As a part of my teaching of AI, I have to think a lot about where AI itself is going.
I take an agnostic view of technology
Much of the media gets too carried away both on successes and failures.
One such technology is blockchain
Blockchain has a chequered history
I am not optimistic about blockchain because, in my view, its a technology looking for a use case that can be uniquely addressed by it.
Crypto was one such use case … but crypto itself has its own issues.
NFT was the other.
NFT(Non-fungible token) is supposed to represent a piece of digital content uniquely as part of the Ethereum blockchain, which keeps track of ownership and trade.
In theory, an NFT could be tied to any digital content that supposedly has some value (even a tweet).
This application of NFT (hence blockchain) too has been impacted by hype (Bored ape, Nyan cat, and of course, Metaverse NFT real estate).
It’s in this context that I think that there is an area that could be interesting for blockchain.
Firstly, an observation that in the digital world, aggregation models have taken off a lot.
For example, streaming music is behind the boom in song back catalog deals.
We may actually see something analogous in the large language models.
If you are not familiar with large language models, you should read this from Stanford: how large language models will transform science society and AI.
Large language models essentially scrape the net for training.
But how ethical is this to the creators?
As such, each image in the training set contributes only a tiny amount of information—a few tweaks to some numeric weights spread across the entire network.
But… the people who created these images did not give their consent. And the model can be seen as a direct threat to their livelihoods. No-one expected creative AIs to come for the artist jobs first, but here we are!
leading to a very interesting tweet comparing AI to veganism in the sense of people making their own personal ethical decisions.
This may well work – it does not talk of lots of money for one piece of content (like bored ape) – but rather like the song back catalog models.
And companies who train large language models can afford to pay – every time they train (and indeed retrain!) a model – they would need to pay the original creators of the content used to train the model.
Image source https://medium.com/@sbmeunier/when-do-you-need-blockchain-decision-models-a5c40e7c9ba1
