).
The efficient market hypothesis is how the market should operate IN THE LIMIT! The market provides INCENTIVES for people to provide TIMELY and ACCURATE **INFORMATION** to the markets.
I wasn't addressing cyber attacks, I was addressing mechanisms to deliberately slow down HFT trading and the like. HFT trading *MUST* increase liquidity. Volatility is *natural* in markets (including chaos), and rewards those who reduce it.
HFT is good actually for markets, and good for everyone, because, if you actually examine the operation of markets (as double opposing queues), you will see that it is NEVER possible to harm another trader by placing trades, because you only provide (at least) BETTER opportunities for other traders, or match other traders SOONER than they OTHERWISE would have been.
The problem with HFT is that if you use their tactics against them you go to jail. Where you should be allowed to place and cancel orders and the profits go to the MOST efficient provider of information to a market.
> Here's a cut and paste from someone that has not heard of the 1987 Black Monday crash.
Quite the contrary. Crashes are GOOD things for some people as it provides the opportunity to get stocks at very low prices. Especially transient crashes... Similarly with bubbles. Volatility rewards those who act against it... As the market SHOULD.
The big problem is that everyone wants the stock market to only go UP... This looks good for politicians and everyone thinks that helps their retirement plan and stuff... The purpose of the market is not to always go up, but to provide information and allocate resources efficiently.
What you describe is only possible with "front running" where an actor can intercept orders before they are placed on the queue and made publicly available.
If everyone only has access to orders after they are on the queue, and publicly available, then the (let's use) HFT operators can only either improve the buyer's opportunities by placing a sell order at a lower price than the existing orders, improve the seller's opportunities by placing a buy order at a higher price than the existing orders, or have no effect at all.
If people actually understood the stock market from the game theoretic point of view that it is designed as, they would see that no order can be placed to the detriment of any other actor's orders, and that in fact, every order either increases the value of the market to *some* set of actors in that market or at worse has no effect at all.
There is no such thing as a market order that can unfairly affect anyone else's position in the market (with the exceptions of insider trading and front running, both of them are rightly illegal).
The "fixes" people propose to the market actually tend to make it worse. For example the, "best execution" laws have made the system worse, as they have attempted to implement the physically impossible by force of law, where simple arbitrage already provided the mechanisms and market incentives to do this. (As a result of this "fix", something like front running becomes possible for the select few).
Slowing down the market like you suggest is another unnecessary and actually harmful solution to a system that is specifically designed and intended to provide instantly up to date prices that reflect all known information about a stock, in the limit (the efficient market hypothesis).
Actually allowing the stock market to operate as it is intended, with all the price volatility that involves, is actually the best available option. Those who correctly act against that volatility will profit.
One more example, there is a law that states an order cannot be placed that you do not intend to fill. A couple of uni students saw patterns and vulnerabilities in HFT bots, that reacted to large orders that were taken down after the uni student's own bots ate up the orders the HFT bots placed in response to them. The market moved wealth from the inefficient HFT bots into the hands of the more efficient uni students just as it is designed to do. However, this upset the HFT bot owners (Goldman Sachs and co)... so the uni students were arrested, charged and imprisoned, instead of being rewarded with the millions they rightfully earned. The IRONY of all this is that the HFT bots place and cancel many thousands of orders per second... but they are owned by ultra rich people and this is not a free market, but crony capitalism.
The correct answer to this, although highly unlikely, is for people like you to learn how these things are meant to operate and understand them before you suggest ideas that actually cause harm instead of solve problems... I guarantee you that a "solution" like this will result in even more money for the likes of GS and co at the cost of everyone else.
10 years since my last formal contract assignment... 1 year on Australian NewStart jobseekers allowance...
Your mentality has the government trying to force me into doing 30 hours work a fortnight for an extra $20 (called "work for the dole!")... Presumably because "society" think it's "good for me to work", at best it's so I'm not "getting something for nothing"... It's beyond retarded.
Firstly, learn to live your own life. If you chose to work, go work, be happy. Don't tell others how to live. Being on welfare sucks, but really it's being "poor" that sucks... but if I wanted to work, and someone offered me enough to do it, i would... but this concept of "needing" to work is stupid. Even more is the idea of being forced to work for no extra reward.
If you had enough to survive... "food and house, enjoy"... then it's up to you to find a way to fill that time. Go volunteer if it makes you happy... do "work" if it makes you happy... but no one has a "right" to be employed, because that requires someone to employ you against *their* will.
I have found most of the problems of being unemployed are due to being poor. When I was just as unemployed but had large stacks from being an engineer, I was never bored or depressed. When I can't go out every night to restaurants and bars, or travel international, and stuff, and just sit in my house all day to stretch out the budget... That's where the problems are... in being poor, not in not working.
Basically the meme that "everyone needs work to be fulfilled" is propaganda by capitalists to create cheap labour, or propaganda by communists to create slave labour. Rather, learn the free market model of microeconomics, support a UBI, and follow your own utility (at no cost to others) and BE FREE!
Stop this stupid meme... work is merely a MEANS to an END... If you have the MEANS, you will find that not working is just as fun if not more so. Look at the idle kids of the ultra wealthy for plenty of happy unworking people. It's never the work that makes me happy, it's in the spending!
I figure that even if robots become supernormally attractive, some people will still prefer to know that they are with actual humans instead of robots.
Also, by the time we get supernormally attractive sex robots, we will probably have replaced AI engineers, as it will probably require a level of engineering that humans would struggle at. So, I don't think it will be one of the first, but one of the later (think of all the things AI is currently doing).
I also just realised that having a human 'in the loop' will probably protect athletes too. An AI car might out perform an F1 driver, but we like to see humans competing. Maybe also for entertainers?
On the other side, some people already prefer machines! I just think there will always be a significant number of people that would prefer a human (at least some times) to a beyond porn star quality robot... simply because they are not robots.
> When, eventually, the computers can think, why would they want us around at all?
And this is exactly the problem.
Firstly, while I think computers will best humans at any and every intellectual and physical activity (almost all work will be obsoleted), I don't think they are likely to be truly autonomous (we better damn well hope not)... they don't have to "want" anything the way we do, because we will build into them what they "want" to be doing... Instead, they will continue to be simply machines, or rather, capital, owned by corporations and governments. Corporations will eventually be able to task the AI with the function "make money" and then the 0.1% won't have any need for you at all.
It's not that the computers won't want us around, it's more that the elite that own the computers won't have any work for us, and we will become obsolete the way the internal combustion engine made horses obsolete (it hasn't worked out well for horses at all, btw)... We will need to change to some basic concepts about our capitalist economic system if most of the humans are to survive (and thrive). We can either end up in a utopia with AI automation, or a dystopia...
My personal beliefs on this are that we need to let go of putting so much emphasis on "work" and "jobs" as being the primary means of existence. We don't have to throw away capitalism, and can keep all of it's incentives, we just need to ensure that the wealth generated by AI is distributed better than we currently distribute the gains in wealth enabled by capitalism. I propose a UBI, funded by taxes on income, capital gains, and wealth (because wealth concentrates, it floods up).
Really I wish more people understood microeconomics so that these fixes would be comprehended, obvious to all, and politically viable... but I'm probably dreaming... hope you can all enjoy your starvation instead.:)
As far as your philosophy goes, it's called solopism, and what you claim is impossible to prove for computers is just as impossible to prove for humans.
Prove that *you* have your own thoughts and aren't just responding to stimuli. Basically you can't, and that *is* a philosophical problem, but irrelevant.
What matters is whether AI can perform as well or better than humans at any given task... And with very few exceptions (maybe prostitution) I can't think of anything that an AI won't be able to best humans at. It's just a matter of them being sufficiently advanced. They are already better at law than law graduates (can find the relevant laws and case history, etc, faster, from natural language problems), better than doctors at certain diagnostics, better than experts even at image recognition problems... over time the things that AI is better at than humans is only going to explode, probably at an exponential rate. Once we teach them to do AI engineering, it's all over.
By the way, "thought vectors" are a term used in AI/ML today.
> Chess is a deterministic problem as most games are, I never believed computers wouldn't ever be better than humans.
Right, but you probably grew up in an age where computers were already better at chess than humans. Before they were better, Kasparov (the chess grand master), was famous for saying that computers would never beat humans because "Chess is a unique cognitive nexus, a place where art and science come together in the human mind and are then refined and improved by experience.".
> I'd be willing to bet (tooth picks) that playing poker with AI would not be as deterministic, that the humans would win at least some times
Poker was, until very recently, "unsolved" with regard to AI. Only in the last couple of years has AI bested poker experts. Again, people said it couldn't be done, and now it's done. The best heads up texas no limit poker players are now AI. No, they don't win every hand, what they do is beat human experts statistically and consistently over many hands. It is extremely unlikely humans will ever be able to take back the position of best poker player, just as it is unlikely that anyone will ever again beat computers at chess.
> In sort, some things can be learned and trained, but throw a poker playing computer into a game of Gin Rummy would not garner very many wins
We don't do it yet, but again, there is no reason why we can't reuse the same AI to play multiple card games. Furthermore, with enough advances there is no reason a more generalised AI won't be able to pick up new games simply by reading the rules and a bit of practice (internally, against itself). Think of it as changing the goal posts from building an AI to play poker, to building an AI to play cards in general. We're definitely on the path to something as 'simple' as that to being feasible in the next few years.
Your arguments are as simple minded as Kasparov's... just because it hasn't been done yet is no reason to believe it won't be done (and soon I bet!).
I actually think the next big step in AI is training the same neural network to perform very many and varied tasks, as opposed to training many and varied AIs each to do one task. Some form of multimodal AI. Then it will be able to apply cross domain knowledge, and do exactly what you are currently claiming it cannot do... including learning new games by simply being told the rules for it.
> But I'm not going to be able to prove this to you.... Why?
Because your simply wrong.
This is the history of AI, people saying that computers will never do X, then people building computers that do X.
There is no reason to believe we can't make AIs that be given the task to build software or robots to perform tasks we explain to them in natural language. It's just another 'game' after all.
Already they can make natural language descriptions of images... We should be able to have them do the inverse. We can teach them maths, then we can teach them software engineering.
It's not a question of philosophy, or even if, it's a question of when.
By zero shot learning, it was able to learn the task by only a few examples and extrapolated the rest from previous experience. So, on day four you tell it has a new task, and show it just one purple object, then it knows that the key is purple objects from then on.
Absolutely no different to how humans would deal with it.
A multimodal AI with enough modes and experience will not have the limits you are used to thinking about for AI. Think of using the same AI to control robots, drive cars, fly planes, make financial decisions, play chess, go and poker, manage a business, teach us new maths, design, build and maintain robots, software and better AI, make medical diagnoses, interpret law, etc, etc...
In fact, the whole deal with AI is finding the frontier of what computers can't do, and every time large numbers of people say it can't be done, and every time we smash through those frontiers. (AI will never beat humans at chess, image recognition, go, poker... etc... and now AI is better at all of these than the best humans... this trend will continue).
We don't know the "rules" that define the differences between a cat and a dog, yet neural networks can work them out through experience.
Nor do we know the "rules" that make up a winning go strategy (as opposed to the rules of the game itself)... yet AlphaGo managed to work those out itself too.
Everything is just games from some point of view. Eventually AI will be better at them than any human (or group of humans) can be (including at your job).
This machine has different types of inputs, and different types of outputs, and the same AI learns many different tasks as opposed to just one very narrowly task.
Apparently this machine is capable of performing many different tasks at equivalent to where the state of the art was just five years ago. I don't believe it will take long for it to 'catch up'.
> Computers will be useful for specifically defined tasks, but they will neither recognize the boundaries of what they know or choose to stray outside their limits. They will always just respond as they have been programed to.
Now the real interesting thing about it is that in some cases it was able to use information it had learned from previous tasks in new tasks that it was given... and so was actually able to perform so called "zero shot learning"... it already new enough about the world to perform well on tasks it had never been trained to do!
So, as AI evolves, I think that AI will be very capable at tasks outside of its original training.
Go cannot be exhaustively searched the way chess and many other games can. The branching factor (number of possible moves) is simply too high.
AlphaGo is a breakthrough because it learns what is basically an "intuition" for good moves and board states by learning from examples and playing against itself, in a way similar to how we train networks the differences between cats and dogs.
> but we won't *ever* get a computer to think, feel and decide based on intuition.
That's the thing with AlphaGo though. The search tree for Go is simply too large to deal with in a traditional manner (minmax algorithm) the way we can with chess. Instead AlphaGo analyses the board holistically and basically develops an "intuition" for the best move.
The current state of the art image recognition systems are not built on rules, they "learn" the rules themselves from examples... In fact, we don't even have a good grasp on *how* they recognize the various images, just that they do. (For example, why does the network think it's a cat and not a dog... we just don't know beyond "it lights up these 'neurons' we trained with back propagation").
That's pretty damn close to decision by feel and intuition.
> But we humans MUST have some sort of work. We evolved to work and without it, we lose a sense of purpose and we resort to drinking.
No we don't.
If you WANT to work then go do voluntary work or whatever. If you don't, then why work?
People should really learn some microeconomics to understand utility.
Also, the point of UBI is to provide welfare for everyone, but without the trappings of welfare, like having to forego work in order to be eligible for it. With welfare, you can end up with less money for working, or at the very least suffer very high effective marginal tax rates. With a UBI, every dollar you earn is more money in your pocket... It increases the incentive to work compared to standard welfare systems.
I know what you mean by regret. The market doesn't have regret, the stupid actors do, and the smarter ones have joy!
The market is already the best predictor we have!
Knowing the best predictor at any instant would pick the one that said the winner will win... likely the best predictor at any instant is an actual idiot. It does 'learn' the long term best experts.
You had it with exponentially adjusted weighted sum of experts.
I read a chunk off of your pdf... if you can't exclude users who were getting their information from actual exchanges to inform them of their play money bets, you shouldn't be surprised when they work... just bet the difference between the play money market prediction and the real market and you'll profit.
Useless internet points do provide utility even when they aren't money... but they don't provide utility like actual money.
> The claim is that a prediction market can act as well as the best participant of that market, i.e. have zero regret.
The market has no regret... some agents will, others won't.
> - If the claim were true, you could dump a lot (and I mean an extreme amount) of comparatively simple algorithms into a virtual prediction market, and by the claim, the market would act just as well as the very best algorithm of the lot, producing a super AI.
That doesn't follow... it doesn't do better than the best prediction on any given outcome... it's over time the best on average, You show me your exponential set of simple algorithms that contain at least one super AI, and the market outcome will be the super AI. It's just a matter of TPS for the market...
I think a virtual market type setup would, for example, find the long term optimal weighted sum of AI expert predictions...
I think why you wouldn't use a virtual market for training an ensemble, is that you have all the information already... you can condition an ensemble on the input... but in the real world we do not have that information, we just have our information, and markets create the incentive to provide that information, because we can profit from that it.
The real market is already a super intelligence... collating more information than any individual human (or not) expert can by being the weighted sum of expert human opinions already!
You have to show me your AI that can profit from what already exists first!
Funny thing is, they do exist and the best of them are thriving! Still I don't think we have any super AIs yet!
BTW: Swarm AI just seems to be an equal weighting of random 'experts'... and assigns it as win. That's what I'm saying, it can't possibly beat the market... unsurprising it failed to pick a 1 in 70k event... at least that's what the market thought, right? If anything it's a consensus creator...
> Suppose, for instance, that there's a prediction market on when the next act of terrorism will occur in the US, as defined by some criterion
Yeah, in this and the keynesian beauty contest example the outcomes are dependent on the market itself... so, you can use them to create assassination markets or can you get weird feedback loops.
So, where the outcome is independent of the market itself they work. You're not going to change tomorrow's weather by betting on it, yet that would be useful for many people, but the pope might get shot if the odds go long enough!
I would think that it should be regulated to that extent... I think people know the difference... other than that, why are they illegal for?
Slashdot Mirror
No.
This is not how it works.
If I provide a more competitive price or product than you, you will not make a sale, and you will see that as a 'loss' to your business. You don't have a right to profit, and *should* lose to the more competitive supplier.
Traders increase liquidity and decrease market spread. These are good things.
I disagree. I think the answer is to *allow* placing orders that you might not *want* to be filled.
Cancelling orders is normal even for manual trading. Normally the price moves against you and you have to place another order at a worse price for yourself. Sometimes you are at the front of the queue, but you can see your side dropping off, so you cancel to place a better price for you (but this happens less often).
Given that you almost always must be able to cancel your orders, and given that your orders *may* get taken even if your don't *intend* for them to be taken, let's just drop the shenanigans and allow people (and bots) to place and cancel orders as fast as they possibly can given that we now have the technology, and that they are always at risk of having their orders matched.
Then smart uni students can compete with the HFT bots and take their money when the uni students provide more efficient pricing than the HFT bots (by definition, that's how the market is supposed to work).
You don't actually trade!
I've built bots that trade on markets... you are ignorant of their fundamental operation as well as microeconomics.
The efficient market hypothesis is how the market should operate IN THE LIMIT! The market provides INCENTIVES for people to provide TIMELY and ACCURATE **INFORMATION** to the markets.
I wasn't addressing cyber attacks, I was addressing mechanisms to deliberately slow down HFT trading and the like. HFT trading *MUST* increase liquidity. Volatility is *natural* in markets (including chaos), and rewards those who reduce it.
HFT is good actually for markets, and good for everyone, because, if you actually examine the operation of markets (as double opposing queues), you will see that it is NEVER possible to harm another trader by placing trades, because you only provide (at least) BETTER opportunities for other traders, or match other traders SOONER than they OTHERWISE would have been.
The problem with HFT is that if you use their tactics against them you go to jail. Where you should be allowed to place and cancel orders and the profits go to the MOST efficient provider of information to a market.
> Here's a cut and paste from someone that has not heard of the 1987 Black Monday crash.
Quite the contrary. Crashes are GOOD things for some people as it provides the opportunity to get stocks at very low prices. Especially transient crashes... Similarly with bubbles. Volatility rewards those who act against it... As the market SHOULD.
The big problem is that everyone wants the stock market to only go UP... This looks good for politicians and everyone thinks that helps their retirement plan and stuff... The purpose of the market is not to always go up, but to provide information and allocate resources efficiently.
No, you are simply wrong.
What you describe is only possible with "front running" where an actor can intercept orders before they are placed on the queue and made publicly available.
If everyone only has access to orders after they are on the queue, and publicly available, then the (let's use) HFT operators can only either improve the buyer's opportunities by placing a sell order at a lower price than the existing orders, improve the seller's opportunities by placing a buy order at a higher price than the existing orders, or have no effect at all.
You don't trade.
If people actually understood the stock market from the game theoretic point of view that it is designed as, they would see that no order can be placed to the detriment of any other actor's orders, and that in fact, every order either increases the value of the market to *some* set of actors in that market or at worse has no effect at all.
There is no such thing as a market order that can unfairly affect anyone else's position in the market (with the exceptions of insider trading and front running, both of them are rightly illegal).
The "fixes" people propose to the market actually tend to make it worse. For example the, "best execution" laws have made the system worse, as they have attempted to implement the physically impossible by force of law, where simple arbitrage already provided the mechanisms and market incentives to do this. (As a result of this "fix", something like front running becomes possible for the select few).
Slowing down the market like you suggest is another unnecessary and actually harmful solution to a system that is specifically designed and intended to provide instantly up to date prices that reflect all known information about a stock, in the limit (the efficient market hypothesis).
Actually allowing the stock market to operate as it is intended, with all the price volatility that involves, is actually the best available option. Those who correctly act against that volatility will profit.
One more example, there is a law that states an order cannot be placed that you do not intend to fill. A couple of uni students saw patterns and vulnerabilities in HFT bots, that reacted to large orders that were taken down after the uni student's own bots ate up the orders the HFT bots placed in response to them. The market moved wealth from the inefficient HFT bots into the hands of the more efficient uni students just as it is designed to do. However, this upset the HFT bot owners (Goldman Sachs and co)... so the uni students were arrested, charged and imprisoned, instead of being rewarded with the millions they rightfully earned. The IRONY of all this is that the HFT bots place and cancel many thousands of orders per second... but they are owned by ultra rich people and this is not a free market, but crony capitalism.
The correct answer to this, although highly unlikely, is for people like you to learn how these things are meant to operate and understand them before you suggest ideas that actually cause harm instead of solve problems... I guarantee you that a "solution" like this will result in even more money for the likes of GS and co at the cost of everyone else.
TLDR; Learn microeconomics.
10 years since my last formal contract assignment... 1 year on Australian NewStart jobseekers allowance...
Your mentality has the government trying to force me into doing 30 hours work a fortnight for an extra $20 (called "work for the dole!")... Presumably because "society" think it's "good for me to work", at best it's so I'm not "getting something for nothing"... It's beyond retarded.
Firstly, learn to live your own life. If you chose to work, go work, be happy. Don't tell others how to live. Being on welfare sucks, but really it's being "poor" that sucks... but if I wanted to work, and someone offered me enough to do it, i would... but this concept of "needing" to work is stupid. Even more is the idea of being forced to work for no extra reward.
If you had enough to survive... "food and house, enjoy"... then it's up to you to find a way to fill that time. Go volunteer if it makes you happy... do "work" if it makes you happy... but no one has a "right" to be employed, because that requires someone to employ you against *their* will.
I have found most of the problems of being unemployed are due to being poor. When I was just as unemployed but had large stacks from being an engineer, I was never bored or depressed. When I can't go out every night to restaurants and bars, or travel international, and stuff, and just sit in my house all day to stretch out the budget... That's where the problems are... in being poor, not in not working.
Basically the meme that "everyone needs work to be fulfilled" is propaganda by capitalists to create cheap labour, or propaganda by communists to create slave labour. Rather, learn the free market model of microeconomics, support a UBI, and follow your own utility (at no cost to others) and BE FREE!
Stop this stupid meme... work is merely a MEANS to an END... If you have the MEANS, you will find that not working is just as fun if not more so. Look at the idle kids of the ultra wealthy for plenty of happy unworking people. It's never the work that makes me happy, it's in the spending!
I figure that even if robots become supernormally attractive, some people will still prefer to know that they are with actual humans instead of robots.
Also, by the time we get supernormally attractive sex robots, we will probably have replaced AI engineers, as it will probably require a level of engineering that humans would struggle at. So, I don't think it will be one of the first, but one of the later (think of all the things AI is currently doing).
I also just realised that having a human 'in the loop' will probably protect athletes too. An AI car might out perform an F1 driver, but we like to see humans competing. Maybe also for entertainers?
On the other side, some people already prefer machines! I just think there will always be a significant number of people that would prefer a human (at least some times) to a beyond porn star quality robot... simply because they are not robots.
> When, eventually, the computers can think, why would they want us around at all?
And this is exactly the problem.
Firstly, while I think computers will best humans at any and every intellectual and physical activity (almost all work will be obsoleted), I don't think they are likely to be truly autonomous (we better damn well hope not)... they don't have to "want" anything the way we do, because we will build into them what they "want" to be doing... Instead, they will continue to be simply machines, or rather, capital, owned by corporations and governments. Corporations will eventually be able to task the AI with the function "make money" and then the 0.1% won't have any need for you at all.
It's not that the computers won't want us around, it's more that the elite that own the computers won't have any work for us, and we will become obsolete the way the internal combustion engine made horses obsolete (it hasn't worked out well for horses at all, btw)... We will need to change to some basic concepts about our capitalist economic system if most of the humans are to survive (and thrive). We can either end up in a utopia with AI automation, or a dystopia...
My personal beliefs on this are that we need to let go of putting so much emphasis on "work" and "jobs" as being the primary means of existence. We don't have to throw away capitalism, and can keep all of it's incentives, we just need to ensure that the wealth generated by AI is distributed better than we currently distribute the gains in wealth enabled by capitalism. I propose a UBI, funded by taxes on income, capital gains, and wealth (because wealth concentrates, it floods up).
Really I wish more people understood microeconomics so that these fixes would be comprehended, obvious to all, and politically viable... but I'm probably dreaming... hope you can all enjoy your starvation instead. :)
As far as your philosophy goes, it's called solopism, and what you claim is impossible to prove for computers is just as impossible to prove for humans.
Prove that *you* have your own thoughts and aren't just responding to stimuli. Basically you can't, and that *is* a philosophical problem, but irrelevant.
What matters is whether AI can perform as well or better than humans at any given task... And with very few exceptions (maybe prostitution) I can't think of anything that an AI won't be able to best humans at. It's just a matter of them being sufficiently advanced. They are already better at law than law graduates (can find the relevant laws and case history, etc, faster, from natural language problems), better than doctors at certain diagnostics, better than experts even at image recognition problems... over time the things that AI is better at than humans is only going to explode, probably at an exponential rate. Once we teach them to do AI engineering, it's all over.
By the way, "thought vectors" are a term used in AI/ML today.
> Chess is a deterministic problem as most games are, I never believed computers wouldn't ever be better than humans.
Right, but you probably grew up in an age where computers were already better at chess than humans. Before they were better, Kasparov (the chess grand master), was famous for saying that computers would never beat humans because "Chess is a unique cognitive nexus, a place where art and science come together in the human mind and are then refined and improved by experience.".
> I'd be willing to bet (tooth picks) that playing poker with AI would not be as deterministic, that the humans would win at least some times
Poker was, until very recently, "unsolved" with regard to AI. Only in the last couple of years has AI bested poker experts. Again, people said it couldn't be done, and now it's done. The best heads up texas no limit poker players are now AI. No, they don't win every hand, what they do is beat human experts statistically and consistently over many hands. It is extremely unlikely humans will ever be able to take back the position of best poker player, just as it is unlikely that anyone will ever again beat computers at chess.
> In sort, some things can be learned and trained, but throw a poker playing computer into a game of Gin Rummy would not garner very many wins
We don't do it yet, but again, there is no reason why we can't reuse the same AI to play multiple card games. Furthermore, with enough advances there is no reason a more generalised AI won't be able to pick up new games simply by reading the rules and a bit of practice (internally, against itself). Think of it as changing the goal posts from building an AI to play poker, to building an AI to play cards in general. We're definitely on the path to something as 'simple' as that to being feasible in the next few years.
Your arguments are as simple minded as Kasparov's... just because it hasn't been done yet is no reason to believe it won't be done (and soon I bet!).
I actually think the next big step in AI is training the same neural network to perform very many and varied tasks, as opposed to training many and varied AIs each to do one task. Some form of multimodal AI. Then it will be able to apply cross domain knowledge, and do exactly what you are currently claiming it cannot do... including learning new games by simply being told the rules for it.
> But I'm not going to be able to prove this to you.... Why?
Because your simply wrong.
This is the history of AI, people saying that computers will never do X, then people building computers that do X.
There is no reason to believe we can't make AIs that be given the task to build software or robots to perform tasks we explain to them in natural language. It's just another 'game' after all.
Already they can make natural language descriptions of images... We should be able to have them do the inverse. We can teach them maths, then we can teach them software engineering.
It's not a question of philosophy, or even if, it's a question of when.
By zero shot learning, it was able to learn the task by only a few examples and extrapolated the rest from previous experience. So, on day four you tell it has a new task, and show it just one purple object, then it knows that the key is purple objects from then on.
Absolutely no different to how humans would deal with it.
A multimodal AI with enough modes and experience will not have the limits you are used to thinking about for AI. Think of using the same AI to control robots, drive cars, fly planes, make financial decisions, play chess, go and poker, manage a business, teach us new maths, design, build and maintain robots, software and better AI, make medical diagnoses, interpret law, etc, etc...
In fact, the whole deal with AI is finding the frontier of what computers can't do, and every time large numbers of people say it can't be done, and every time we smash through those frontiers. (AI will never beat humans at chess, image recognition, go, poker... etc... and now AI is better at all of these than the best humans... this trend will continue).
We don't know the "rules" that define the differences between a cat and a dog, yet neural networks can work them out through experience.
Nor do we know the "rules" that make up a winning go strategy (as opposed to the rules of the game itself)... yet AlphaGo managed to work those out itself too.
Everything is just games from some point of view. Eventually AI will be better at them than any human (or group of humans) can be (including at your job).
It's already happening... and AGI may well happen sooner than you think.
As a counter example you should look into multimodal AI.
https://research.googleblog.co...
This machine has different types of inputs, and different types of outputs, and the same AI learns many different tasks as opposed to just one very narrowly task.
Apparently this machine is capable of performing many different tasks at equivalent to where the state of the art was just five years ago. I don't believe it will take long for it to 'catch up'.
> Computers will be useful for specifically defined tasks, but they will neither recognize the boundaries of what they know or choose to stray outside their limits. They will always just respond as they have been programed to.
Now the real interesting thing about it is that in some cases it was able to use information it had learned from previous tasks in new tasks that it was given... and so was actually able to perform so called "zero shot learning"... it already new enough about the world to perform well on tasks it had never been trained to do!
So, as AI evolves, I think that AI will be very capable at tasks outside of its original training.
Go cannot be exhaustively searched the way chess and many other games can. The branching factor (number of possible moves) is simply too high.
AlphaGo is a breakthrough because it learns what is basically an "intuition" for good moves and board states by learning from examples and playing against itself, in a way similar to how we train networks the differences between cats and dogs.
> but we won't *ever* get a computer to think, feel and decide based on intuition.
That's the thing with AlphaGo though. The search tree for Go is simply too large to deal with in a traditional manner (minmax algorithm) the way we can with chess. Instead AlphaGo analyses the board holistically and basically develops an "intuition" for the best move.
The current state of the art image recognition systems are not built on rules, they "learn" the rules themselves from examples... In fact, we don't even have a good grasp on *how* they recognize the various images, just that they do. (For example, why does the network think it's a cat and not a dog... we just don't know beyond "it lights up these 'neurons' we trained with back propagation").
That's pretty damn close to decision by feel and intuition.
> But we humans MUST have some sort of work. We evolved to work and without it, we lose a sense of purpose and we resort to drinking.
No we don't.
If you WANT to work then go do voluntary work or whatever. If you don't, then why work?
People should really learn some microeconomics to understand utility.
Also, the point of UBI is to provide welfare for everyone, but without the trappings of welfare, like having to forego work in order to be eligible for it. With welfare, you can end up with less money for working, or at the very least suffer very high effective marginal tax rates. With a UBI, every dollar you earn is more money in your pocket... It increases the incentive to work compared to standard welfare systems.
At some point the robots will be better at engineering better robots than humans too.
There is no reason AI won't be able to design, build and maintain robots... and be better at it than human too.
As soon as AI can do the work of an average AI engineer, most human intellectual work will be obsolete.
The time to start thinking about UBI is now.
I know what you mean by regret. The market doesn't have regret, the stupid actors do, and the smarter ones have joy!
The market is already the best predictor we have!
Knowing the best predictor at any instant would pick the one that said the winner will win... likely the best predictor at any instant is an actual idiot. It does 'learn' the long term best experts.
You had it with exponentially adjusted weighted sum of experts.
I read a chunk off of your pdf... if you can't exclude users who were getting their information from actual exchanges to inform them of their play money bets, you shouldn't be surprised when they work... just bet the difference between the play money market prediction and the real market and you'll profit.
Useless internet points do provide utility even when they aren't money... but they don't provide utility like actual money.
> The claim is that a prediction market can act as well as the best participant of that market, i.e. have zero regret.
The market has no regret... some agents will, others won't.
> - If the claim were true, you could dump a lot (and I mean an extreme amount) of comparatively simple algorithms into a virtual prediction market, and by the claim, the market would act just as well as the very best algorithm of the lot, producing a super AI.
That doesn't follow... it doesn't do better than the best prediction on any given outcome... it's over time the best on average, You show me your exponential set of simple algorithms that contain at least one super AI, and the market outcome will be the super AI. It's just a matter of TPS for the market...
I think a virtual market type setup would, for example, find the long term optimal weighted sum of AI expert predictions...
I think why you wouldn't use a virtual market for training an ensemble, is that you have all the information already... you can condition an ensemble on the input... but in the real world we do not have that information, we just have our information, and markets create the incentive to provide that information, because we can profit from that it.
The real market is already a super intelligence... collating more information than any individual human (or not) expert can by being the weighted sum of expert human opinions already!
You have to show me your AI that can profit from what already exists first!
Funny thing is, they do exist and the best of them are thriving! Still I don't think we have any super AIs yet!
BTW: Swarm AI just seems to be an equal weighting of random 'experts'... and assigns it as win. That's what I'm saying, it can't possibly beat the market... unsurprising it failed to pick a 1 in 70k event... at least that's what the market thought, right? If anything it's a consensus creator...
See subject for more information.
> Suppose, for instance, that there's a prediction market on when the next act of terrorism will occur in the US, as defined by some criterion
Yeah, in this and the keynesian beauty contest example the outcomes are dependent on the market itself... so, you can use them to create assassination markets or can you get weird feedback loops.
So, where the outcome is independent of the market itself they work. You're not going to change tomorrow's weather by betting on it, yet that would be useful for many people, but the pope might get shot if the odds go long enough!
I would think that it should be regulated to that extent... I think people know the difference... other than that, why are they illegal for?
Also, play money doesn't work... it''s not money.