Bitcoin Forum
September 23, 2019, 08:26:17 AM
 News: If you like a topic and you see an orange "bump" link, click it. More info.
 Home Help Search Login Register More
 Pages: [1]
 Author Topic: Theory of intelligence as a form of Hedging  (Read 745 times)
BenRayfield
Sr. Member

Offline

Activity: 316
Merit: 250

 May 13, 2014, 02:45:53 AM

What is hedging?
Hedging means to balance risk where you are not trying to apply force. For example, before attempting an axe murder, you make sure your axe is sharpened to balance against the risk of injury to yourself (due to a delay) which really has nothing to do with your goal of murdering Hitler. Or before most are allowed to drive a car, we must buy a reasonable amount of insurance against if we hit someone, so we can accomplish the goal of driving without being as much affected by if we crash or not. You could also say the law to buy car insurance is other people hedging on our crashes too, that they do not want to be affected as much by such a crash. If only Hitler had hedged against his loss of popularity due to murder and stuck with his goal of improving the world by selective genetics, he might have convinced people to choose different people to have kids with. Maybe. The point is balancing risk goes far beyond existing theories of economics and is core to all intelligent behavior.

The simplest way to measure Turing Complete (general) intelligence

http://en.wikipedia.org/wiki/Rock-paper-scissors is an excessively complex game that uses 3 options instead of 2. It can be played by 2 players after agreeing which player will win if their choices EQUAL instead of OPPOSITE, for any number of rounds in this game, choosing YIN or YANG. Either way, repeated plays of the game let the players get a feel for eachother, and over time its a game that a supergenius will beat a genius at slightly more often than randomness would allow. There is no upper limit on the amount of intelligence Rock Paper Scissors (or its 2 option form) can measure by comparing 2 players at a time statistically.

Now lets expand that to any number of players. Each player moves their mouse left or right and looks on screen to see what direction they should have moved in the last fraction of a second to win the previous round. The winners are those who move in the MINORITY PREDICTION, which is the prediction of left or right that is made by less of the players that round. If theres an odd number of players there is always some MINORITY PREDICTION. Each person may join the game as any number of players, for example their mouse left/right movements and mouse forward/backward movements as 2 players. We may extend the game to continuous/scalar numbers of how far you move the mouse, but it works at its simplest when everyone is just a single bit like YIN vs YANG or LEFT vs RIGHT.

If many people are moving the same mouse cursor, they would learn to HEDGE against the mouse movements of many other players by forming a mental model of those players which allows everyone to stabilize the shared mouse cursor, or at least during the times they are trying to move it a certain direction.

Consider what would happen if we put such a shared mouse cursor into a game, like Pac Man, and more of the people had to move together in sync else get eaten. Do they learn to predict where others will push so they can push opposite of that direction plus where they think the pacman on screen should move next? Or do they move back and forth randomly like a stock price in equilibrium?

How about a simpler model? All 7 billion people find themself moving 1 shared mouse cursor between a few icons on screen which say "launch nukes" and other parts are empty space on screen (not that we would ever let this happen)... Do the people learn to navigate the shared cursor through the empty space even if its a complex shape? Don't they have to HEDGE against the other players' movements that would otherwise jump around chaoticly?

I propose the theory that this is all the same as Hedging on Rock Paper Scissors, and we can test any theory of economics or intelligence on a group of volunteers willing to play such games or sufficiently smart artificial intelligences but I dont think we have any that can fill Humans place just yet in such a flexible way. Any system which wins the game more often than randomness would allow is more intelligent than less.

How can we use this in a real world experiment? How might we get something started in a context (what kind of game like user interface, what do the players do, etc) that would motivate expansion into more advanced forms? How can we proceed as a species into these many possible futures while hedging against any specific economic models which happened to work for some time but are not founded in a mathematical definition of intelligence?

1569227177
Hero Member

Offline

Posts: 1569227177

Ignore
 1569227177

1569227177
 Report to moderator
1569227177
Hero Member

Offline

Posts: 1569227177

Ignore
 1569227177

1569227177
 Report to moderator
Advertised sites are not endorsed by the Bitcoin Forum. They may be unsafe, untrustworthy, or illegal in your jurisdiction. Advertise here.
cbeast
Donator
Legendary

Offline

Activity: 1736
Merit: 1002

Let's talk governance, lipstick, and pigs.

 May 13, 2014, 02:59:22 AM

I probably would have started with the Rock, Paper, Scissors rather than the ax murderer analogy, but I get ya. Nuance is important in game theory. Understanding human patterns is what competitive games are all about.

Any significantly advanced cryptocurrency is indistinguishable from Ponzi Tulips.
twiifm
Hero Member

Offline

Activity: 784
Merit: 500

 May 13, 2014, 03:02:34 AM

Sounds like you wanna read up on game theory and Nash equilibrium.

But what you are are describing isn't a hedge.  A hedge is when you take both sides of trade.

For example your car insurance story.  Insurance is just insurance and not a hedge.  But suppose you wanna frame it as a trade.... The insurance company is betting you won't get in accident.  If you crash they lose the bet by paying out more than they collect in premium from you.  If you don't crash you lose the bet each month paying premium.  A true hedge is if you bet someone else you wont crash and each month you don't crash this person pay you cost of your premium.  You are hedged in a sense that whether you crash or not you never lose money
BenRayfield
Sr. Member

Offline

Activity: 316
Merit: 250

 May 13, 2014, 04:13:59 AM

I understand Nash Equilibrium, any state of a game where all players have no reason to change their strategy.

Hedging is the balancing part. Risk taking is whats left minus the context of what was hedged on both sides.

cbeast, I am, among other things, an artificial intelligence programmer trying to understand the most basic properties of intelligence in general so we can speak, email, and paint it to eachother, whatever form the intelligence takes. Dunbar's Number, the number of people/organizations each person can keep track of the relations between, is estimated around 150, far below the 7 billion we need it to be. To fill that gap, we can afford no extra layers or complexities in a theory of intelligence, such as the unjustified belief that we are separate life forms instead of a statistical bio computer which memes flow through.

Cryptofun2.0
Newbie

Offline

Activity: 11
Merit: 0

 May 13, 2014, 04:23:28 AM

Quantum physics -  observer shapes reality
twiifm
Hero Member

Offline

Activity: 784
Merit: 500

 May 13, 2014, 04:41:29 AM

What do you wanna achieve your left right mouse game?  Is it like prisoners dilemma?  Each player has 2 choices, and they want to be in the minority to win?  Seems easy enough to make this game.  But what will the data tell you?

I still dont think you are using the term hedge correctly though because you aren't quantifying risk.

BenRayfield
Sr. Member

Offline

Activity: 316
Merit: 250

 May 13, 2014, 04:51:41 AMLast edit: May 13, 2014, 05:27:17 AM by BenRayfield

It has been observed in physics that when a particle may have come from 2 different pools of particles and theres no way to tell which (physically lacking that information, not just ignorance), the 2 pools are entangled as a result. Could this be extended to the "observer" putting force toward being less aware of which pool it came from (if it was between entangled and not, or say more entangled by more probability-amplitude on the ket-vectors of the combinations)? The makers of Quantum Radar thought so, even though they say their experiment didnt turn out well, I think they didnt take it far enough.

There was a TED video, I forget what it was called, where everyone in the audience had a button they could hold down or not at any time, controlling 1 pixel on the big display everyone could see. Acting toward the same goal together, said by someone on stage, they were able to quickly form basic shapes and letters. This is the most misunderstood ability we have. Why hasnt it been further explored?

I want to create a Nash Equilibrium where the best strategy for everyone is to play game-theory games through the Internet, with the mouse or other forms of interaction like brain chips or Emotiv Epoc mind reading game controller, whatever form of vector input and output in general, and most importantly these vectors will amplify eachothers prediction ability by helping people to predict eachother recursively, so when some of the players watch movements of wind in the weather, stock prices, or anything needing prediction, the network would pull prediction from other people like neuromodulation.

Neuromodulation is simply the difference between my mouse movements predicting yours and yours predicting mine, whichever short term memory gets Nash Equilibrium'ed into place temporarily, the supply and demand of information flow between neurons, some acting out of surprise (See Jeff Hawkins "surprise goes up the hierarchy" of neocortex) and needing more info and others acting toward that surprise with hedged stability to push against.

This is why I need to understand games of many people moving the same mouse cursor. Its literally neuromodulation except between people instead of just brain cells, but of course brain cells will do it while playing the game, as if we were all one big brain.

Quote
I still dont think you are using the term hedge correctly though because you aren't quantifying risk.

In such a game, left/right of the mouse is like buy/sell, driven by whatever could motivate the players to try to move like the MINORITY PREDICTION as displayed on screen at the time. The risk is the other players moving their mouse in a way that puts your prediction in MAJORITY so you see it move opposite direction on screen and you lose that round, so you would lose influence among the other players who are trying to be more like those who are able to predict in the minority more often. The hedging is if the players would learn to mostly half of them move left while half of them move right, while some of them move as needed to get some other game actions done with such movements. A player would hedge against others' mouse movements by finding a way to get the total mouse movements to go a certain direction regardless of what the other players are doing, by having influence on them at the time through the subtleties of this dancing together with the mouse. The stock market is like a dance, but you might not see it that way if it takes months or years to make a single move. I'm going for something more realtime as it works on all scales.

Whats stopping me is how to remove the players (theoretically not built yet) which do not add anything to prediction ability because they just do something random or copy others or hook in a random number generator and spam the network in the worst case.

There is an important part of math... A bunch of random bit vars, observed many times together, has an XOR between each pair of equally as often 1 or 0 (the XOR finds opposite or not), but if you do that on a group of bit vars held to have the same number of 0s and 1s while being random in all other ways (like you randomize them then see how many 1s and 0s then choose some random bit vars to adjust toward standard deviation 0) then you get XOR of slightly more 1s than 0s (they are opposites more often because once you observe one of them the others are known to have more of the opposite bit). But if you take the XOR between 2 groups each of standard deviation 0 within the group, its still balanced XOR between them, unbalanced within either alone. This is what the MINORITY PREDICTION is based on. Your prediction is 1 of the bits, and the others must have slightly more of the other direction to cancel.

I need a Nash Equilibrium between all these mouse movements, but I also need a money-like system to reward good predictions so the randomness gets overpowered. Bitcoin's "proof of work" design can provide seed money, in a very basic form maybe just do a SHA256 until you get a few bits of improbability to allow a network connection, and gamble on from there.

But its not even that simple. I need it to branch into many games of this MINORITY PREDICTION since just 1 would leave a very boring game where you took commands from many others but had very little influence on your own. Instead it needs to flow together as many smaller games partially intersecting, like a bunch of stocks vibrating and related to eachother.

So I know how to measure prediction in the simplest case of mouse movements in one game, and a few other cases in artificial intelligence, but these more advanced forms I'm not sure how to proceed.

twiifm
Hero Member

Offline

Activity: 784
Merit: 500

 May 13, 2014, 05:15:20 AM

Sounds pretty cool.  I trade options and I'm really interested in the behavioral economics aspect of markets.  I like to study bubbles.  I try to discover assets that have formed bubbles and compare the shapes of their charts.  So far I found one shape of Apple 2012 bubble.  I found at least 4 current stocks that has formed a similar shape.  Tesla, Yelp, Facebook, I forgot the other one could be Netflix or Amazon.  This interesting thing is that these tickers have no correlation to each other and that the bubbles occur at different times yet they form similar shape in similar timeframe

I've been reading Robert Shillers works on irrational exuberance which is taken from Keynes "animal spirits" ideas.  Would interesting to create a game experiment to test if bubbles are indeed psychological
 Pages: [1]