it's not what you think
I was recently asked by a traditional tech investor for the “galaxy brain” idea on how crypto and AI will combine to form a mega-trend. I’m very interested in each of bitcoin and traditional tech, which I think is what prompted the question.
A lot of people are trying to form narratives that smash crypto and AI together today. I can see why these narratives are attractive. There might be something in the future where AI machines do a bunch of training and inference work and the demand for their services is run on an open, competitive market of suppliers mediated by some sort of cryptographic monetary and payments technologies.
Sharing such narratives will definitely whip up a base of excitement on Twitter. Crypto is exciting. AI is exciting. If we collide them together wouldn’t it be even more exciting? 1+1=3!
But I’m guessing each of these technologies will need to make independent progress before we’ll collectively figure out how they should work together. Obviously some experimentation with crossing these ingredients can be fun and help us learn about the problem space, but we still face a lot of questions with each. For example:
What do we need for machine-to-machine payments to work?
- Traditional payment providers work today for machine-to-machine payments, but don’t achieve any of the goals of cryptocurrencies (low fees, fast transactions, censorship resistance).
- Bitcoin has superior monetary properties compared to other cryptocurrencies, but has so far fallen short on most attempts at scalable payments.
- Being a fast, scalable cryptocurrency payment technology is relatively easy to launch if you’re willing to compromise trust minimization, but the whole idea of using cryptocurrency is to avoid all the trust required to make these kinds of things work.
What are the winning services and delivery models for AI?
- Will the transformer long-term continue to be the most important AI architecture upon which all applications get built?
- What are the model/data/weight sizes required for various applications?
- How much are these open source/open model versus closed models/weights?
- How will the costs of training be amortized over the pricing of inference?
- Who will operate the infrastructure?
- What kind of hardware architectures might enable new applications?
- Who will be getting paid for which contributions?
- How much do winning solutions end up being delivered primarily by centralized versus decentralized providers?
The future collision is inevitable. But before we get to the unified theory, I’d say the “galaxy brain” idea is to help these technologies mature independently for a while. Primarily I’d like to see bitcoin achieve more scalable payment solutions and I’d like to see which AI applications (and underlying models) achieve PMF as utilities in our daily lives. Cryptocurrency payments and AI applications are fundamentally different layers of technology. The competition is fierce at each layer. Competing across layers today seems like a fools errand — you end up compromising too much. The winning formulas are almost certainly, as yet, undiscovered. Who operates what infrastructure, supporting which applications, and who gets paid for what?
I’d guess the collision happens a few years out from here and the fireworks will be magnificent to watch! For now, let’s keep building the best payment technologies and the best AI infrastructure/applications we can.