Alive - but still mostly in the early whitepaper/idea stage.  The design on the wiki is still evolving and we are working on a demonstrator of some of the key initial tech (the neural engine).  We are going to slowly spread awareness in a very limited fashion to targeted groups, refining the wiki along the way.

This is all leading up eventually to a main website launch which will have a better intro to the idea, a video or two, press release, etc.  The neural engine itself - when appropriately demonstrated and validated - will also generate press, probably timed with the main website launch.

Graham - what probability would you assign to AGI-5 in 5 years?
0.1%? 1%? 10%?

Five years ago, it looked like human level visual recognition was impossibly far in the future.  Many said that hitting that performance bar would require fundamental new insights into AI.  It did not.  All it required was minor improvements to existing ANN algorithms - combined with scaling up to appropriately large amounts of neurons and training experience.  There are still challenges in vision - video, 3D object localization, etc - but almost nobody in ML now believes that solving these subproblems will require huge new insights.

Likewise five years ago, few believed that ANNs could be used to build a complete complex AGI.  Pattern recognition is important, but there is much more going on in brains - complex sequential planning, attention, episodic recall, etc etc.  Many believed AGI would require a mix of different algorithms.  Then the deepmind guys trained an ANN to play atari and showed that you could achieve human level play across a wide spectrum of games using deep reinforcement learning across an appropriate initial architecture.

All of this has been achieved by small teams of researchers running models with 10 million neurons or less on just a few GPUs.  It has actually required new insights to get this far - many of them in fact - but all fit within a general optimization theory of intelligence that has been solid for a while.

We can achieve AGI-5 by:
*scaling up to millions of GPUs each running billions of neurons (embedding in a video game, for example)
*scaling up the number of researchers to thousands, augmented with new global opt/research automation tools
*creating a new framework to organize the code/design exploration process and manage incentives.

The above combination will accelerate research by a factor of about a million, and will lead to a reasonable probability of AGI-5 in 5 years.

Accomplishing the above will not be easy, but there is literally nothing else more worthwhile.  Even if it only has a 1% chance of leading to superintelligence in 10 years, 1% of a 100 trillion dollars (lower bound on the value of superintelligence) is still a trillion - and justifies almost any investment.

Everyone is making a bet, like or not.  If you dont invest in AI (and in the right project) in the next 5 to 10 years, you will find your wealth/job/livelihood eroded away regardless.

Higgins, I've read through some of your posting history and I can see that you are an intelligent - and normally civil - guy.  There's no need to get angry.  


How much computational neuroscience literature have you read?  10 hours worth? 100 hours worth?  1,000 hours worth?  I only ask this to speed up the conversation.  Your reply of 'bollocks' indicates to me that it is unlikley that you are up to date on computational neuroscience literature.

The general uniformity of the laminar cortical architecture has been a central key finding of neuroscience since the days of Hubel and Weiss's discovery of columnar maps in the primary visual cortex of cats.  The same general 6 layered microcircuit with its characteristic distributions of excitatory and inhibitory cell types and initial wiring patterns is repeated throughout the cortex of all mammalian brains, with varying degrees of small differences across cortical regions (motor and frontal cortical regions have some key differences) and between species.  At the higher macrocircuit level the cortex is partitioned into (mostly) genetically predefined modules, each of which repeats the same general wiring strategy but has unique intermodule connectivity.

The most significant and striking direct evidence of cortical microcircuit uniformity comes from intentional or accidental rewiring experiments.  If the normal 'auditory' cortex is rewired to receive input from the retina, the microcircuit develops the 'software' wiring for visual processing instead (although - as expected due to various reasons - not at the same level of capability).1

The detailed microwiring changes dynamically during development and learning - encoding acquired mind 'software' if you will - and thus ultimately is unique per individual.

In blind humans, the 'visual' cortical modules typically develop advanced audio processing capabilities, including even active echolocation (sonar) in some cases. 2 . 3

The general uniformity of cortical architecture leads to the 'one learning algorithm hypothesis', the key insight behind 'deep learning' (just another name for large cortex inspired ANNs), and the success of that field is strong indirect supporting evidence for the principle.  The same general deep CNN architecture that has now achieved primate/human levels of performance in vision is also leading to similar breakthroughs in a wide variety of other pattern recognition challenges such as speech recognition and automated translation.

See this video by Andrew Ng, it summarizes the importance of the universal cortical learning hypothesis and how it has (and is) revolutionizing machine learning.

Also - just to be clear - there is still a great deal of research to do in learning algorithms.  The brain uses a complex mix of supervised, unsupervised, and reinforcement learning, and in general its learning machinery is more complex and powerful than anything we have yet in DL/ML.  

Even so, our current learning algorithms are already powerful enough to rival isolated equivalent brain circuits in some cases - at least for tasks that are simple enough to construct an ideal training environment.  The 'commonsense reasoning' that you mention below will probably require embedding an AGI for virtual years in a complex simulated 3D environment (running the simulation much faster than realtime is key - its a performance issue).

...


Performance is constrained by the complexity of ANNs in terms of size and speed.  It takes about a few years of visual experience for a state of the art DANN to achieve human level performance, which requires running the simulation dozens of times faster than real-time so that training takes a more reasonable few weeks.  Everything is performance constrained: with more compute power we can train and experiment with ever larger ANNs.  Model complexity is thus increasing exponentially concomitantly with performance.

...

...

Human level general intelligence will require very large ANNs with complexity similar to that of a human brain, along with a similarly complex and lengthy educational development process.  Essentially we will need to actually raise and educate AGIs - somewhat like children.  The key is accelerating the learning speed, which is already essential for training visual circuits.

There is still a great deal of research to do in unraveling many of the brains macrocircuits and principles.  On the other hand, we also have a huge backlog of theories and wiring data - far more than we can test.  Massively accelerating experimental throughput is the key to accelerating progress.



The exact timeframe is irrelevant - the question is simply to gauge your current general intuition for where the principle locus of uncertainty/difficulty in achieving AGI resides.  Is it in getting to infant AGI? toddler AGI? adult AGI? etc.

For example, once we achieve AGI-5 (toddler AI), it is still uncertain that AGI-30 (adult) is near.  It could be that there is a huge remaining complexity gap.  I doubt this - instead I believe most of the complexity/difficulty is in getting up to AGI-5.  In that - at least - it appears we agree.



Now I sense that you are intentionally misreading me.  By 'simulating an infant brain', I meant simulating a brain the size and complexity of a human infant - without yet understanding the correct seed wiring architecture.  It would be a brain dead infant, essentially.  There is still a huge amount of research work to do going from "now we can simulate an infant brain" to "now we understand all of the initial wiring/algorithms that infant brains need to become adults".


A while back the AI/AGI community split into a few directions. One group believed that AGI could be achieved by advances in logic and traditional computer science.  Another group believed that AGI will require artificial brains - based on understanding the learning mechanisms in real brains.  The former group's thesis has essentially been disproven and is a dead-end, and all advances in AI at this point are actually coming from the latter group (neuroscience/machine learning).

General 'commonsense' reasoning requires a general internal predictive model of the world, which is something that comes around or after AGI-5.  Its not an isolated problem you can solve independently before hand - it's essentially AI complete.

Cheers

-Alix

I think your idea is interesting, the rental idea is orthogonal and could be combined with it.

I agree that in the current market, renting half of the hashing power for Bitcoin would be really tough.  But that is only because the Bitcoin ASIC market is inefficient in the economic sense.

The rental attack idea probably already works - and may have already been used - for GPU mining coins.  Even there though there really isnt yet an efficient market for GPU computation - at least not yet.

An efficient market for GPU computation would be the situation where almost everyone runs some smart app on their GPU which is intelligent about earning money through any means - renting out compute time for any of a wide variety of coins/projects/etc when they are asleep/away, but only when the pay is worth the power cost, etc.

I think that market efficiency will tend to increase over time.

That article is not really a whitepaper and was misnamed: it really should have been called "Brain Hardware Parity" and now is.  The actual high level network architecture for decentralized/distributed mind simulation is here: http://aishare.co/wiki/Cryptagoric_Cortical_Networks (warning: work in progress).

As to the "tinfoil hat grade", I'm not quite sure what you mean by that: this has nothing to do with conspiracy theories, it's just a basic review of recent computational neuroscience focused narrowly on performance comparisons.  We now have ANNs that can replicate some specific cortical functionality very well (visual recognition), and we can use that - along with various low level metrics - to estimate the performance necessary for replicating all of the cortex, as it has a uniform architecture.


As a field, machine learning is already making steady exponential progress.  You can ignore the evidence for that progress - that's up to you.

This is simply a plan to greatly accelerate ML progress through a combination of:
* more efficient use of local compute resources (GPU neural simulation efficiency)
* an efficient distributed/decentralized architecture to leverage a wide network of GPUs
* a more efficient incentive structure

I consider myself a rationalist, and you probably do as well.  The essence of rationality is updating on evidence.

I get the impression that you assign a low probability to the general possibility of near future AGI, and or this proposal in particular.

So, what future evidence would cause you to update that to a high probability?

Say in 5 to 10 years, you hear about an AI 'app' that has true adult human level intelligence - like Samantha from Her.  You try it - it works, and it appears to be smarter than anyone you know.  Do you still not believe?

Say in 5 years, you experience an AI 'app' that has intelligence - but only that of a 5 year old.  How much would this evidence shift your belief on whether adult level AI is near?

Say sometime later this year, you read about some new neural simulation engine in the news which can simulate a full human brain equivalent with billions of neurons on a single GPU.  The code/design is proven and its exciting to researchers, but initially its just simulating infant brains that don't do much, and certainly aren't interesting yet.  How much would this shift your belief on whether 5-year old AI is near?


That latter quote was sincere - not sarcastic - and any offense was unintentional.  I actually really do want feedback and criticism.

Thank you for your feedback.

Cheers
-Alix

I somewhat misunderstood you earlier.  As you say, a pure simple PoS that inflates in the way you describe is in fact pointless, as it doesn't really perform any redistribution.  The bond example I gave earlier is a form of redistribution from short-term holders to long-term holders.  Its probably a good mechanism to have, but its not sufficient by itself as a redistribution strategy.  You actually want the new coins to go to those who work for them in some fashion.  PoW does that - albeit the work is pointless busywork.

Of course it would be better if the work done in exchange for new coins actually increased the long-term value of the coins: an idea I argue for at length elsewhere.

Agree with much of your other points, but using linear historical comparisons for timeframes can be misleading, due to various vaguely exponential technological acceleration effects.


Yes, but ecological disaster is perhaps not the best way to look at it, for even if the ecological disaster were averted, its still a giant economic malinvestment.  It creates a bunch of hardware and computation that doesn't actually lead to technological/economic growth.

The other problem with the whole "bitcoin as reserve currency" idea is the incompatibility between the bitcoiner ideal of a currency and world free of taxation, and the practical necessity for governments to fund themselves.

For example, consider a world where bitcoin became the reserve currency, and then - hypothetically - a giant asteroid was found on a collision course that would destroy the world in the next 5 years, unless we could spend 10 $trillion on some massive spaceship intercept missile program.

Now, in a world where governments can arbitrarily commandeer 50%, or 80% of the GDP, the disaster can be avoided.  But in the ideal bitcoiner world, you have a problem.  In reality governments aren't going to just give up and die.

So given all that, the more realistic scenarios would involve massive tax increases in other areas to compensate - such as much higher property taxes, or whatever.  The end result is you still have something similar to the present wealth tax - just in a different form.

Imagine, if you will, a forum with actual intelligent discussion - instead of endless spam variations of "you're a troll!".

You say you've digested what I've said, but in the linked posts I provided earlier I made a very strong argument (I would say a knockdown argument) that no PoW or PoS based cryptocurrency design can possibly be coup-complete.  Coup-completeness requires immense value creation orders of magnitude beyond what existing designs can offer - indeed far beyond the value of any transaction system.

The dollar, obviously, has already accomplished coup: it became the world reserve currency.  It did this because the united states became the world's superpower, by actually winning several wars against its competitors on the strengths of superior net technological intelligence (the source of economic power).  The dollar reflects the value of the US economy.

Proof-of-Work (PoW) currencies like Bitcoin and Ethereum are like distributed governments that have very high early inflation which eventually tapers off, and they use this seniorage income to subsidize enormous worthless hashing computations.

Reductio ad absurdum: Imagine, for sake of argument, that a PoW currency was universally adopted by the citizens of a nation - as Bitcoiners dream.  The national currency would then be abandoned, and tens of trillions worth in wealth (for US or Euro sized examples) would be transferred from the current 1% wealthy elite to a new 1% (or more likely, 0.1%) elite.  The new elite would be composed of some mix of two groups: 1.) currency investors, and 2.) miners.  Crucially, wealth is simply transferred from one group to another, without actually creating new wealth in proportionate trade - as in new valuable technology.  

From the perspective of the entire nation, this change is at best a marginal economic improvement, and adds significant new increased costs for tax collection, enforcement, etc.  It's simply a wealth redistribution that favors anti-government libertarians (early investors) at the expense of pro-government liberal types (late investors).  You may have better luck attempting to just pass legislation that redistributes all the dollars to libertarians.  Those cryptocurrency advocates who naively think that everyone else will just 'wake up' to the obvious superiority of the new regime are suffering severely from the mind projection fallacy.

In a capitalist system, wealth distribution goes to those who provide value.  We've had many partial coups - Google is worth $600B (3% of the entire stock market) because they actually created a bunch of new technology to dominate markets - they actually created value.  A real coup requires something about 30X more powerful than google in terms of value added to the economy.  Hashes are not valuable (PoW).  Neither is passive rent-seeking (PoS).

Could a cryptocurrency be coup-complete?  Yes - but it would have to also function more like a crypto-stock, where all of the new wealth redistribution actually creates an enormous plethora of immensely valuable new technologies.  A PoW coin that sucks in $10 trillion doesn't create anything useful but $10 trillion worth of useless mining hardware and waste heat.  If you instead spent $10 trillion on general AI, you could potentially takeover the world: because an AI economy is actually worth that much (and more).



You are now arguing against a strawman model of our future vision - one unrelated to our model.  I ask instead that you attempt to steelman my arguments, and I'll attempt likewise.

My actual model - in simple form for our purposes - was defined earlier as:


I'd like you to actually answer those questions and focus specifically on that possibility.

To make an accurate assessment about the near term prospects of general AI, one needs to actually have real knowledge in the fields of machine learning and computational neuroscience, and especially specific knowledge concerning new developments in deep learning.

This article is a good introduction to that body of knowledge.

Understanding the future crucially hinges on understanding the state of the art in these fields.


This we largely agree with, as will many other AI researchers.  There is a growing awareness in the field that embodied development is important, and all the recent state of the art research involves a form of virtual embodiment during training.


This is pseudoscience that appeals to those who lack sufficient grounding in the requisite technical fields.  Quantum computing is enormously misunderstood and overhyped by the general media, and the brain is absolutely, provably not a quantum computer.

Quantum computing would enable some large speedups, but only for some very specific types of computations.  It could have relevance to AI oneday, but that day is far away, and we certainly don't need QC for superintelligence.

We already have achieved superhuman performance in many narrow domains (Deep Blue - chess, Watson - jeopardy).  

The most recent key development is we are closing in on parity with the primate visual system, which is a much much bigger development than Deep Blue or Watson.  The visual system is 25% of the monkey brain and about 10% of the human brain, and the cortex uses essentially a uniform architecture.  So figuring out the visual system leads directly to figuring out the rest of the cortex.


So ... you don't believe in evolution?


There's some contradictions here.  You argue first that strong AI is impossible because quantum-woo, but then according to you we can create "extremely sophisticated forms of artificial intelligence", but these somehow "fail to live up to our deeper expectations".  And then you also seem worried about handing over power.

aiShare's goal is simply to create "extremely sophisticated forms of artificial intelligence".  It doesn't matter whether they have 'true consciousness' or quantum-woo qualia, or whatever other non-scientific philosophical attributes you arbitrarily choose to associate only with human brains.

What matters in the end is economic value - can they do our jobs?  And the answer, of course - as you have more or less already admitted - is yes.


I already have spent a reasonable amount of mental energy attempting to clarify the need for scalability.

I already gave you one example of an interesting application that requires enormous transaction volume : a nano-payment based file sharing platform.

The other big future application category is distributed autonomous prediction markets.  These are a key mechanism for a future machine economy and central to our vision.  Imagine millions of agents each engaging in thousands to hundreds of thousands of nano-trades per second to predict anything and everything: the weather, traffic, crime before it occurs (haha - half joking), and of course: search and advertising.  The key problem there is predicting what a user wants based on their text query, what actions they will take next, and how much influence is worth in each case.  So essentially - a decentralized competitor to google.

These applications require or benefit from billions to trillions of transactions per second.

Stop thinking narrowly about just human scale transactions - there is very little untapped potential there.

Those prediction examples are all just external applications, where the most important long term applications are internal - distributed prediction and optimization markets to drive AI research itself.

A market is the most powerful global optimization engine - and its performance is measured in transactions per second.  So TPS is everything.  From this perspective, Bitcoin and Ethereum are like vacuum tube technology.

You are talking about going from 7 tps to 1,000 to eventually 10,000.  I am talking about going directly to the range of a billion to a trillion tps.  You seem happy catching up to Visa transactions levels, I would be only moderately satisfied with exceeding the entire transaction throughput of the current world financial markets combined (all the HFT stuff at NASDAQ, BATS - we can go far beyond that).


Sidechains/Treechains are still O(N) unless you have a sidechain/treechain per user (at least as I understand them) - which of course sacrifices all fraud robustness.  They are a half-step in the right direction, but they are starting very far from the optimal.

We are claiming to have solved O(C) decentralized transactions - which of course is optimal.  That's a strong claim to make, but that is the claim we are making.  

Optimal Scalability

Optimality means there is no better possible design (in at least the measured criteria).  The best that other networks can do is just eventually evolve into the same O(C) scalable design.

If Ethereum was based on the optimal scalable design, and there wasn't some large cost to using their network, we wouldn't have needed to work on the transaction problem and could have just used their system.  But alas, we had to solve that subproblem


Indeed - but which bits on which network?  That is the question.

EDIT: inefficient -> efficient, whoops

This is interesting, and gets more interesting if there is an efficient rental market for cloud hashing.

The attacker doesn't need to actually buy the hardware, they just need to rent it for a few hours or a day or whatever - just long enough to launch a 51% attack and steal money from a major exchange.

1. Rent enough hardware to create a temporary 51% pool.
2. start mining a private fork that gets ahead
3. Send in $X worth of currency A to the exchange (darkcoin say)
4. sell all of the A-coin for $X worth of B-coin (BTC)
5. withdraw all of the B-coin and wait for that to clear on the B-network
6. reveal your private fork to roll back the transactions and get all of your A-coin back

The attacker nets: $X + mining_revenue - mining_rental

In an efficient market, the mining rental cost becomes the same as mining revenue, so this scheme almost doubles the attacker's money.

The somewhat good news for bitcoin is that this attack is still easier for smaller market-cap coins, which would presumably start suffering from it first.

Nexus 9, we have a similar long term goal.  You say :


In particular, see our reply here concerning coup-completeness.

Ethereum has great marketing, but the core of their technology design - what we actually care about - is not a huge advancement beyond Bitcoin and alternatives compatible with it (Mastercoin, Counterparty, etc).  They may end up having the best practical implementation of a scriptable blockchain, or maybe just better marketing - but its still just a scriptable blockchain.

It will be able to process what - 1,000 transactions per second? 10,000 if they really compress transactions and or limit themselves to fiber nodes - maybe?  The core problem is O(N) transaction cost, so performance doesn't scale as you add more nodes.  That's not an improvement versus current technology, its a step back to the 80's.

Will you be able to build some sorts of applications on that?  Yes, but only by dropping almost everything off the blockchain and using some side database - essentially abandoning the design - which then opens up fraud and attacks again.  This just isn't interesting from a technology standpoint; it's a deadend.

To take just one simple example: BitTorrent is kinda cool, but it'd be way cooler if it could be integrated with a pirate-bay style database and a cryptocurrency, to create a market for valued file data.  That requires enormous quantity of transactions per second.  Its just not going to happen on anything like a blockchain.

But that's not the biggest problem.  Imagine if a country/region (say europe) actually adopted ether as their reserve currency 8 years from now.  The inflation of about 10% that year would amount to about a trillion euro being spent on pointless hashes.  The world is probably not that stupid.

Nor does this problem go away with PoS schemes - for the distribution problem is not technical, it is economic.  For a coin to go to reserve status, it has to redistribute wealth massively - and just about any such redistribution will be an enormous economic malinvestment.  PoS just redistributes it in a very concentrated fashion to the early stakeholders - who don't add any additional value thereafter - hardly ideal.  To avoid that problem, you actually must have some sort of strategic plan for the economy: something to invest all that monetary energy in which results in actual return and real GDP growth.

Accepting it is one thing, actually holding it in their cash reserves is another.

In most cases 'Accepting' just seems to mean autoconverting to dollars with bitpay - except for overstock which is actually claiming to hold some - are there any others?

http://unqualified-reservations.blogspot.com/2011/04/on-monetary-restandardization.html



Moldbug is at least partially correct, currency domination is coup-complete.

So, let us imagine - what would a real coup look like?  Firstly, it would be something created specifically for that goal - designed to destroy the current world order and rebuild it.  Secondly, it needs a decisive strategic advantage: an edge so overwhelming that it can takeover the world, without even requiring largescale adoption from everday folks.

What is the real source of power?  This isn't a trick question.  It is intelligence itself - that is the ultimate source of all power, all technology, and all value.

Bitcoin will not conquer the world because its Proof of Work distribution strategy consumes precious economic investment and outputs waste heat.  It is maximally unintelligent - pure entropy.

Imagine for a moment an alternate world where instead of Bitcoin we had a more optimal cryptocurrency appear at the same time.  Imagine - for sake of argument - that this alternate design was perfectly efficient and scalable, such that trillions of transactions per second are possible and double-spending is solved effortlessly without requiring proof-of-work.  Naturally this presents another potential problem (at least when viewed from within the narrow perspective of Bitcoin fundamentalism) or rather an opportunity - how should the coins be distributed?

The optimal distribution is simply that which maximizes the long term value of the coin.

Intelligence is the generator of all value, so the optimal distribution algorithm would be one that spends every coin in such a way as to maximally increase the intelligence of the network at every step. 

Instead of a huge supercomputer doing pointless hashes, you have a huge supercomputer doing AI research: deep learning, global optimization, etc.  And the end outcome is the generation of a new virtual world, economy, and distributed government of superintelligent agents.  Imagine tech similar to SIRI of today evolving towards something like Samantha from the movie Her  - and then beyond - all enabled by a distributed cryptographic economy, and of course, a new cryptocurrency. 

The coup is simply - the Singularity.

This project actually exists, and is just beginning.

Imagine a world with a currency that came in various bond forms, which we can divide into 3 rough categories:
* base currency which slowly loses X% of its relative value per year - where X is 3% or something
* mid-term bonds which stay perfectly stable in relative value
* long-term bonds which gain in relative value at X% per year

Normally this system would be inflationary, but that's kindof arbitrary - one can easily imagine just converting the currency into relative units.

Holding base currency gives you the most liquidity and is the best option if you may need to make big purchases in the near future - but you slowly lose wealth to 'inflation'.  Holding long term bonds gives you a return (negative 'inflation' that is the counterpart to the currency value loss), but you have to plan expenses carefully. Most people would hold some mix that suites their needs.

Now this system may seem 'strange' but it actually has some interesting pareto optimality going on - everybody is generally better off.  Inflation isn't a problem because the people holding lots of base currency don't expect to hold it for long anyway - and more importantly - the interest drives demand for bonds which removes liquid money supply which generally lowers prices.

So the currency holders are better off because they get lower prices compared to a system without bonds, and the long term bond holders are better off because they get interest.  Everyone wins.

Actually - and here's the secret - our current system is already kind of like this (but with a bunch of added extra complexity).  We just don't renormalize the money base, and moreover - the bond gain doesnt actually match the monetary inflation, so the current system 'leaks'.

EDIT: fixed changed URL link


Thanks!  We agree!  


Most new big technology starts out as fiction - literally in many cases, such as satellites.  And AI was first conceived in fiction long before it became a field.

As for the actual science, its mostly all there, it just takes some time to read and digest.  We have a more 'fun' introduction to the science which compares the brain to computing technology (with cool pictures) and has links to the important research:

http://aishare.co/wiki/Brain_Hardware_Parity

If your smarty pants friends have some specific smart criticism - we'd love to hear it.

As for the big team/funds - read more and give it time to set in, and eventually to snowball.  The project certainly doesnt need massive funds/developers right away, it just needs to get an initial network started, get the coin traded, incrementally prove some of the key innovations, make some headlines, and then eventually growth is organic and can continue as awareness spreads.

Nexus 9:

We understand you are a believer in Ethereum and are invested in that project (at least emotionally), and we sympathesize with the decentralization movement and thus have goals in common, however . . .


We are somewhat doubtful of both premises.  Our focus on scalability and performance is motivated by a very different vision for the use and applications of the network.  Ethereum - as we understand it - is mainly focused on building decentralized Web applications.  Aionic is focused on enabling new applications of AI, deep learning research, and so on: a very different immediate niche.  There is naturally overlap - both designs encompass programmable smart contracts - but the visions are very different.

Like it or not, Ethereum started without a solution to scalability/efficiency.  To his credit, Vitalik Buterin honestly admitted this from the beginning.

Our vision/plan for Aionic begins with a fully scalable design (driven by the specific needs of machine learning) - and it turns out that this requires a radically different foundation; it can't be based on a blockchain, even the principles and philosophy are quite different.  Scalability necessarily implies a partially observable environment, and that is exactly the situation where traditional computer science techniques break down and you move into the realm of AI and reasoning with uncertainty.


No - but it is reasonable to expect a strong foundation that at least points in the right eventual direction.  What hard problems exactly did ethereum solve from the beginning? Do you consider adding a scripting language for contracts solving a hard core problem?


So basically you are saying : "There can only be ethereum, even if we didn't solve any of the hard problems to start out with, ah sorry but we are the monopoly now."  Well sorry, but this is market evolution.


From the perspective of AI research, Ethereum is a non factor - not even on the path.


We are essentially making a bet that at least general AI is possible in less than ten years (software intelligence better at humans at programming, science, etc. that runs on a single machine with a couple GPUs and thus costs < $1,000 a year).  What do you think the probability of that is?  Do you have a pretty deep understanding of machine learning and neuroscience?  So pick some probability for that scenario.

Now, conditional on the above, what is the market value of general AI?  It's close to the market value of the entire economy : we are talking trillions to quadrillions.  It's technically more than that, because general AI will rapidly increase GDP - making the entire economy more valuable.

Web X.0 whatever doesn't lead to general AI.

It Begins with a Seed

The grand Bitcoin experiment has proven - against all odds - that a decentralized currency is both technically possible and also capable of bootstrapping monetary value into existence seemingly from nothing.  Yet despite this success, Bitcoin and all of its kin are flawed, for: 1.) blockchains do not scale efficiently (an intrinsic design limitation), and 2.) proof-of-work is an enormous waste of computational resources that could otherwise be allocated to vastly more profitable uses.  The full potential of the future machine economy is almost beyond imagination, but this vast potential will not be realized by any network of dumb nodes following a simple Blockchain protocol.  The key problems run deep down to the technical foundation, and the required solutions go far beyond simple improvements such as fully programmable contract scripting (ala Ethereum).  Yes, that too is important, but also pointless without solving the deeper scalability and efficiency problems for decentralized consensus that - once solved - will provide the foundation for the new machine economy of the Singularity.

Aionic is an upcoming agoric programming language, platform, and network designed around the concept of communication between and with intelligent autonomous agents.  It extends the concept of probabilistic programming (itself the most significant recent development in programming language theory) to include the concept of utility(value) - the foundational concept that makes markets and economies possible.  In traditional imperative programming, the developer describes exact logical instructions to a 'dumb' logical interpreter or compiler.  In agoric programming, the developer instead communicates high level goals and or rough program descriptions to a distributed economy of probabilistic reasoning agents, essentially contracting out the work to find a concrete implementation and execution on one or more machines.  In its most sophisticated eventual form, agoric development fully automates programming itself.

We are building Aionic as the foundation for aiShare : the decentralized research project to advance general AI towards benevolent artificial superintelligence, crowdfunded through a new truly scalable cryptocurrency - the Aion.  Every new coin issuance incentives actions and thus funds some change in the world.  Bitcoin, Ethereum and other proof-of-work coins fund worthless hashes.  Out of the space of all possible distribution strategies, the optimal distribution is simply that which maximizes the long term value of the coin.  There is nothing more valuable than superintelligence.

Our goal is to decentralize the Singularity - and in short - to save the world (literally).  The future is much too important to leave control in the hands of the few, the current elites.  The future needs your help; we ask for the community to join us as founders, developers, investors, evangelists and volunteers to ensure that the vast potential of general AI is used for the benefit of all mankind.

Find out More, and Join Us

main site: www.aishare.co

reddit: http://reddit.com/r/aiShare

aiShare was first announced on Friday, December 19, 2014.  Crowdfunding began on that date and is ongoing in the form of an Aion presale implemented through Bitcoin - full details are available on the main site.

This project is in its early days - all feedback is welcome.