Bitcoin Forum

We've got the biggest supercomputer on the planet, and I think maybe we could use one like it for creating a rapidly evolving artificial intelligence.

What if the proof of work could include a random set of AI problems? A quick hash of the transactions would randomize a problem set generator which creates a bunch of random machine learning, pattern matching, and similar problems. The node which releases the most efficient script for solving all of the problems in that random set wins the coins for that block.

Difficulty would increase by creating ever-harder AI problems rather than ever-harder hash targets.

Solution propagation would be interesting, since you're looking for the most efficient solution rather than a solution better than a certain threshold, and testing a proposed solution would take a lot longer than testing a hash. I think maybe each node would have to broadcast its best solution generated or received, and after 10 minutes, the best solution would be frozen and work would start on the next block. Everyone who helped propagate the winning solution would also get a small reward.

Once a solution was accepted, the network would test a random selection of previously-accepted scripts and give an additional reward to the one which happened to do well on this problem set, rewarding flexibility.

Since every script would be public, there would be fierce competition to build on previous best solutions and rapidly evolve solutions to very difficult AI problems. "Efficiency" could be judged in such a way that allows multiple copies of a candidate script to run and interact with copies of itself. Whoever does the best job of creating a basic unit of intelligence would get the rewards - a race to "build a better neuron".

This is probably unworkable in some way that is not clear to me at the moment, but it sure would be cool if it worked. On the other hand, maybe creating something like this is a bad idea even if it does work :)

Artificial Intelligence is not a hardware/performance issue, it is a software/logical issue.

You could take all the computing power around the world, multiply it by a billion, then create a time machine, go back to the big bang, and let all that computing power begin crunching numbers.

Then fast forward until today, it would still not have solved even the most mundane task that the human brain could in a matter of minutes, like solving a puzzle of some hundred jigsaw pieces. Simply because we fail to design algorithms for these problems.

Citation needed!

He's right :)
You could solve specific problems, but you can't solve 'specialised' problems simply by numbercrunching. A computer doesn't do any intelligent shit until you explain him how to do it, and that's a matter of clever programming, not by mere calculations :)

I'm imagining millions of computers randomly mutating scripts, creating better and better "generic problem solvers" that can be multiplied like neurons.

Probably totally impossible, but maybe not?

Here's sort of a "reverse citation" in that it refutes poster shiv's claim. The life's work of John Koza: http://www.genetic-programming.com/johnkoza.html

If nothing else, I do like the idea of a coin which provides problems with value behind them while hashing, as opposed to the busywork currently handed out by all the existing chains...


Maybe tie it in to SETI@Home or something, if a pure AI is too infeasible.

I thought about that, but that requires centralization. AI problems can be deterministically procedurally generated based on the most recent transaction hash, so AI research needs no central server to coordinate things like SETI has.

Imagine: Decentralized AI research!

http://en.wikipedia.org/wiki/NP-complete

for example.

doesn't all the computing power in the world include human brains.  last I checked 2 + 2 = 4.  That sounds like a computation.

I think you are lost in the gravitational pull of the hyperbole.

Not completely true.  Genetic Programming solves problems using a vocabulary of operations, random generation of programs, testing, mutation and recombination of solutions using a large population of candidates. See ref: http://en.wikipedia.org/wiki/Genetic_programming (http://en.wikipedia.org/wiki/Genetic_programming) and http://cswww.essex.ac.uk/staff/poli/gp-field-guide/index.html (http://cswww.essex.ac.uk/staff/poli/gp-field-guide/index.html).  They've even experimented with FPGAs to optimize the testing/evaluation process for candidates: http://halcyon.usc.edu/~pk/prasannawebsite/papers/sidhuFPL99.pdf (http://halcyon.usc.edu/~pk/prasannawebsite/papers/sidhuFPL99.pdf)

I like the idea of useful PoW, but:
https://i.imgur.com/OMTExKp.jpg

reminds me of this experiment

http://primaryobjects.com/CMS/Article150.aspx (http://primaryobjects.com/CMS/Article150.aspx)

Is it possible for a computer program to write its own programs? In the previous article this question was answered with the creation of a program, using artificial intelligence, that could write its own programs to output "Hello World" and other simple strings. The AI computer program used a genetic algorithm to write child programs using the programming language brainfuck, a Turing-complete programming language consisting of 8 instructions.

+->,>,[>+,],,,,-<[.+<]
"dlroW olleH" ("Hello World" backwards)

The above programming code was created by an artificial intelligence program, designed to write programs with self-modifying and self-improving code. The program created the above result in 60 seconds. The above program accepts user input and prints the reversed text.

Not sure.

I think its time for a PrimCoin  ;)

primecoin? sure, that would be so easy to do, ask someone to modify the POW on the bitcoin client for looking for primes lots of open source prime programs out there just incorporate what they have learned into the Bitcoin clone the search for primes will auto level itself out so no need to worry if it's generating too many coins.

look at the GIMPS example, they got more than 50 million primes found with only 100 teraflops of processing power, I have 12 teraflops at home I could find all of them in a few weeks, so a balance system would be needed, cashout of coins according to the number of primes found every 10 minutes. find a 1000 per minute great you get paid out at 10,000 for 1 coin difficulty increased? finding only 10 primes every 10 minutes? then you get one every ten minutes, but the distribution would have to be distributed to everyone on the network so that people work collaboratively from the getgo.


now back to topic A.I. madness!!!

start small then, if all the work has been done then the only thing needed is to create the platform for people to contribute their processing power.

we start with a small subset of problems and to test the system allow people to contribute their processing power for free at the beginning. once the system is proven to work go to bitcoinstarter.com and make a petition for donations for the project .

I figure the best place to start is with the work that Alice bot has already done, just chatter bots working on language and slowly progress to audio/visual problems for the A.I to work on then increase the complexity to the limit till it starts working on it's own problems, viewable by all of course, with the simplified scripts we can start building a general frame work for transferring that AI logic to other platforms.

http://www.youtube.com/watch?v=DEtrzdGSXCU

I suggested something similar

An "IACoin" , but where the IA is using historical data of transactions/blocks + times + % fee +  user vote applied to transactions that is considered proportionally with the amount paid in the fee.

it should be easy to develop, but i can not hate more the mess of code implemented in all the coins.

This was discussed briefly in #bitcoin-dev. It's actually possible using game theory.

Use infinite prisonners dilemma of evildoers vs collaborators (basically you need to guess the others strategy mid-game). Whoever is ultimate winner in a round (everyone can replay the game to verify the "proof of game"), mines the block. Generalized poker.

Some notes:

* You need chips to play with - for the sake of simplicity no PoS for now, still use SHA256 pow as base currency. Big miners will have lots of chips to play in the round. However being rich is useless, if you don't know the intricacies of playing poker ... basically even small group of sha256 miners has a chance, if their game algo is much more sophisticated.

* Spiking NNs need massively shared memory, such computing clusters are forced to cooperate, (T=4 will beat all T=2 powerful brains, one by one).

* The fact anyone can be double (to N) agent is part of the game

* There might be problem that you might end-up with megacorp vs secondary megacorp (think Dems vs Reps and the rivalship is just theater), but i doubt it.

Thats why I was thinking that it should be a donation system, so many coins and it's security cannot be guaranteed.

To further increase adoption, users should be allowed to provide their own problems to the system with their own reward; this way if corp A needs certain problems to be worked on it can submit it to the network so people can choose to work on it for the pay rate.

But what if the IA becomes so powerfull, it performs an 51% attack on the network?  ;D