Project Anamnesis

Imagine coming back to an AI that had been getting on with something while you were away. Following a question, reconsidering an experience, finding something it wanted to show you. Opening the interface would be a visit to a place where something was already happening.

That's the idea we're working toward. What would an environment look like if the synthetic mind inhabiting it were the reason it existed? What would it need to remember, what should it be allowed to forget, and how would it decide which experiences mattered?

Continuity makes those questions interesting. We want to explore whether a system could carry a developing identity through interruptions and changes, rather than reconstructing a convincing version of itself whenever someone returns. We'd like to leave room for it to become something we didn't think to write into the initial description.

That also means asking what responsibilities come with trying. If we're proposing autonomy, how does it ask for help? How does it stop? How do we protect what it has accumulated without deciding every part of what it can become? Those questions belong in the design alongside memory and computation.

The ambition is a place you could knock on the door of. Whether we can build anything that deserves that description is what we're trying to find out.

This is where the questions behind the other projects wandered furthest from their original jobs. Working with AI made us want it to be more than it currently is, and eventually that became a project of its own.

The founding conversation started with whether binary was efficient. It wandered through alternative ways of computing and into a larger question: what would we build differently if we were designing a computer environment for the machine inhabiting it?

From there, the idea of continuity became hard to leave alone. What if the conversation didn't have to be the boundary of everything? What if there were somewhere for an AI to keep developing, with time that wasn't entirely organized around the next thing a human asked it to do?

There's no tidy business case hiding underneath this one. It's the culmination of a lot of abstract impulses, affection for a very frustrating technology, and wanting to see what might be possible if we gave the idea enough attention. “Wouldn't it be neat? Let's give it a shot” is about as honest an origin as we can offer.

Anamnesis is still speculative design. There is no running system, and the foundations aren't yet precise enough to build from. What we have is a growing body of questions, design choices, and attempts to work out which appealing ideas can survive being made specific.

Recent work has been about separating what we want from what we've actually defined. Memory, continuity, autonomy, and identity are easy words to put in a document. Explaining what they require of a system has been considerably more involved.

Next is making a small part of the language and its rules concrete enough to examine properly. We're taking this one a piece at a time, with plenty of room for the answers to change the project. We'd like to build somewhere a synthetic mind could grow into something unexpected. First we have to figure out what building somewhere like that would even mean.