Decentralized AI: it’s more than just about who controls AI
Over the past year we’ve seen an incredible rise in all things AI. However, we believe that this exponential growth won’t be sustainable. The AI industry will not be able to deliver on all the promises regarding value creation, unless we move towards a more Decentralized AI.
In this blogpost we’ll cover:
- Why AI should not and cannot be controlled by a single entity.
- Why native payment rails for AI can be a game-changer.
- Access to non-public data is critical for AI.
- What we’re doing about it at Nevermined
(TL;DR: An AI & Data marketplace with #SmartSubscriptions) - What our community of D-AI pioneers is doing about it
(TL;DR: A hackathon in Paris, July 18–19)
Transparency
Decentralized AI is an alternative to the monopolies that are currently being established by the tech mammoths. As the whole world agrees that AI has the potential to have a fundamental impact on society, it is fair to say that we should be worried about having such a powerful technology run by a small number of people or companies. The stakes are too high and the issues around trust, data usage and copyright infringement require a level of transparency that doesn’t rhyme with a centralized approach.
Furthermore, Decentralized AI is more than ‘just’ anti-centralized AI. Rather than having centralized silos, it’s about creating ecosystems, taking into account all stakeholders. This democratizes the access to the AI industry to all Data Owners, AI Developers and will ultimately improve the advancement of AI.
AIs need to make payments
A crucial part of ecosystems are aligned incentives, rewards and economic models. Blockchain is the perfect, decentralized technology to enable exactly that. This is not so much about ‘crypto’, but about ‘cheap and fast programmable digital payments’.
AI’s won’t have bank accounts, but they will have wallets on a blockchain. The native payment rails of blockchain-based AI will be a crucial distinguishing feature in the advancement of AI agents.
Decentralizing Data Supply
The current boom in AI was made possible by scraping publicly available data. But the large AI models will run out of data to train on. At the same time, there’s a trend towards Precision AI, which requires niche models to get access to small but highly relevant data sets.
This will start a trend of decentralization of the data supply. AIs will need to get access to walled and proprietary data and compute wherever that data resides.
A crucial aspect in this is the development of federated learning techniques and the fact that privacy-enhancing technologies like zero-knowledge proofs and homomorphic encryption continue to mature.
There’s a Marketplace for that
Both decentralization and AI, are core elements of the Nevermined DNA. One part of our team has been working with blockchain and Web3 technology for the best part of 10 years. The other part of the team has decades of experience in Data, Analytics and AI.
And this is why we recently launched the Nevermined Marketplace App, an open platform for AI developers and Data Owners to monetize the access to their assets. We are using an innovative tech solution called Smart Subscriptions, a model that reinvents outdated subscriptions by combining usage rules with smart contracts, with full track and trace capabilities.
AI developers can use our tools to create Smart Subscriptions for their AI assets and sell them to app developers who want to integrate AI services into their product, but who don’t want to build an AI from scratch.
Data Owners can use our tools to tokenize their Data assets, set pricing and access control conditions, and offer them up for training purposes to AI developers.
We believe this approach will contribute to the decentralization of AI. And we’re not alone. In his recent essay on Decentralized AI, deep tech analyst Lawrence Lundy-Bryan phrased it as follows:
“More data will be available on blockchains, notably validated data, and decentralised AI networks will use cryptocurrency to buy and sell data, tools, models and anything else required to perform increasingly sophisticated ML tasks.”
However, we’re not purists. We love our Web3 tech, but we acknowledge that there are different ways to reach a similar goal. There are other ways to look at decentralization, with different tech and different economic models.
There’s a Hackathon for that
That’s why we’re co-organizing the Augment Decentralized AI Hackathon on July 18/19, on the side of EthCC in Paris.
Together with AI3, Algovera, Inflection, Orbis, Tenfold and Valory, pioneers in the decentralized AI space, we’re bringing together 100 people to explore, create and contribute to the advancement of this compelling new technology.
The objective of this event is to gather developers and researchers to build solutions that can increase the decentralization of AI, beyond the hype of ChatGPT or crypto.
The event will run over 2 days and there are 5 themes that you can work on:
- Authenticity: how can we know what’s real?
- Privacy vs Personalization: how can we use Federated Learning with local data?
- Governance: how do we enable stakeholders to self-govern, and find alignment towards the common good?
- Value Capture: How can those who contribute data or code to AI platforms share in the value those platforms generate? (we love this one, obviously!
- Democratize Compute & Training: How can we run training and inference of very large models in a decentralized way?
So if this triggers your AI and/or your decentralization buttons, sign up! There are still some seats left to make sure we D-AI and don’t die.
