AI Personal Memory persistence, the future of the browser and the barbershop
Privacy future-proof architectures: will the browser be the host and orchestrator of personal AI agents and data stores?
Following last weeks video on on device LLMs, today wanted to share some further thinking on personal Ai and memory storage.
As my dear friend Francesco reminds me with the story of the barber shop, trust and memory are two essential parts of great service in retail. Last time he called to book an appointment, the barber kindly suggested his son to come with him, adding a humorous comment on how his son’s 3 month long locks. No agenda or digital bookkeeping was required, only an experienced barber that remembers his clients with empathy and kindness.
This is precisely what AI agents are trying to do next, memory persistence that will make us feel as if they already know us. However, in the case of many GP AIs, such as ChatGPT agents the trust element is missing. How much of my personal data are they storing? for what purposes? how can i delete it?
I find fascinating how many AIs are starting to separate personal data storage, critical for training, memory and alignment, from the foundational models. This is one of today’s biggest privacy-by-design AI architectural challenges.
On this topic, I enjoyed the views that Brendan Eich, founder of Mozilla and Brave, shared last week on the future of the browser at Consensus 2024.
The browser as a local, on-device AI driven conversational experience that remembers your preferences and acts both as as a host and an orchestrator of multiple personal AI agents. PODs as the privacy-first extension of Local Storage. Data and AI private and local.
Would love to hear from you guys. Are the privacy-by-design architectures hybrid too? Combining the best of central 1st party CDP repositories with decentralised on-device (browser or native app) PODs + personal agents?