Alphaville on Bitcoin

I'm not a regular reader of the Financial Times, so I really regret I hadn't noticed that Izabella Kaminska and others at the FT's Alphaville blog have been posting excellent work in their BitcoinMania series. For a taste, see Bitcoin's lien problem, in which Kaminska discusses the problems caused by the fact that the blockchain records the transfer of assets but not the conditions attached to the transfer:

For example, let's hypothesise that Tony Soprano was to start a bitcoin loan-sharking operation. The bitcoin network would have no way of differentiating bitcoins being transferred from his account with conditions attached - such as repayment in x amount of days, with x amount of points of interest or else you and your family get yourself some concrete boots — and those being transferred as legitimate and final settlement for the procurement of baked cannoli goods.

Now say you've lost all the bitcoin you owe to Tony Soprano on the gambling website Satoshi Dice. What are the chances that Tony forgets all about it and offers you a clean slate? Not high. Tony, in all likelihood, will pursue his claim with you.

She reports work by George K. Fogg at Perkins Coie on the legal status of Tony's claim:

Indeed, given the high volume of fraud and default in the bitcoin network, chances are most bitcoins have competing claims over them by now. Put another way, there are probably more people with legitimate claims over bitcoins than there are bitcoins. And if they can prove the trail, they can make a legal case for reclamation.

This contrasts considerably with government cash. In the eyes of the UCC code, cash doesn't take its claim history with it upon transfer. To the contrary, anyone who acquires cash starts off with a clean slate as far as previous claims are concerned. ... According to Fogg there is currently only one way to mitigate this sort of outstanding bitcoin claim risk in the eyes of US law. ... investors could transform bitcoins into financial assets in line with Article 8 of the UCC. By doing this bitcoins would be absolved from their cumbersome claim history.

The catch: the only way to do that is to deposit the bitcoin in a formal (a.k.a licensed) custodial or broker-dealer agent account.

In other words, to avoid the lien problem you have to submit to government regulation, which is what Bitcoin was supposed to escape from. Government-regulated money comes with a government-regulated dispute resolution system. Bitcoin's lack of a dispute resolution system is seen in the problems Ross Ulbricht ran in to.

Below the fold, I start from some of Kaminska's more recent work and look at another attempt to use the blockchain as a Solution to Everything.

Last month Kaminska, the author of the excellent From the annals of disruptive digital currencies past, reported on the emergence from stealth mode of a startup called 21 Inc. whose plan is to put Bitcoin mining hardware into every device in the Internet of Things. As she explained in a second post entitled 21 Inc. and the plan to kill the free Internet the underlying goal is to replace the current "free" advertising-supported model with one based on making every Internet-connected device mine Bitcoin to provide the funds:

This isn't about disrupting fiat money, central banks or the existing financial rentier system. It's about making the internet much more like the financialised real world. Namely, by adding an energy and scarcity cost to digital transfers on the web so that information can't be as easily exploited as it is today.

Up for grabs, notably, is the marketshare of Google, Facebook and Twitter and their ilk, due to their dependency on free consumer data to drive their advertising-based revenue.

Now, I agree with the quote from this interview with Monica Chew that I used in Preserving the Ads?:

[Chew] further argues that advertising “does not make content free” but “merely externalizes the costs in a way that incentivizes malicious or incompetent players.” She cites unsafe plugins and malware examples that result in sites requiring more resources to load, which in turn translates to costs in bandwidth, power, and stability. “It will take a major force to disrupt this ecosystem and motivate alternative revenue models,” she added. “I hope that Mozilla can be that force.”

A replacement business model that lacked incentives for bad behavior would be a good thing. Perhaps Bitcoin mining in every device could be the major force, and would lack incentives for bad behavior, but I am doubtful.

Lets start by noting that the "free Internet" can't be killed because it doesn't exist. Advertising supports much of the content of the Web, but attempts to provide advertising-supported "free Internet" connectivity have never succeeded at scale. We pay for our connectivity at home, Stanford pays for our connectivity at work, Starbucks pays for the connectivity underlying their "free WiFi".

Perhaps 21 Inc. has the idea that an ISP could provide a home router that paid for its connectivity by mining Bitcoin. To pay for my home connectivity it would need, assuming current Bitcoin prices are stable, to mine one Bitcoin every 4.5 months for as long as I owned it.

Lets assume that 21 Inc. could initially deliver hardware with insignificant cost and power consumption that could mine at that rate. One problem has been that, driven by Moore's Law, the hash rate of the Bitcoin network was increasing rapidly, which meant that the useful life of Bitcoin mining hardware was very short, perhaps only 6 months. Equipment in the IoT, such as routers, has a long service life. No ISP is going to replace all its customers' routers every 6 months, so on 21 Inc's model they would indeed be providing most of their customers with "free Internet" and thus losing money.

To be viable, the router would have to mine enough Bitcoin in its first six months to pay for connectivity for say 6 years, or about 0.6BTC/week. A current mining ASIC, the AntMiner S5, mines about 0.1 Bitcoin/week. So we would need 6 times the performance of state-of-the-art chips. One S5 consumes 590W and costs $370. Over 3.5KW and $2,200 make for a rather expensive home router. It would take some serious chip magic to get this level of performance out of something that an ISP could afford to put in a router, and 21 Inc. would make more money using their magic chips to mine on their own behalf than they could extract from the ISP market.

For the last six months the hash rate has not been growing; it has been between 300 and 400 PHash/s. Doesn't this invalidate the argument that mining hardware has a short life? No, because as the introduction of the S5 shows, the development of new, better hardware has continued. Miners who can afford to buy state-of-the-art hardware mine more Bitcoin for a given energy input. The stable hash rate is not the result of big miners stopping getting bigger, it is the result of smaller miners with less efficient hardware being forced out of the market. The value of mining hardware (including 21 Inc's) is still dropping rapidly.

21 Inc. believes, as they say in this image, that they have finally made micropayments feasible. A Satoshi is 10^-8BTC, so a chip that can mine a million Satoshi a year at current prices generates an income of $2.25/yr. We are truly talking about micropayments. Not even close to enough to pay an ISP for connectivity. Not enough for three $0.99 songs. At my rate of $0.096/KWH if the mining hardware drew 2.7W it would generate no net income.

The claim that micropayments are now feasible has been made many times in the last quarter-century, but they have never succeeded. Emin Gun Sirer points to a remarkable, decade-old paper by Nick Szabo, an early participant in Bitcoin development, called Micropayments and Mental Transaction Costs. Szabo writes:

We have seen how customer mental transaction costs can derive from at least three sources: uncertain cash flows, incomplete and costly observation of product attributes, and incomplete and costly decision making. These costs will increasingly dominate the technological costs of payment systems, setting a limit on the granularity of bundling and pricing. Prices don't come for free.

Notice that 21 Inc's chips haven't addressed any of the issues Szabo raises. The owner of a device will have only a rough idea of the value it will generate. Because the chips are so feeble the chance that the owner's chip will ever actually mine a Bitcoin is effectively zero. They make sense only as components of a huge mining pool, which the device owner has to trust. The value of the chip's mining will depend not merely on the market for Bitcoin but also on how large a proportion of the total mining power the pool represents. Vendors offering services to chip owners, such as the ISP in the example above, will not know what their revenue will be worth. The chips add huge amounts of risk to everyone.

If we're going to replace the business model behind both the current Internet and the current Web the replacement needs to provide not just lower risk but also better security. The business model underlying connectivity is quite secure; ISPs don't seem to suffer significant fraud losses. That isn't true of the Web advertising business. Although there have been recent advances in some areas of blockchain-related security, such as in providing two-factor authentication for the wallets in which the bitcoins must be stored, fundamental problems remain.

In a longish but must-read piece entitled A Machine For Keeping Secrets? Vinay Gupta, who is working on a blockchain startup called Ethereum, points out the most fundamental problem with basing commerce, the Internet, or anything else important on the blockchain. It is the bane of all "reliable" distributed systems - inadequate diversity of the underlying technology:

There is more at risk than individual users being compromised and having their contracts spoofed. In a distributed system, there is a monoculture risk. If we have individual users being hacked because their laptops slip a version behind bleeding-edge security patches, that's bad enough. We have all heard tales of enormous numbers of bitcoins evaporating into some thief's pockets. But if we have only three major operating systems, run by >99% of our users, consider the risk that a zero-day exploit could be used to compromise the entire network's integrity by attacking the underlying consensus algorithms. If enough computers on the network say 2+2 = 5, the nature of blockchains is that 2+2 not only equals 5, but it always will equal five.

The blockchain is a brittle system. How realistic a problem is this? An attacker with zero-day exploits for each of the three major operating systems on which blockchain software runs could use them to take over the blockchain. There is a market for zero-day exploits, so we know how much it would cost to take over the blockchain. Good operating system zero-days are reputed to sell for $250-500K each, so it would cost about $1.5M to control the Bitcoin blockchain, currently representing nearly $3.3B in capital. That's 220,000% leverage! Goldman Sachs, eat your heart out.

Brian Arthur's work two decades ago showed that technology markets with increasing returns to scale would be dominated by one or a small number of players. Events since then (CPUs, GPUs, search, social networks, Bitcoin mining pools, ...) have amply confirmed this. I agree with Gupta that work on capability systems such as Capsicum has the potential to greatly improve the security of the systems we depend on. But the underlying economics of the markets that provide us with these systems make the monoculture risk he describes unavoidable.

The forces squeezing diversity out are the same ones I discussed in Economies of Scale in Peer-to-Peer Networks, using Ittay Eyal and Emin Gun Sirer's analysis that Bitcoin and similar protocols need be incentive-compatible:

the best strategy of a rational minority pool is to be honest, and a minority of colluding miners cannot earn disproportionate benefits by deviating from the protocol

This is a game theoretic concept. In a recent series of posts Arvind Narayanan has been pursuing a game-theoretic analysis of Bitcoin:

• Consensus in Bitcoin: One system, many models "The branch of math that studies the behavior of interacting participants who follow their incentives is called game theory. This is the other main set of models that's been applied to Bitcoin. In this view, we don't classify nodes as honest and malicious. Instead, we assume that each node picks a (randomized) strategy to maximize its payoff, taking into account other nodes' potential strategies. If the protocol and incentives are designed well, then most nodes will follow the rules most of the time."
• Bitcoin and game theory: we're still scratching the surface shows using very simple arguments that the "block withholding" attack is profitable for the attacker, and thus that the Bitcoin protocol is not incentive-compatible.
• Bitcoin is a game within a game analyzes Bitcoin as a two-level game, the lower-level game played among the miner software implementations, and the upper-level game played at much lower speed among the humans running the miners.

In Economies of Scale in Peer-to-Peer Networks I asked the following questions:

• Does incentive-compatibility imply income linear in contribution? The answer is that anything faster than linear is not incentive-compatible.
• If not, are there incentive-compatible ways to deter large contributions? In principle, yes, but in practice networks that used them would not succeed.
• The simplistic version is, in effect, a static view of the network. Are there dynamic effects also in play? Yes, and they favor large miners.

My conclusion was that economies of scale pose fundamental problems for peer-to-peer networks like Bitcoin. They depend on being distributed enough that a conspiracy controlling a majority of the mining power is infeasible. But unless there are no significant economies of scale they will end up being dominated by a small number of large players, as we see with Bitcoin mining pools.

It isn't clear to me that 21 Inc. has abolished economies of scale. What they are trying to do is to externalize some of the costs of mining (capital, power, communication) to device owners by burying them in the costs of running the device. Externalizing costs in this way has already been tried. Malware that infects home routers and PCs with mining software is fairly common. The malware suppliers costs are low, so even the small amount of Bitcoin they can mine is profitable. The cost base for 21 Inc's "hardware malware" is higher.

The end of Kaminska's post loses me when she writes:

As it stands today, nobody really has an idea of what their data is worth because there’s no mechanism by which processed or unprocessed data bundles can be valued.

Yet we know for a fact that personal data, especially when processed in aggregate form, does have a value in the open market. If it didn’t, Facebook and Google wouldn’t be the multi-billion dollar organisations that they are today.

But the data market trades much more like highly illiquid OTC commodities than anything akin to an open market exchange. Deals seem to be done bilaterally, on a bespoke and opaque basis. There is no “processed data”-value index.

She goes on to suggest that:

While it’s true that the processed data in question is light on both information (due to so much of it being pseudonymous in nature), there’s no denying the energy it took to create it.

It’s this base energy cost that can now be used as a public benchmark to price more information-intensive data against.

This makes us think the key objective of the high-order bitcoin enthusiasts (as opposed to the financial speculators) is mostly about giving consumers a choice. On the one hand to pay for specially designed web services with spent processing time (and energy) that helps support the public digital commons (which acts as a glue that links up all sorts of different datasets). Or, on the other hand, to pay for open web services directly with personal data on a trust basis.

Either way, once a transparent price is established for the former, it stands to reason a price for the latter can also be equally derived, opening the door to a meritocratic paid-for internet and a market where all personal data has a clear market price.

Unless I'm totally misunderstanding her argument, it seems to be full of misconceptions. The value of "processed" data isn't linear in its size, or in the energy expended in "processing" it. It is true that the Bitcoin market sets a value on a very specific, but otherwise completely useless, computation that is related by the current state of chip technology to the lowest cost of electricity for data centers across the world. But apart from being highly variable, the energy costs of the Bitcoin computation have nothing to do with the energy costs of the computations that, for example, Google does to add value to the raw data they collect.

In 21 Inc's world, the idea is that end users generate Satoshi to pay sites not to collect their data (and how do the users know the site isn't?), not that sites pay end-users to be able to acquire data. The amount is set by 21 Inc, presumably uniformly, whereas in the real world the value of personal data depends strongly on the person. Personal information about, say, Warren Buffet or Barack Obama is worth a lot more than personal information about me.

Cathy O'Neill points out another problem with this concept. Current end user licenses allow the vendor to change the terms of the license at will:

As everyone knows, nobody reads their user agreements when they sign up for apps or services. Even if they did, it wouldn’t matter, because most of them stipulate that they can change at any moment.

The customer's only recourse is to stop using the service, and switch to a competitor, like that other eBay just down the street. But increasing returns to scale means that the competitors will be much smaller, less useful, or even non-existent. So even if your hardware is generating Satoshi to pay for the services you are using the power imbalance between eBay and you, or in this case your home router, hasn't changed.