Quote from: Nexus 9 on December 22, 2014, 06:26:02 AM

Alright Alix, Ive digested what youve said.

What you and others are referring to as coup-completeness or regime change is interesting. I think this is a real possibility with present technology, as Ill attempt to elaborate. But before I do that, I want to comment on some of your expectations for the future.

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).

Quote

I sympathize with your utopian dream of an all-powerful, all-valuable, all-technological intelligence bootstrapping itself into existence and saving humanity, but . .

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:

Quote

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.

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.

Quote

The most significant reason is that it ignores or denies the scale of innovation required to solve the embodiment problem.

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.

Quote

While its true sophisticated quantum computers might provide the required granularity of material detail to house true AI,

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.

Quote

Hence, we face the vicious circularity of requiring an all-powerful intelligence before we can design systems capable of producing an all-powerful intelligence.

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

Quote

Does that mean we cant create extremely sophisticated forms of artificial intelligence? No, of course not. Were already accomplishing that. . . .Systems of this sort will always fail to live up to our deeper expectations. For this reason, I would be extremely careful in deciding what kind of power we hand over to such systems.

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.

Quote

With that said, lets take stock of what we can actually do right now, or at least what we should be able to do in the near future. You argue there is an intrinsic limit to current blockchain design, in the neighborhood of 1,000-10,000 transactions per second. Im not clear why you think this is a significant problem.

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).

Quote

The limit on transactions is only going to continue decreasing with advances in computational power. We also know this can be solved using pegged sidechains and interoperable currencies, giving us the enhanced functionality, scalability, and efficiency you seem to think were lacking.

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

Quote

And it all begins with something as silly as dumb nodes trading bits on a network.

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

Alix Istek

Posts

Merits