Re: Identity

The Baileys (
Wed, 25 Nov 1998 04:32:27 -0500

Eliezer Yudkowsky wrote:

>Suppose you make a recording of a Turing machine, and play it back - load
>state and tape from each tick back into memory. Is that an instantiation?

I assume the playback would take place on a Turing machine as robust as the recorded subject since only a device equal in complexity could perform a faithful playback. I'm further assuming that by "instantiation" you are meaning that the process of playing back the recorded machinations of a conscious intelligence instantiates that conscious intelligence (assuming defined steps of computation are involved). If I'm getting any of this wrong, let me know and I'll revise my observations.

The example above is not instantiation of a conscious intelligence. The playback machine is not given the chance to experience its own consciousness and allow that experience to develop its mind. The playback process, assuming the recording was at the appropriate detail, would constantly overwrite every connection, memory, heuristic, etc. There would be no instance of consciousness in the playback machine. Once the playback ends, however, and if the recorded mind is not deleted from the playback machine, it could be conscious then. Going forward, starting with the final mind state of the recording, the playback machine (no longer playing a recording) would be a developing mind.

>Suppose you run the Turing machine again, except this time you compare it
>the recording and destroy it if it differs. Is that an instantiation?

This example appears to be instantiation of a conscious intelligenc, albeit one in a very miserable plight. The difference with this example is that the tolerance of difference, at least momentarily, implies the playback machine is given the chance to develop on its own but only those developments that compare with the recording are allowed to remain. A similar scenario could be the ultimate form of mind control. You are reading the manifesto of a new totalitarian group. This group happens to possess the power to monitor every change in a mind and manipulate it if it chooses. The group monitors your mind as you read its manifesto. If you react negatively or reject a concept, they change your mind accordingly to agree with the concept. You are conscious, but the trajectory of your mind over time is constantly being adjusted to direct it down a particular path.

I found the analogy of a musical artist improvising a cadenza helpful in thinking through these. An artist has been asked to improvise a cadenza on a guitar lasting one minute. The artist does so, employing all of her dexterity, musical knowledge and creativity. A recording is made of the cadenza. Additionally, the cadenza is meticulously transcribed (every note, nuance, timbre change, etc.)
Do any of the processes below exhibit creativity, i.e. instantiate creativity?

  1. Playing back the recording. Answer: Obviously no.
  2. An equally-skilled musician performs the cadenza using the meticulous transcription. Answer: No, the musician is employing musical knowledge and dexterity but not creativity. The musician could employ creativity but is not.
  3. The original musician replays the performance using the transcription. Answer: No, for the same reason as in #2.
  4. A musician is told to improvise a cadenza, if his improvisation differs from the original improvisation he is killed. Answer: Yes, the musician is employing creativity. Whether the results of his creativity matches that of the original musician is an ancillary concern. (Notably, this example does not completely match with your "test and rewrite" example above but I could not determine a neat cadenza example that matched well. However, using the recorded mind, if another mind was allowed to develop and if it differed from the recording it was destroyed, that would be a similar example. In that case, the mind would be conscious because it would be allowed to develop on its own, albeit closely monitored.)

Doug Bailey

>If these processes are AIs, which of them are conscious?
>It's questions like these, admittedly carried a bit farther, that converted
>from my former position of Strong AI to a Penrosian noncomputationalist.
> Eliezer S. Yudkowsky
>Disclaimer: Unless otherwise specified, I'm not telling you
>everything I think I know.