Topic: Weak / Narrow Blockchain AI - page 4. (Read 5669 times)

The Ai is meant to be decentralized, but to work within given rules. Each node acts like a neuron, broadcasting data to others over the network.  Yes, i think i will allow each Ai node to be named by the owner, however i will enforce versioning to prevent attempts at manipulation.

There are currently a couple of sticking points that i am trying to solve hence the delays in releasing an alpha version.

1) random forest decision trees (was using a crap implementation, decided to go pro) for individual nodes and better simulations
2) how to put together a decentralized neural net that can tolerate nodes coming and going ?


I am curious though, do users prefer PoW or PoS ? The code is capable of both or hybrid.

I agree with you, sometimes when you hit a limit, you better go check something else, i use to find solutions while doing something else, totally different.
Even i have never coded an AI, i can imagine how diffuclt and tricky it can be, keep it up, i know you will finish it, but just a question, will this AI be carried in all the wellets ? or will it have a central brain ? will users be able to rename the AI ? or provide some modifications to it ?

How I understand! :-)
Even basic hashing algos can be brain melting, at times ;-)

Hi

Thanks for reminding me, I've been working on it since monday and have made great progress. I was running scenarios yesterday and the predictions were ~ 40 accurate, i would like to bring this up to 85 % before release.

Due to the brain melting business of AI and simulations, I take breaks and use the time to pursue other ideas and goals such as wallet mods and other items i find interesting.

I agree with pallas, a plain background is better then a transparent one, maybe you can customise the collors to reflect the project, a black/darkblue gradient may be appropriated regarding the project's aim.
Had you you been able to work again on the kernel ?
I like why you did regarding the wallet generation, it is true that customising more than the first letter in the wallet address is a nice update.

Maybe because I multi task a lot and needed to be able to monitor the wallet, i found it useful. Here's what i will do:- I will keep the default and add optional transparency.  

ugh.

plain wallet please. that see through background is distracting.

My 2 cents: don't mess with the UI, leave it to the default.

I think thats pretty neat and its another first in crypto currencies right?
I KNOW WHAT I'M ABOUT TO SAY IS NOT REALLY TOP OF THE LIST IN DEVELOPMENT you could also add style sheets so people can choose their own wallet colors.

I was having difficulty getting focused over the weekend so i decided to do something a bit easier:-



Do you like it or is a see through wallet a bad idea ?

Now trying to figure out how some of the changes I made to how the AI operates can be propagated across a network in a timely fashion. Not yet confident at real time self-correction , so will start of with once a day and increase the frequency accordingly until we reach a safe but viable number.

Hi

Thanks, i already can confirm full and uninterrupted headers sync for PoS. The AI is being developed to be completely decentralized. If you download the client, you have a "AI node" that contributes to the whole AI by running simulations, submitting best found solution and comparing with others, culminating in a vote by active AIs to select the most popular strategy.

I will release a test version in a couple of days when i finish up with the kernel. 

Hmm,

Interesting Topic, if you can get headers sync with PoS working that would be valuable to many.
Pandacoin tried one , but the blockchain would keep corrupting on them , so they use the original method instead.

Regarding Negative Interest Rates, it won't work, you can have the interest fluctuate all way down to 0 ,
but igotspots tried one that actually deleted coins, pissed off the whole community last I remember.

You could probably get away when playing with a fluctuating fee system as a way to decrease the amount of coins.
If that is something you wanted.

By placing it on the Oracle AI, is that a centralization or will there still be separate windows or linux wallets.
Nice to see some new ideas. 


 

So i got bored and tired of the harder work on the kernel, decided to pursue one of my pet projects and managed to apply it to the testing client. The Oracle client now produces custom wallet addresses.

Most wallets only specify the first character, eg Bitcoin with "1" litecoin with "L" . I thought why not make a really distinct prefix for example the Ticker code for the tokens.

A early morning test case worked : -



So once i settle on a final prefix, all wallets will produce addresses with the custom prefix by default.

I'll wait for you in this project and inform successful for you.

I've since resolved most of the problems with PoS and am now totally focused on the AI kernel. There is still some signing error on some stake transactions, but i will retrun to that later. Expect a testing binary release soon. Binaries only, no source.

Headers first does indeed complicate PoS 

I would be happy to participate to this project, if you need anything contact me, i may provide a lot of support depending on what you need. Feel free to send me a PM. By the way, if you need a team i am your men.

i've been waiting for this, i completely understand why you would want to keep the source closed. i'd like an invite, how do i go about that??

An overview of the design is laid out in the technical paper :- ORACLE AI TECHNICAL DRAFT


Oracle AI is an Intelligent Blockchain that analyses, self-corrects and self-regulates rather than the hard-coded system behavior typical of bitcoin and clones. The AI acts as a guardian for the blockchain, monitoring it and adjusting aspects of the system to ensure viable conditions are always maintained. The Ai takes over control of all system policies, such as block reward, proof of stake rewards, fees, money supply management etc. The AI will self-evolve beyond the initializing parameters using a combination of Trial & error and Build-Evolve-Destroy.

I have decided on ten goals/ steps in my road to achieve this kind of Expert System/Narrow AI

1) Blockchain
2) Data Acquisition
3) Kernel v1
3.1) Learning (Not sure if i can teach it to teach itself)
4) Simulator
5) Self replication
6) Inter-AI communication
7) Network wide AI (each individual AI rather than being independent, operates as part of the whole, each nodes vote weight is based on # of iterations and correct previous history )
Reasoning and Explanation
9) Extended Simulation
10) Cross-Chain interaction

I am likely to keep this closed source, i cannot see a compelling reason to open the source to others , especially with the prolific cloning i see around here. For example, no-one has bothered to update the basic Proof of Stake code since 2013/2014, so i had to take bitcoin 0.12.x (2016), update it to use c++11 and start baking PoS into it from scratch. I am currently testing a very basic version on my private home network ironing out bugs, while i write the data acq and kernel code. I will release binaries to curious, interested or enthusiast parties for a test run in a few weeks, if successful, i will proceed, dedicating my free time to meeting the above ten goals and/or any others that may arise during development. For those who will potentially be invited to participate in the private test run but are wary of using closed source code, all i can say is, get a VM and sandbox it.

Regarding my choice of 0.12.x  over the newer 0.13 version of the bitcoin codebase:= Segwit is thus far untested and is not even used in the bitcoin protocol yet. That coupled with the PoS code and many other issues such as CLTV and CSV which no-one has explored in PoS, leaves me with more than enough sleepless nights.

As the code is based on Bitcoin, there will be a token, however if this proves unsuitable, i may opt to develop a new code-base altogether and tack on the previous balances. Services are still just conceptual at the moment, and the level of multi-threading, ram and processing requirements would probably end up making service hosting a very costly thing. Service will be able to access the AI via API and each unique service instance can be propagated across the network.

I am dreaming, this is one of the projects i mostly dream to participate to, at last someone had the idea to develop an AI blockchain, could you please give more details about the project, how will it evolve ? will the code be open sourced or private sourced ? will it be a coin or service driven ? what do you expect from it to do ?