Summary: Deflation (or so I heard) is considered harmful because it stifles growth. Central banks have been fighting it to keep economies healthy. Cryptocurrencies can be designed to be deflationary. If the market capitalization of cryptocurrencies becomes big, central banks may have no way to contain the deflation. This may be catastrophic in many ways – but might it also slow AGI development and buy safety research more time?
I’ve been wondering what the “crypto endgame” may look like. Crypto may just turn out to have been a bubble or may continue for decades unchanged at its current limited level of significance. But we’re probably sufficiently prepared for those scenarios already.
Instead the crypto market capitalization might take off at a superlinear pace. Bitcoin is currently on rank 14 among the world’s currencies and holds 0.7% of the aggregate value of all of them. All cryptocurrencies together hold about 1.1% of the world’s monetary value.
The crypto space is rife with network effects. Or it is one huge network effect. These are wont to show superlinear growth if they show any growth. Superlinear growth often doesn’t last long enough to be significant, but in other cases it does. So maybe crypto’s market capitalization will grow by an order of magnitude or more within a few years, so before AGI is developed. That might happen if more more companies follow the examples of MicroStrategy or Tesla and invest their spare cash into a cryptocurrency. Phil Bonello, director of research at Grayscale, also mentions a scenario in which governments start buying crypto and thereby force all other governments to follow suit. I don’t know the reasoning behind that, so I don’t know how forcible that effect will be, but it doesn’t seem to be hard to come up with plausible growth scenarios. (At least if you’re incentivized to come up with them the way Grayscale is.)
Some cryptocurrencies, notably Bitcoin, are designed to be deflationary. If companies across the board find that they generate greater profits by investing all their spare cash into deflationary cryptocurrencies rather than reinvesting into their own production, innovation may stall. (Isn’t that one of the typical reasons why deflation is considered harmful?)
So that brings me to my questions:
Conditional on the cryptocurrency market growing by another 1–2 orders of magnitude in the next five years or so, what happens to the global financial system and the world economy? (Might this future look like a cyberpunk dystopia where a rich elite retires on their capital gains, and anyone who didn’t enter the period with enough capital has very few avenues to sell their labor and so lives in poverty?)
And might this stalled innovation also carry over to AGI in that technology companies will find it more profitable to invest into cryptocurrencies rather than compute or staff, thereby buying safety research a few more years or decades to solve ethics, decision theory, and international coordination?
So I have yet to see what utility crypto provides besides “transactions are irrevocable and semi anonymous”. (as a miner you are just a tulip farmer and are gaining only temporarily). Do you know of any? Smart contracts provide a way to do some limited forms of escrow.
So if you want to buy something online, presently:
You can use a credit card. A temporary virtual one. If the transaction isn’t satisfactory you almost always get your money back simply by complaining. Fees are approximately 3% though hidden from you. A record of the transaction is held by your bank and can be obtained by investigators with a court order.
You can use crypto. Transactions are irrevocable and it is possible to get your money stolen from various forms of phishing or hack. Crypto may decline in value in real time while you are making the purchase. Fees can be tens of dollars including both exchange fees and transaction fees. The recipient may be using a temporary receiving address, so even if you use an exchange that keeps a record, investigators with a court order may not know what the money was used to purchase.
Basically, I don’t know what legitimate use crypto has for those who aren’t trying to break the law in some way. Whether the law is restricting contraband items (drugs and weapons and stolen information) or taxes or international funds transfers.
As a consequence, at some future point, Western governments may outright ban crypto exchanges. This has already happened before—look what happened to e-gold. They can’t ban crypto directly, but banning the exchange of currencies controlled by western government for crypto and vice versa would crash it’s value and reduce the number of users to a tiny number of people willing to go through the risks and hassles of evading the ban.
Here’s a legitimate application, buying PornHub Premium. https://news.bitcoin.com/pornhubs-premium-services-crypto-payments-13-digital-assets-supported/.
Online payment processors are an oligopoly and can at any moment revoke a businesses ability to receive online payment even if they’re not breaking the law. Thus what business is and is not permissible online is entirely up to the whims of this oligopoly and the law. Crypto provides a way around this.
That is a narrow niche but “I’ll tentatively allow it”. I forgot about this—that in certain sectors, corporations have gained quasi-governmental powers. Social media sites being able to ban entire ideas from being discussed, the entities able to provide DDoS protection (which requires very large scale, creating a natural oligopoly) can choose not to protect sites that are expressing unpopular ideas.
I guess it was “first they came for the ..” because I am not completely certain letting outright hate groups and conspiracy theorists coordinate with each other is a net societal good. But yes this is a problem right now, like you said, it’s not illegal under present law for someone to run a porn site that accepts user submissions, even if occasionally this means illegal content will remain on the site until reported. This is also what Stormfront/etc are as well—technically the illegal content gets removed after complaints are raised. (after giving the site viewers plenty of time to read it, of course)
Thanks! If I read your reply correctly, then those are reasons why the extreme growth scenario that I’m worried about is unlikely and why, therefore, my conditional question is unimportant. They’re also similar to the reasons why I didn’t invest into crypto until 2016. I took it for an obvious speculative bubble about to pop, one with no commensurate underlying value.
I suppose my default assumption should still be that most of the current crypto price and market capitalization is purely speculative. But the more boom and bust cycles the market continues to survive without collapsing entirely the more I’m starting to think that I might be wrong about that in ways I don’t anticipate or that it may not be important in the end.
So I’m still interested in what might happen if the above scenario comes true and lasts for long enough to make a difference for safety research.
To be honest, I think crypto is either neutral or even slightly positive for the growth of AGI. The reason is that crypto mining is one of the major drivers of GPU and specialized ASIC designs. While it does cause shortages in the long run it actually helps fund the development of more chip fabs and even more powerful GPUs. It indirectly is making AI R&D cheaper because it is lowering the (long term) cost paid for powerful GPUs, and software stack improvements made to support GPU compute for the crypto market carryover to the inference and network training market.
Moreover, one company that got it’s start making crypto ASICs (bitmain) has used the money and chips it developed to get into the AI market.
However, at least for the niche I work in, AI systems that “take off” seem very far away. I think years of work are needed just to develop the platforms to make development and deployment of AI systems that do relatively well defined things (driving a car without crashing, maneuvering a robot to do manufacturing or logistics) reliable and economical.
Once we have this kind of stuff working reliably maybe we can start looking into the higher level abstractions that artificial sentience would need. But I don’t think it will be simple, and I don’t think it will just accidentally wake up from a random developers computer and eat the internet. Probably.
Good point. A broad switch away from proof of work (as it seems to be happening) may change that dynamic.
What incentive is there for a broad switch to proof of work?
Away from proof of work. :-)
Sorry that’s what I meant to ask
Preventing 51% attacks. Maybe also others. And for the environmentally minded people, there’s also the reason to decrease the power consumption. I heard a region of China has banned new mining facilities because of their energy consumption. If that continues, 51% attacks may become easier again unless more blockchains switch away from proof of work.
See that’s why I asked what’s the incentive to switch to proof of stake and not why it’s better. Like with climate change, this is a coordination problem.
Or banning prediction markets. A prediction market would have allowed business owners last year to effectively insure themselves against lockdowns.
If you buy a drug on Hydra there’s independent quality testing. If you buy a generic drug or supplement in an US pharmacy under the watch of the FDA there isn’t independent quality testing. Hydra does human trials for this that might be illegal to do in the US in the unbureocratic way Hydra does them.
Sure. But everything you said is illegal. Maybe not immoral but not legal. And maybe we would have a better world where people could do these types of illegal things without the government being able to stop them. But my major point is the government has to stop them—it can’t permit widespread flouting of its authority—and it has a mechanism to do so. Just order banks to not allow funds transfers into crypto exchanges. You could get around this by finding a black market crypto seller and paying cash, but this would still basically kill crypto.
People have argued that for a while and it hasn’t happened.
It took a long time for the government to prohibit hard drugs and machine guns. Crypto is brand new and was a little known toy less than a decade ago.
The most obvious use case is as a store value; “digital gold” as Peter Thiel likes to call it. Bitcoin is limited in supply and has enough network effect behind it to succeed in the long term, other cryptocurrencies much less so regardless of any technical advantages. I don’t see crypto getting banned because A) there is too much institutional investments in it and B) it’s not a threat to Western governments any more than gold is.
Regarding B: this is not true. Someone who accepts crypto for their business can easily set up the system in a way that it is optional which transactions they report. Sort of how cash only businesses—or at a lower level, people paid cash tips—probably nearly all cheat to some degree on their taxes. Some of the cash can be spent on personal expenses without any records kept. Crypto makes this way easier—no longer is there large sums of cash around, it’s harder for the IRS to audit, and the same ‘benefits’ apply—someone can accept crypto for a transaction to a randomized new account they control, and sometimes they import that transaction into the books they show the government, and sometimes they don’t.
The government has to collect money to operate, it can’t have everyone cheating. It leads to failure of the state when too many people are cheating the system.
It also allows for purchase of things that governments don’t want citizens to have far more easily than before. Yes, in any major city, you can probably find a drug dealer for whatever recreational drug you are seeking. But can you find the parts for a specific model of machinegun? A rocket or even guided missile launcher? (ok I’ve never heard of one of those for sale on the dark web but it’s possible). A hitman who isn’t definitely a cop? (I’ve never actually heard of the dark web successfully being used for this, but combined with Smart Contracts for escrow it’s at least possible ). Can you get a data set of stolen credit card numbers, or a zero day for a major type of computer system?
None of these illegal goods were easy to get before because they are so specialized. It’s the same problem where if you were doing hobbyist electronics and shopping at radio shack, you were limited to very basic designs. The government being able to control access to weapons—which both heavy weapons and high end hacking tools count as—are part of their monopoly on the use of force and again a threat to them.
Anyways I’m not disputing (A) but I can see a possible crash of crypto if (B) ever leads to a major incident, such as a major government VIP assassinated with a guided missile, all arranged through the dark web with payment through crypto. Governments can regulate crypto, as mentioned—they merely need ban Western banks from knowingly allowing funds in the bank accounts from being used to buy crypto, or allowing the proceeds from a sale of crypto sale to be deposited in them. This would pretty much kill crypto.
The Russian dark website Hydra was used for it.
Neither of which is a meaningful challenge to state monopoly of violence. You can even legally own tanks, but these weapons are very powerful only in limited contexts. Sure you can blow up a few people with it before being subdued, but an assault rifle or even a truck would do the job just as well. You’d have to go to WMDs before the situation becomes problematic.
The more fundamental problem with this argument is that, once important state secrets/WMDs have been stolen, the damage is done; the fact that someone is trying to make an extra buck off of it afterwards via crypto is rather trivial.
The problem with hiring a hitman on the web is first and foremost that there is no incentive for the hitman to follow through. There is obviously no legal recourse for the buyer, and using cryptocurrency actually disincentivizes the hitman from doing his job even more. If you’re untraceable, you’re also reputationless. I don’t see what problem is being solved by smart contracts here; at the end of the day you have to interact with the real world to enforce your contracts.
You could do the same thing with gold, yet no one bothers to do it. That’s basically my knock-down argument against crypto being revolutionary in general: there is no important aspect in which crypto differs from gold. I guess we’re not really disagreeing here; I think governments will just slap some regulations on coin exchanges and that will be end of the story, no need for more drastic measures.
I don’t see what problem is being solved by smart contracts here; at the end of the day you have to interact with the real world to enforce your contracts.
The smart contract would specify the identities of other network accounts who will be able to vote on the outcome of the contract. So there could be, say, 10 third party accounts who represent “observers” who vote on whether or not they believe the target was killed. (from reading the paper). These “observers” would have a reputation built up over past predictions and might not actually be human beings who can be arrested. (or they might be living in countries where this action is not considered a crime).
This means the hitman can either hunt down each anonymous observer and force them at gunpoint to vote the hitman’s way (which other mechanisms would need to zero out their reputation score for this false observation) or kill the target or forfeit the money back to the buyer when the contract expires.
You’re just delegating the problem away to an observer reputation system that has the same problem one level deeper. Who actually has incentive to align reputations of observers with what actually happened?
This is a thorny problem, and I’m not working in this space. Having thought a bit about the problem and rejected many other possibilities, what I arrived at is this:
day 0, no one has a reputation but n accounts ‘volunteer’ to be judges
Day n, each “judge” has a history log of the (evidence, decision). Automated tools detect a corrupt judge by looking at the log and looking for decisions not justified by the evidence, and then the buyer and the seller agree on a possible list of non-corrupt judges, and a random sampling of them is chosen. (simplest way is to look at a judge making a different decision from another judge, but determining who is “right” when the majority is wrong is a difficult unsolved problem)
There are some difficulties with this, namely that a judge can only make decisions on publicly available information. For example, you could in theory use it to place a bet for an event that will later happen, and these judges vote whether or not your bet is good.
The incentive that the judges have is the longer the history log of correct decisions, the more that judge is “worth” and the larger the fee they will get.
The main problem with deflation is mainly not about holding M1 but promises of payments that are denoted in the currency. When currency inflates it’s relatively easy to raise the salary of employees to match the value of the currency. On the other hand when deflation happens, making salary cuts is much harder.
It’s possible to write laws that force salaries to be payed in the currency that’s preferred by the government for most legal employment.
You can write laws that make loans in crypto currencies not legally enforceable which prevents a lot of current loans that are backed by the governments power to seize future earnings of a person.
That said, even if you want to use crypto for denoting future payments of money using a stable coin like reserve makes much more sense then using bitcoin.
Thanks! Yeah, I don’t imagine that a currency that deflates will be used for everyday payments like salaries. Stablecoins or fiat seem more appropriate for that. But that doesn’t undercut the worry that a currency that deflates may be preferred over reinvestment and hence stall innovation – the scenario I sketch above. Or does it?
There were a few good points distributed over many answers and comments, so I’ll collate them here. I’ll ignore answers to other questions, such as “Is it likely that cryptocurrencies will get big?” I’ve paraphrased all of these.
Question
Bitcoin is not actually deflationary by design but its popularity is increasing faster than its supply. That will stall at some point, at which its value will not grow anymore. That’s probably going to be 100 years from now, so maybe not relevant to AI, but a cyberpunk scenario is unlikely to be stable that way unless a more deflationary asset becomes dominant. (H/t Olomana)
A 1–2 OOM increase in market cap may not be enough for crypto to have much impact on society. It would still just be one more biggish assets class. There is not that much value in currencies that becoming as big as all currencies combined would have these vast societal effects. (H/t romeostevensit)
Question
Proof of work generates demand for GPUs and ASICs, allowing them to be produced at greater scale (and maybe smoothing out demand volatility). That hardware is also useful for AI, so proof of work at least may accelerate AI. (H/t Gerald Monroe) Bitcoin is posed to remain dominant, and it may be too conservative to switch away from proof of work. (That leave the avenue of promoting non-PoW blockchains so that tokens on those blockchains can displace Bitcoin to some extend.)
Did I forget any?
I think that market growth isn’t well defined. I don’t believe market capitalization is particularly meaningful. Suppose we look at the sum value of all transactions made in crypto and we compare that to the sum values of transactions made in other currencies. For arguments sake lets say this is currently 1:10,000 and a two order magnitude change will bring this to 1:100 . So what will happen? Probably nothing. I would think for this to happen the successful cryptocurrencies will need to end up looking a lot more like todays payment methods. If somehow something akin to bitcoin becomes the ‘official’ world currency, I imagine those who do the labor would find alternative means to trade amongst themselves.
If crypto (or else) turns hugely profitable over night I’d expect resources and talent to leave AGI in favor of crypto. However I would also expect a similar reduction in resources within safety research; for every year you buy safety research, they’d take another year to get the same work done.
Thanks! I can’t argue for the importance of market capitalization, but I don’t think transactions are a good proxy either. For something to actually be used as currency, its value would have to be somewhat predictable, so I think that role will fall toward cryptocurrencies like the stablecoins that we have. The deflationary cryptocurrencies are more likely to be used as “stores of value,” just bought and held by most owners, and traded only by traders on markets where you don’t actually exchange the underlying coin (like most centralized crypto exchanges today).
Hmhmm, yeah maybe… I suppose it should be the default assumption that sectors behave the same unless there are differences between them. But I can think of some differences. Crypto finance may be highly automated, so there may be few jobs there. The energy and hardware sectors may grow, but perhaps also not to the point where they can absorb all labor. So people will either fall into the category of (1) desperately looking for employment or (2) not in need of employment. The assumption is that AGI development will slow because companies are not incentivized to invest in it – compute or staff. Just as an observation, AI safety people will enter this period with capital. Also they’re not motivated primarily by profit but by saving the world. So they’ll be motivated and able to pay the people in group 1 and retain the people in group 2, who may not even ask for a salary. That could still lead to a differential speed-up.
Why do you think that transactions are a poor proxy? I don’t fully buy into the “store or value” idea. My uneducated opinion is something can only store value if it has an underlying use. Facilitating black markets is certainly an economically valid use.
I’m confused by your second paragraph. Perhaps some of it is intended as a response to my answer to your first question? The primary argument which I can identify is that AI safety is not motivated by profit, but by saving the world. Saving the world requires money. Many people working on AGI probably think it will have an enormous positive impact on the world, so I can equally claim that they might not even ask for salary.
I don’t find you’re argument convincing. It requires a lot of conditionals and appears to contain much room for error/oversight. You might be right but it doesn’t serve me to speculate deeply into this topic, I’m content with not knowing.
Oh sorry, I didn’t see your comment.
There is the existence proof of gold, silver, platinum, etc. But people have mostly convinced me already that 2 OOM are not going to be enough to have much of an effect.
I think you understand my argument correctly. :-) You argued that AI safety will not differentially speed up compared to AI capabilities because it will slow down at the same rate. Maybe not literally the same rate but the same rate in expectation. But that still seems unlikely to me since I can think of reasons why it would slow down less but can’t think of reasons why it would slow down more.
Yes, precious metals are used as stores of value, however they are also used in electronics and in jewelry (which functions as a way to signal wealth). So I don’t consider them counterexamples.
It’s fair for you to believe that, but you’re probably using a lot of weak evidence so it’s hard to communicate that evidence to me. I’m just saying that what you did share wasn’t sufficient to shift my opinion.
Bitcoin is not deflationary. It is slightly inflationary, much less inflationary than fiat currencies, but it is not deflationary.
Fascinating, thanks! I found this article. Is that roughly what you’re referring to? It sounds like the author would agree that it is deflationary so long as the user base grows faster than the supply. In that case, my scenario above should self-correct eventually, unless a more deflationary coin catches on.