Your Data Can Tailor The Future - Private AI

take the first step, write your own story, own your AI future

We find ourselves at a critical juncture in the rapidly evolving landscape of artificial intelligence. As reported by The Economist in July 2024, AI firms are facing an unprecedented challenge: the looming exhaustion of the internet's data for training large language models. This scarcity of novel, high-quality data threatens to slow the pace of AI advancement, potentially bringing us to the brink of an "AI winter."

“Epoch AI, a research firm estimates that by 2026 the stock of high quality textual data on the internet will all have been used. In the industry this is known as the data wall...”

So I thought, couldn’t they just talk to each other and learn more? Well, to add to the challenge, a study published in Nature (2024), ‘AI models collapse when trained on recursively generated data’, analysed the results of LLMs being trained on their own data, revealing that fine-tuning AI models with personal data may be the only way to significantly improve performance and relevance over time and other reality shifting outcomes.

“Model collapse is a degenerative process affecting generations of learned generative models, in which the data they generate end up polluting the training set of the next generation. Being trained on polluted data, they then misperceive reality.”

A new frontier in AI development is required, one that shifts focus from broad, generalised models trained on publicly available data to highly personalised, context-rich, private AI, specialised assistants.

Now, I know what you’re thinking,

‘My data’ is an elusive, mythical creature I have never fully appreciated, captured, controlled, and used in a consolidated way. It’s just a big scary mess, and it's probably mostly all signed away, anyway.

We get it, and the worst part is that this narrative is mostly true, but it doesn’t have to be. Time and place have more of an impact on the value of your data than you think. Your digital self will be valued and revalued the more it touches AI, and I think you should be rewarded for this contribution.

The use of personal data in AI training is not without controversy, especially when it is harvested and used without consent, which I’d argue is one of the more critical issues for AI to tackle. As noted in a Scientific American article:

"The companies creating these models are effectively treating the entire internet as a free-for-all data set, and 'your' data—the pictures you post, the text you write—has very likely been used to train models without your knowledge or consent." - Scientific American, July 2023

To tackle these problems, a new paradigm in AI development that respects individual privacy and data sovereignty is urgently needed.

---

If done right, privacy-preserving AI built using decentralised, open-source infrastructure could offer a solution.

A fabled two birds, one stone, 100x solution to censorship and centralisation of AI models. Breaking the data walls and, in turn, oligopolies of AI.

I’d like to contribute to the refinement of AI as well as be valued along the way.

Trending toward a dignified, loyal AGI outcome.

Take the First Step - a Private Data Bridge

A secure, user-controlled conduit that connects an individual or entity's scattered digital footprint to power truly tailored, context-rich and uniquely personalised AI experiences. This approach aims to address the privacy concerns raised by current AI training practices by putting control back in the hands of users.

Data bridges will become a commonplace tool for interacting with AI. In the same way that you bridge digital assets between chains depending on use cases, you should be able to bridge your intellectual and behavioural data and value across AI platforms and models.

It’s also becoming apparent that for AI Agents to interact with real value, it must be on-chain. So, naturally, your data should be stored privately and accessed using a blockchain-compatible wallet.

The good news is that Verida has built most of what you need to get started using private AI trained on your own data. This is the best image I found of how it all works.

Building a Personal Data Bridge to Enable Personal AI

How can I prepare to cross the bridge into Private AI territory?

To help you prepare for the path of Private AI, consider these steps:

1. Audit your digital footprint: Take stock of where your personal data resides across various platforms and services.

2. Explore data portability options: Familiarise yourself with tools and platforms that allow you to export and control your data.

3. Stay informed about privacy-preserving technologies: Monitor developments in areas like homomorphic encryption and secure multi-party computation.

4. Engage with privacy-focused platforms: Start using services prioritising user data control and sovereignty.

As we venture further into this new era of AI, it's crucial to understand both the potential benefits and the ethical considerations. While personalised AI promises enhanced efficiency and relevance, it also raises questions about data privacy, algorithmic bias, and the potential for over-reliance on AI systems. As users and developers, we must remain vigilant and proactive in shaping this technology to serve humanity's best interests.

In the next post, I’ll explore some of the benefits and mechanics of private fine-tuning and data bridging for AI and share some of the opportunities and applications you could build that leverage the advantages of private AI - starting with the private data bridge.

…but how could private AI be better than the rest? 👀