I think it's worth putting into perspective though that machine learning kicked off in the 60s, whereas there wasn't any notable progress on the byzantine generals problem until 2009 (and even the concept of proof-of-work wasn't formalized until 1993). Distributed ledger technology is still relatively young compared to machine learning.
AI has followed the classic Gartner hype cycle [1]. There was the trigger in the 1960s-90s, the peak of inflated expectations in 2000-2010 with every other company bragging about "AI this" or "machine learning that" while having nothing of any use to show for it, then finally the "slope of enlightenment" from the 2010s-present where AI is actually becoming useful to people in their everyday lives.
Crypto may be on the way out - or - it may be following a similar curve on a shifted timeline. For example: Trigger in 2009, peak of inflated expectations 2017 (cryptokitties), trough of disillusionment 2017-2030, slope of enlightenment 2030-2040.
PyTorch: Initial release September 2016; 6 years ago
Tensorflow: Initial release November 9, 2015; 7 years ago
> In 1983, a research paper titled "Blind Signatures for Untraceable Payments" by David Chaum introduced the idea of digital cash.
> The first artificial neural network was invented in 1958 by psychologist Frank Rosenblatt.
In the limit, the dates don't really matter. I'd wager both fields have had roughly the same human capital afforded to them - I'd be happy to be shown otherwise.
Neural networks really only had a scaling problem to overcome, and most people knew it. What's the technical problem with crypto?
* The waste of Proof Of Work (in terms of both on-chain economics and off-chain externalities). Introduced in 2009, solved in 2023 with Ethereum's final Proof Of Stake update (earlier if you look at smaller cryptos or if you count Ethereum's half-finished PoS algo starting in 2020).
* The inability to scale transaction throughput. Became a problem with Bitcoin in 2015, partially solved by Ethereum in 2022 with Layer 2 chains, to be fully solved in 2024-2027 with data availability sampling and generalized zero knowledge roll-ups.
* The inability to reach quick finality. Bitcoin provided 30 to 60 minute probabilistic finality. Ethereum provided similar probabilistic finality with much more predictability. Ethereum's PoS update gave it 15-minute economic finality. Future updates promise 12-second on-chain economic finality. Zero knowledge rollups promise instant economic finality.
* Miner extractable value (MEV). Too deep a topic to get into here, but it's a problem that plagues existing public smart contracts and that has solutions (like proposer-builder separation) that are currently in concept phase (2-5 years until wide implementation).
* History and state expiry. Basically, the ability for a blockchain to prune very old data that is no longer needed to keep consensus. 3-5 years out imo. This will allow for blockchains that don't constantly grow in size the longer they run.
This is incorrect. During 90s NNs were just one of the approaches to machine learning. It was not even clear how far ML as a field could go.
You can twist it any way you want, but AI/ML is older than cryptocurrencies.
Applying your same "modern AI is less than 10 years old", modern cryptocurrency really dates no older than Bitcoin. That took off say maybe 2010-12 or so, while the AI research was really kicking off around 2012-2014.
They're roughly the same age on both scales, Bitcoin is probably a bit older than modern AI and AI fundamentals are probably a bit older than cryptocurrency fundamentals, it sounds like. But this is arguing about 2 years +/- for the modern stuff and 5y +/- for the 70s/80s stuff.