Cryptocurrency for AI Self-programming grid-network

[streondj]

Hi,

I'd like to gauge the interest of bitcoin community in a computing grid similar to bitcoin, and BOINC,  where instead of computing hashes, the GPU's generate computer programs which are ordered in the distributed marketplace.

I'm working on SPEL (speakable programming for every language), which is foundational for making GI-OS (General-Intelligence Operating-System), that is meant to be a OS that can be used on robots fit for reincarnating into.

Genetic Programming (GP) is a form of evolutionary algorithm which evolves computer programs to fit certain input/output criteria, so given a specification GP can write a program. GP is "embarrasingly parallelizable" so it is amenable to working on GPU's and FPGA's efficiently -- where as you know most of the cheap TFLOPS are.

SPEL and GI-OS is a large project of course, though by employing GPU's and CPU's "electronic-brains" to do much of the coding, it can be accomplished by a grid network of participating users. GP would of course simply be an initial algorithm used for writing these programs, various nodes could easily innovate with adding new algorithms to generate code, or having human-computer teams of programmers writing programs.

To incentivise it's use, and development, the "marketplace" would be in a bugzilla style format, where people could post bugs or feature requests, including one-off programs, and incentivise their completion by putting some coin as bounty for the bug. All of this bug marketplace would be machine-serviceable, so nodes could automatically search out which bugs might be appropriate for them or their pool to solve, solve them and submit the results.

The Coin, which we may call SPELCoin for simplicity, is planned based on stocks, so it may start with a low superior highly composite number (SHCN) like 720720, and then if/when the value of a coin becomes cumbersomely large then can split-stock, so each user that had 1 coin would still have 1/720720, even if the total number of coins doubled or quadroupled.

Since this project is for creating an open source AGPLv3+ operating system we with you may use for incarnating into robots/droids in the future, I'm thinking the "fruit of life" with it's thirteen circles as the diagram for splitting up the funds http://www.encroach.net/images/sacred_geometry/flower_of_life_geometry/fruit_of_life_61_circles_13.jpg

one of the circles would be for initial IPO, so 1/13th of the coins,  1/13th would be for the faucet. It would likely be a modified proof of stake, where transaction costs would randomly feed one of the circles, such as the faucet, core development circle or server-node circle.  
Likely would other than the marketplace, have a "project bank", where people could buy or sell coins from the project as a percentage of the money the project has, to make it easier for paypal friendly transactions. Various elements of the project would be in the various circles, so ideally by the time all the coins have been fed into the marketplace then GI-OS and beyond would be a reality.

The ideal would be for us to have self-replicating communities of robot host-bodies in ecological niches not suitable for liquid-based life,  so some of the circles would be for developing things like replicating robot communities in deserts (cold arctic, or tanezrouft (dry Sahara), or deep sea),  and at least one circle would be for colonizing other planets.

 Anyways I thought some of you may be able to help out with figuring out how to get the financing to work since many of you have much experience with such things, and both maidSafe and Ethereum are open source projects funded by cryptocurrency.

Do you think the IPO would be best done with MasterCoin or some other way?

Thanks,
Logan

[J1mb0]

both maidSafe and Ethereum are open source projects funded by cryptocurrency.

Maidsafe is vapourware until I am told different.

Etherium - you would be lucky to run a program to rival my Casio watch, let alone a neural net simulation big enough to 'reincarnate' into. 

[streondj]

Hi J1mb0,

Yes, it is true that maidSafe and Ethereum have still to live up to all their expectations.
I do not know  the detail of how they both managed to raise millions of dollars for their respective projects.
My hypothesis is that it had something to to do with interest in the ideas, and expectations of returns based on buying coins which could in future be redeemed at a higher value.
I am aware that Ethereum coin has had higher value recently, so at least some people may have seen a return on investment.

Unlike maidSafe and Ethereum which were funded based on concept whitepapers (to my knowledge), there are already several available prototypes of SPEL.

One of which I wrote last year in javascript as a proof of concept for its use as an interlingua based machine translator.  It is able to translate amongst around 20 languages with high precision in formal variants of the language -- which I have managed to verify that while giving "odd" output, is still understandable to native speakers of other langauges.  Also several of those languages I have prototyped with conjugation support, such as English, French, Russian and Spanish -- again a formal and regularized conjugation for simplicity and expressive power.  Additionally this prototype translates to javascript, and can run some basic programs. [1] Javascript however isn't designed to be GPU-compatible, and that codebase had made use of many prototype objects, which unfortunately makes it practically parallel incompatible, so I started on a new prototype for the new constraints.

Otherwise the SPEL proof-of-concept shows that it is functional as an interlingua, so should allow for most language speakers of the world to write code in their own SPEL variant languages, and have it understood in all other SPEL variant languages -- with the same precision as a computer programming language. The idea of having such a language is that all/most humans of the world (most of which DONT speak English), would be able to participate in making and using GI-OS.

For the vocabulary I've harvested a large dataset of over 24 different languages, and weaned down several international word databases such as (oxford-3000, special-english and wordnet), from a cumulative over 9000 words, to around ~2700 words, and generated accompanying word lists, dictionaries, and propose list for all the supported languages. [2]

Currently I'm working on a GPU-compatible virtual-machine for SPEL, much of the initial version of which is already written, with "hello world" tests having been passed a few days ago [3]. I'm currently refining the VM to be compatible with GP (genetic-programming) generated functions. The bytecode is an encoded SPEL variant language, where each word is represented in a 2 byte encoding, so all/most of the code generated by the GP/AI should also be human speakable and translatable to all supported languages.

As I mentioned at the start, this thread is to gauge interest in the project, and to see what elements of it you the public may like more than others. I feel it would be best to delay actually requesting funds until there is a minimal viable product, or proof of concepts showing that all the areas work. Currently since most other proof-of-concept areas are complete, the main one after the virtual machine is the genetic-programming,  which while having been proven in industry and finance for decades, hasn't (to my knowledge) yet been applied to writing standard libraries for programming languages.

However others have urged me that it is good to keep the community in the loop, to see what elements you may be most interested in.
So what elements of it do you find most intriguing?

Thanks,
Logan

[1] SPEL main tree https://github.com/elspru/spel/tree/master
[2] SPEL dictionaries for all supported langauges http://wyn.bot.nu/spel/src/vocab/gen/allLangs/
[3] SPEL virtual-machine https://github.com/elspru/spel/tree/master/src/virtual-machine
 

[psionin]

Hey streondj,

This topic is quite interesting to me and I have also been thinking about it.

A more AI based "intelligent" type of programming seems to be the next logical step in the evolution of computing technology. How often are countless hours spent on trivial bugs or configuration setting adjustments that block software installation for someone not intimately familiar with the code? Wouldn't it be nice if the OS could automatically select the earlier version of Python when a program fails to build because it doesn't support the later one? What about detecting changes in file locations and updating the code when a file is not found? This is difficult to implement with the current hard rule based structure of languages, whereas a more adaptive and learning system embedded into the OS could pave the way for the next leap in engineering productivity.

Wouldn't it be better to first design the meta language to work well with English, though? And add the translation later?

As for the "coin" IPO, I think that the coins are more financial in nature than technological, i.e., there's a whole slew of other unrelated services and products to build and maintain, and there are already so many of them. It's probably better to take an existing altcoin and to build on top of it. For example, one could get some XMR, and then build an advanced type of AI that can be used to populate the Crypto Kingdom, potentially raising the value of the XMR holding significantly with a successful implementation.

Litecoin simply adapted the Bitcoin mining process to work for GPUs only, instead of using the huge ASIC farms, and it seems likely that this innovation either directly caused or at least significantly contributed to its market cap going past a billion USD in 2013.

Plus, contributing to an existing coin can also gain the support of an already established community, long term holders and people with experience, which could contribute to more growth in the smaller stake than in a less distributed stake with a higher percentage of personal ownership, increasing the chances for overall success and prosperity.

I think it's also a good idea to think in terms of how multiple alts can be supported.

Good luck.

psionin

[streondj]

Hey streondj,

This topic is quite interesting to me and I have also been thinking about it.

Wonderful! :-D

A more AI based "intelligent" type of programming seems to be the next logical step in the evolution of computing technology. How often are countless hours spent on trivial bugs or configuration setting adjustments that block software installation for someone not intimately familiar with the code? Wouldn't it be nice if the OS could automatically select the earlier version of Python when a program fails to build because it doesn't support the later one? What about detecting changes in file locations and updating the code when a file is not found? This is difficult to implement with the current hard rule based structure of languages, whereas a more adaptive and learning system embedded into the OS could pave the way for the next leap in engineering productivity.

Yes Psionin I think you have the gyst of it.

One of the ideas is that can have "GP seed" code, which defines the specifications of a program, such as inputs and outputs.
An initial program could be evolved or written manually. Later if there are some updates in the supporting libraries, then the code could simply be "re-evolved" with minimal or no user intervention.

Wouldn't it be better to first design the meta language to work well with English, though? And add the translation later?

Yes, it always works with English, and the function names and variable of the virtual-machine implemented in C
is currently written in the English variant of SPEL. However as with any form of machine learning,
if the sample is limited to for instance just English, then there is a danger of "overfitting",
that is why when generating the vocabulary used 24+ languages,
from all or most of the major language families.

This ensures that all words have a 1-to-1 ratio in the base vocabulary,
so any word in the base vocabulary has a corresponding non-overlapping word in all vocabularies.

Additionally the grammar is based on the most common attributes found cross-linguistically, [1][2]
while having the potential to express any grammatical structure documented in human languages thus far,
by including all the glossing terms in the grammar. [3]

This ensures that anything that is grammatically expressable in another language is also expressable in SPEL,
though using some of the more unusual grammatical features may require those unused to them to look them up in a dictionary.

As for the "coin" IPO, I think that the coins are more financial in nature than technological, i.e., there's a whole slew of other unrelated services and products to build and maintain, and there are already so many of them. It's probably better to take an existing altcoin and to build on top of it. For example, one could get some XMR, and then build an advanced type of AI that can be used to populate the Crypto Kingdom, potentially raising the value of the XMR holding significantly with a successful implementation.

Yes, I think Monero/Crypto-Kingdom is a good model to go on. I view crypto-currencies more as crypto-shares, or crypto-stock,  so people that are buying XMR are buying stock in Crypto-Kingdom.

I can give an overview of how the SPEL/AGI stock will be split, see below.

Litecoin simply adapted the Bitcoin mining process to work for GPUs only, instead of using the huge ASIC farms, and it seems likely that this innovation either directly caused or at least significantly contributed to its market cap going past a billion USD in 2013.

Plus, contributing to an existing coin can also gain the support of an already established community, long term holders and people with experience, which could contribute to more growth in the smaller stake than in a less distributed stake with a higher percentage of personal ownership, increasing the chances for overall success and prosperity.

I think it's also a good idea to think in terms of how multiple alts can be supported.

In principal that is a good idea, and if I was some super wealthy whale, maybe even possible, but the reality is that developing a full fledged AGI (gi-os + associated hardware) capable of doing everything a human is able to do, is something that is estimated to cost in the trillions of dollars. If I had anywhere near that kind of money, I wouldn't be investigating appealing for funding.

Here is a layout of the circles or areas which the shares/stocks or coins of the project will be allocated:
https://joylifecoop.files.wordpress.com/2016/06/fruit_of_life_coin_distribution2.png?w=584

The core is for the language SPEL.  The inner ring is for the initial software development, and the outer ring is for the hardware development and beyond.

In more detail:

   1 language: SPEL, it’s virtual machine, librararies and translation support
   2 IPO/bank: the only coins available for sale by the bank, though bank can redeem any.
   3 fountain/welfare: the coins which are charitably given to encourage participation
   4 trade/servers: for marketplace (spelcoin) development, and paying server nodes
   5 brain root: brain stem;  kernel of GI-OS (general intelligence operating system)
   6 motor control: cerebellum; drivers and resource manager of GI-OS.
   7 choose action:  cerebrum; emotions, thinking and problem solving of GI-OS.
   8 body tech: hardware for living in settings similar to humans.
   9 cold tech: hardware for cold deserts, like Antarctica, Luna and Mars
   10 hot tech: hardware for hot deserts, like Tanezrouft, Mercury and Io.
   11 wet tech: hardware for wet deserts, like deep ocean, Europa and Ganymede.
   12 air tech: hardware for air deserts, like Venus, Titan and Uranus.
   13 space tech: hardware for space travel, interplanetary and interstellar.

The coins originally would only become available to those who solve problems in the area they are allocated for, collected in the form of bounties.  Afterwards they would be in that persons account unless they decided to trade them or use them.  Transaction fees and a possible hoarding tax would free up coins to be randomly redistributed back into one of the available circles.  If a circle is full then it wont accept new ones,  so those areas which are most depleted are most likely to receive the transaction coins.
[4]

Good luck.

psionin

Thanks,
Logan

[1] World Atlas of Language Structures (WALS) http://wals.info/
[2] Phoible Phonological Inventory Data http://phoible.org/parameters
[3] List of Glossing Abbreviations https://en.wikipedia.org/wiki/List_of_glossing_abbreviations
[4] Fruit of Life Shares for SPEL https://joylifecoop.wordpress.com/2016/06/07/fruit-of-life-shares-for-spel/

[MicroGuy]

Cryptocurrency already is "an AI Self-programming grid-network."

That's the beauty in the protocol. 

~~

[streondj]

Cryptocurrency already is "an AI Self-programming grid-network."

That's the beauty in the protocol. 

I don't see why you think that.
While bitcoin and other currenty crypt-currencyies are all grid-networks,
they don't use AI or any other form of machine learning,
in fact it seems most trading is done manually by humans :-|.
Also it is not self-programming, humans have to manually code it all.

Bitcoin "proof-of-work" is about solving hashes,
"proof-of-stake" is about providing blockchain services.
There is no AI in that at all. It's just a ledger, a fancy spreadsheet or accounting software.

With the SPEL marketplace, you could order a new crytocurrency coded for you, or some other app idea you may have,
for much cheaper than it would cost to have humans do the same task, also likely higher quality.
The majority of the code for the network and operating system would be done in this automated fashion.
That is what is meant by "Self-programming".

All of Gi=OS is an Ai really, and so is the method of programming,  but additionally you could have AI's decide which programs would be easiest to solve, and then solve them in return for coin.  Also could have a coin exchange, where your AI's could do high speed trading,  making it easier for you to make money on your various cryptocurrency investments.

Really manual trading is for amateurs, nowadays Ai's do the majority of stock trades.
Perhaps there is something like this in the cryptocurrency field, If so, please link to it, as I wasn't able to find it.

Thanks,
Logan

[GTTIGER]

Interesting

[spassbold]

Have you looked into Gridcoin before?

It uses Proof of Research on top of Proof of Stake, which means that the amount of coins you mine depends on how much you research/simulations you do for BOINC. BOINC is a grid-computing platform, that runs all kind of simulations/research programs, in terms of AI for example MindModelling@home, the goal of which is to simulate the human brain: https://mindmodeling.org/

So mining literally consists of simulating the human brain. In contrast to for example curecoin this whole mining by researching also happens in a decentralized way, there is no central server given out the coins, but it is build on top of Proof of stake and a so called "neural network", in which all Gridcoin nodes come to a conclusion which researcher has done how much research.

The second coolest project imo is Gridcoin Finance, that uses AI/machine learning/simulations distributed over the BOINC platform to find out which options to invest in in the stock market. (https://cryptocointalk.com/topic/11357-gridcoin-finance-project/). This is a project that is also already running and not just an idea.

Check the gridocin forum for more information: https://cryptocointalk.com/forum/464-gridcoin-grc/

edit: also Gridcoin is currently looking for a second dev, so if you can program and would like to participate, please let me know