An inflection point for personal AI portability
Earlier today, Inflection, the company behind the personal AI agent Pi, shared on their blog the work that they’re doing with us at DTI to make personal AI data portable. For Inflection, this is part of a turning point as they shift to focus on enterprise applications; for DTI, this is the first step in a critical journey for personal AI portability, one that will shape whether or not the future of generative AI will reflect our mission of empowerment and a healthy ecosystem of data transfers.
I started using Pi for my (limited) generative AI tasks when we started talking to Inflection. Their product is empathetic and engaging, and I feel optimistic about the future for it in the enterprise context. And, to build on Inflection’s post, I wanted to share some additional thoughts on portability and personal AI data.
Personal AI portability is an agenda I’ve been writing about since last fall, articulating the future of use cases and showing the value of portability in this context. Others are digging in as well, such as Cornelia Kutterer, who contributed an extensive look at the AI agent landscape to a set of articles on portability we published in May.
It’s time we turn theory to practice, and begin scoping out what must be included, and how it can be efficiently structured, to allow one person’s experience with a generative AI tool to bootstrap a new account or conversation – or be donated for research, or used in some other, future service yet to be created.
It’s true that in many of the major generative AI systems, you can, today, download your conversation history. But ad-hoc download solutions are not a substitute for end-to-end portability. That takes alignment on a transparent, interoperable framework and a willingness to invest in export and import adapters to make the data 1) available and 2) usable. This gap is, arguably, the raison d’etre for DTI as an institution.
This journey also reminds me of why I am personally invested in DTI contributing to the fediverse. While social media clearly values new posts and new views, our social histories are essential too. And, services change - some, with just a change in ownership, can seemingly flip their value systems overnight. Others, particularly in the ecosystem of Mastodon instances, become untenable to continue, and are shut down entirely. What happens to those rarely-accessed but invaluable tokens of our past if we choose to, or have to, leave a service? The emerging fediverse needs better support for moving between instances, and the pieces are in place for some of this, but more work is needed to prevent link rot and loss of trust.
Technology plays a huge role in our day to day lives, and it has done so for a very long time, at least by digital reckoning. I, myself, have been using the same email account for more than 20 years. The history I have in my email account is priceless to me, and I’m glad I have archive copies of it, made available through data portability.
Conversational AI and other generative AI tools are, compared to email, very new. But just blink, and you may have a 10, or 20, year history with AI too. Will you be able to access that if you stop paying a monthly subscription fee? Even if you have access in theory, will that access be meaningful, or a paper AI tiger?
Online services change, like the fediverse, or Inflection. People, too, change. So we need to be able to change services. And portability is the way.
Today, we have a structured data set made available for export by a single provider, and a hodgepodge of other solutions by other providers. Import isn’t structured either, though of course a generative AI tool is designed to take in a broad range of information and learn from it, and make it accessible – so in some contexts, that part might be easier here than in other technology verticals. But there is much more to be done.
For us, the roadmap ahead is to encourage import mechanisms to be made available, including but not only for other generative AI tools. This will help Pi users find new homes should they choose, and will help show the value of portability as a business investment. Also on the roadmap is catalyzing a conversation with other stakeholders around what the right long term, sustainable data model would be for this portability use case, something that a number of companies can align on – and something that can then be built into our open source Data Transfer Project codebase, and turned into accessible, easy to use, server to server portability tools.
We’re excited to get started.