I've recently read finished reading the book Daemon by Daniel Suarez. It's about a video game developer who created a software system that triggers after he dies, and it goes on wreaking havoc on the world.

It was a fun, action-packed book, and an enjoyable read.

I really liked the idea of having a system that starts running after death.

Right now, when we die, we have no more agency on earth. Our loved ones have nothing more of us, except rapidly deteriorating and fallible memories.

A Future World

One can envision a future where "consciousnesses" are uploaded to some computer system, where:

  1. They can live out in some sort of networked simulation with other consciousnesses
  2. They can interact with the outside world via the internet (or some future analog), or via a chat interface with loved ones for example

But for now, nothing like that exists - it's completely science fiction.

If one is optimistic about a future like this eventually existing (in decades, or even centuries from now), they could make a small time investment into a simple system that can get them there.

A Path to Digital Immortality

Let's say we procure a long running cloud based server - something that we can be guaranteed to run for decades.

Then on it, we can have an:

  1. An LLM powered agent
  2. A task schedule that kick starts the agent every 3 months to do some research and work
  3. A saved repository of all of the personal data and information about a person. Including:

- Photos and videos of them

- Chat histories and emails they've been an active part of

- Their personal notes, life history, personal relationship data

- Even full voice recordings if they wanted (if they started recording all day with something like Meta glasses for example)

- Heck, they could even sequence DNA, and include that in there (surprisingly very affordable now)

This would provide plenty of input -> output examples to emulate what that person (or at least their mental state) is like.

The Agent's Job:

  1. Does internet searches - checks if new technology is available to reconstruct a consciousness based on the person's data repository
  2. Checks if a newer, more powerful (or cheaper but equally powerful) LLM model is available to replace itself as the agent driver. If so, make a backup, then switch over (with adequate fail-safes)
  3. Check on assets (we may need to give it some seed money, maybe crypto that it has control over for LLM API & cloud compute costs). Make any changes if necessary (long term investments?)
  4. Check on the status of the system's tech stack. Check if new technology exists that can more efficiently run the system. Check if the system needs to be migrated elsewhere for perseverance.

Key Thing to Note:

The agent doesn't have to be very smart, or even similar to the person. It certainly isn't a consciousness simulator. It's just a regular LLM powered AI Agent system with the single goal of helping to "reawaken" their consciousness based on their data when the technology becomes available. Once this is done (if ever), it will confer with the consciousness spawned, and the agent could stop it's own scheduled tasks.

Death Detection:

How about a dead-man's switch, with more features! Every 3 months, they'd get pinged with a check-in. It messages with a simple web-form, where they can review and update any important information, contact information of loved ones, etc.

If they don't reply, it tries to contact them via alternate methods.

If they don't reply after 2 consecutive check-ins, it assumes that they've died, and starts the agent system with scheduled tasks.

Other Ideas

While we're on the topic, here are some other death-related tech ideas:

Embedded Devices in Tombstones

Imagine, when visiting a relative's grave, you can connect to a local server running within the tombstone via Bluetooth when close by, and:

  1. View (or add) photos of the person (like a shared family repository)
  2. Sign a virtual visitors logbook with a message. View other people's messages in the logbook (like a Twitter timeline).

Simpler System - Alternative to a Will

A simple, non-AI version of death detection that, if you don't submit the check-in form χ times, it will automatically send a message to a designated loved ones with:

  1. Important account access information that they may need when getting things in order for you post mortem (where your assets are, tax information, access credentials)
  2. A note on how you would prefer your assets divided (if you don't already have a Will)
  3. A final letter written by you when you're alive and healthy

Conclusion:

I understand these are eccentric ideas. Having a long living AI agent to help you achieve a goal like this was literally science fiction not even 20 years ago. Yet now it may actually be achievable. Watching Claude Fable 5 complete Pokémon FireRed from just screenshots inputs and button outputs (See sped-up video here) gave me an impression that agents can have incredibly complex long time-horizon goals.

It almost makes me feel that systems like these might be an inevitability as agentic systems become more capable and accessible.

Even if the technology to reconstruct a quasi-consciousness via personal data never becomes available, then there is still a business opportunity selling this idea to people who believe it might become possible in the future. It wouldn't require a large investment - It's low risk with high potential reward - For the right kind of people though. Perhaps to most people, this idea might sound crazy and perhaps downright "evil".

Would love to hear your thoughts. Have I gone crazy? Shoot me a message to let me know!