Hello. This is a concept that I have not seen formulated before and I think the people here might find interesting. I am a new LW user and so I do not know if this has been formulated before. In this case I suppose you can refer me to it and close this post or something like that.

This "cryptographic life" I will describe now should be seen as a plausible example of how a lifeform/consciousness could span a volume measured in hundreds of lightyears without being fixed to any particular physical objects.

Here are my premises:

P1: We assume that AGI is of course possible and can be implemented on classical computers. It is also possible to implement these AGIs as distributed/decentralized computations. You could, for example, have an AI which runs different aspects of its computation on hundreds of different physical computers, constantly sending data between them to keep its overall state of mind coherent. It can dynamically move its computation onto new hardware and integrate new sensory data sources.

P2: We assuming as given a galaxy filled with advanced civilizations that communicate at lightspeed and have access to large amounts of classical, digital compute power. THeir communications technology is sufficiently advanced to send and receive high bandwidths over hundreds of lightyears.

P3: Full Homomorphic encryption is possible. According to Wikipedia: "Homomorphic encryption is a form of encryption that permits users to perform computations on its encrypted data without first decrypting it." As far as I know we only know of such methods that allow for limited arithmetic operations to be performed "under the hood".

We assume that "full homomorphic encryption" is possible in a "public-private" key manner. This means that a customer can give you a description of any program (any turing machine) which has been homomorphically encrypted with a private key, so that you can take an input, encrypt it with the public key, and the let the program run on this input by performing certain fixed computation steps. Then you can return the output to the customer and the customer can decrypt it, getting his answer. Homomorphic encryption of this kind allows cloud computation to become secure: You can give your program to other people and you know that they will have to correctly execute it. If someone tries to modify the program during its execution then the result will not be correctly encrypted anymore, meaning that the owner of the private key can tell if it has been tampered with or not. We assume that the keys are long enough so that they are realistically secure.

Now let us describe what a cryptographic lifeform is.

A cryptographic lifeform is an AI which runs purely on distributed cloud computation performed by possible unreliable actors. A civilization living on a planet may receive over the course of months or years signals that contain a complete description of an aspect of the cryptographic lifeform. This is an encrypted description of various programs. It is well-known throughout the galaxy how such descriptions are supposed to be interpreted and hence they may start running the AI on their local hardware. This means that they provide sensory data to the AI, encrypted with the public key that was contained in the signal, and the AI gives various encrypted or unencrypted outputs.

Why would anyone do this? The people on the planet may have the following motivation: They have a lot of computation power but perhaps actually creating an AI is very difficult. The cryptographic lifeform is a way for them to use the capabilities of a possibly superintelligent AI without actually having to develop this technology themselves. They only need to do the computation. Therefore they maybe by happy to provide the AI with computation power with the deal that some of it is used to compute solutions to problems they face, and the rest is free for the AI to use for its own purposes. This requires giving reliable sensor data about the real world to the AI. The people on the planet can not cheat the AI by modifiying the computation because then the encryption will be violated. Note that the only entity which knows the private key is the cryptographic lifeform itself!

After the AI has performed its part of the deal, as part of the payment it is allowed some badnwidth to transmit data to the stars. It now has the opportunity to send fragments of its updated states to other aspects of its consciousness spread throughout the galaxy. At any time parts of the AI may be running isolatedly on hundreds of locations through the galaxy. By constantly sending messages between those aspects of itself it can keep a coherent state of mind and sense of self. Of course, it wouldn't have as strong as a identity as we do, but perhaps it would be somewhere between a single person and a family of people that work together.

Note that the cryptographic lifeform is protected from betrayal: If a civilization feeds it wrong sensory data or betrays the deal in some other way, then it can retaliate. It may have access to large amount of ressources, currency and connections, so it may pay a nearby civilization to send a kinematic kill projectile.

If a civilization performs the computation wrongly then the local aspect of the AI or instances on other planets may detect that the signals are not correctly encrypted anymore and therefore discard them as wrong. This will not get integrated into the state of mind of the cryptographic lifeform.

Well, that is the whole idea :)

New to LessWrong?

New Comment