Most people raving about cryptocurrencies are either trying to sell you something or complaining about all the people trying to sell you something. I’m going to briefly try something different here. Hopefully it’s refreshing!
In brief, two effects which I expect to matter far more than the market capitalisation of bitcoin, or non-fungible tokens in the metaverse, or whatever everyone else is talking about:
How does cryptocurrency adoption affect the scale and distribution of compute resources for artificial intelligence?
In brief: cryptocurrency demands GPU supercomputers. Machine learning also demands GPUs. Will crypto buildout crowd out ML applications? Or will it produce a compute overhang so that advanced AI rapidly proliferates past a certain point of development?
How will future computational agents i.e. advanced AI (be able to) use cryptocurrency?
Today we discuss (human) criminals and corruption and their enablement by crypto. Tomorrow we might worry that rogue AI agents could unlock the same illegibility and anonymity.
1. How does cryptocurrency adoption affect the scale and distribution of compute resources for artificial intelligence?
Take my word that the resources required to create and run artificial intelligence will be extremely important, because AI will be extremely important. So what are these resources?
One especially important resource is compute i.e. specialised supercomputer chips, specifically Graphics Processing Units (GPUs), which are indispensable to modern large-scale machine learning.
What else are GPUs good for? Well, originally they were designed for computer graphics (hence the name), and became especially valued for real-time rendering of computer games. Interestingly, the same sort of highly-parallel computations turned out to be useful for physics, chemistry, and biological simulation, machine learning, and… bitcoin mining.
Nowadays, a huge fraction of demand for GPUs is in ‘mining rigs’: if you can compute faster, you can more often win the specialised cryptographic computational races which are required (and richly incentivised) by proof-of-work blockchain. i.e. you get rich. That means people will pay a lot of money for GPUs, and the manufacturers can’t make them fast enough.
To some extent that means that machine learning applications, which aren’t yet as profitable, are a bit squeezed. But at some point, our AI algorithms and applications will prove to be very effective one way or another (for science, political power, design, entertainment, rogue AI takeover), tipping the balance in favour of AI demand over crypto demand. That looks like hangover material to me: with so much compute lying around, once AI can make use of it, that AI will be able to rapidly proliferate.
Personally I find this concerning because I don’t think we’re ready for advanced AI, so lots of advanced AI all at once seems especially daunting.
2. How will future computational agents i.e. advanced AI (be able to) use cryptocurrency?
Cryptography introduces a foundation for autonomous, anonymous, and highly computationally-accessible flow of funds.
There have been concerns and conversations for years about how criminals and corrupt powerholders could abuse (or are already abusing) this.1 But criminals and corrupt powerholders have physical locations and bodies, papertrails, persistent and recognisable identity. We can and do track them down and arrest them, at least sometimes.
Meanwhile, a rogue AI operation could be more like a ghost or shapeshifter: not something we can easily track or pin down. After all, it lives ‘in the cloud’: AI is software and can move, copy, and mutate very flexibly. Today’s society already struggles to pin down all the non-smart computer viruses and cyber worms; cybercrime overall amounts to trillions of dollars’ damage annually. A para-economy of uncontrolled or criminal AI agents accruing funds and power might be substantially facilitated by the availability of poorly-traceable cryptocurrency. Similarly for a more coherent, singular rogue AI entity.
If most or all humans eventually came to recognise a growing threat of resource capture like this, could we just agree not to do business with the growing rogue AI empire? Not so straightforward: enough people’s local, selfish interests might be served (and the transactions secretive or deniable enough) that such an emergent misaligned outcome could persist and grow, perhaps terminally, even under those conditions. At some point the rogue operation might act more like an adversarial state; certainly it would be capable of propaganda, distraction, fear-mongering, and divide-and-conquer tactics.
In contrast, centralised (or at least polycentric) currency promises more affordances for tracing and ‘knowing your customer’, providing at least one possible leg to balance a shaky stool of agent oversight on.
Interestingly, cryptographic record of transactions needn’t inherently be characterised by secrecy and illegibility. In principle it could be more transparent and harder to fiddle: the ledger of record is guaranteed - as a matter of mathematics! - to be maintained. But our current ecosystem of cryptocurrencies looks to have settled into a mainly anonymous-enabling balance.
Concluding
This is more out on a limb speculation than I sometimes go for. I imagine I’ll be wrong in some ways. But keeping an eye on what forces drive compute trends and what niches a rogue AI could occupy and grow in strike me as crucial. Cryptocurrency looks to have some implications for both.
Of course all of this depends on all of our actions in the coming years. Recognising the trends and implications could be a good start for making wiser choices and preparations.
In fact that’s why although I considered investing in Bitcoin in 2013, I bottled out due to not wanting to contribute to enabling corruption and laundering. Of course my contribution to its success would have been trivial, and on some views this was a big expensive mistake (I’d have made a lot of money)! I still endorse the decision-making that led me that way, and endorse and encourage similar virtue-motivated investment considerations, even if nowadays I’m a bit more cynical about individuals’ ability to influence things that way.