On Conscious Machines
A thought experiment
In this article, I describe a sequence of machines; each more “conscious” than the previous.
Disclaimer: Since we cannot know if something else is “conscious” my definition and measurement of consciousness is indirect.
Level 1 — Recording
A Level machine has sensors that record the external world: Cameras, Microphones, Touch Sensors, Smell Sensors and Taste Sensors.
We inspect what the machine records, to verify that the machine is sensing correctly. For example, we analyse the videos that it records, to see if those record the actual world.
Level 2 — Reacting
A Level 2 machine can react to what it records. The reactions might be specific or arbitrary.
For example, a level 2 machine “might not like” bring light. If we flash a torch at it, it might emit an annoyed sound, or move away from the light source.
How do we know that it is actually “annoyed”? Or that it moved away because it didn’t like the light?
Well, we don’t. Our definition of “annoyed” is emitting the annoying sound or moving away. That is how we measure “not like”.
Level 3 — Abstractions
A Level 3 machine, not only records the physical world but also converts these recordings into abstractions. For example, it might classify the video frames it records, into having specific dominant colours; like red or blue.
While our machine reacts like a Level 2 machine, its triggers are abstractions. It might “not like” red light, but like other forms of light.
Level 4 — Connections
A Level 4 machine is a Level 3 machine, with the ability to make connections between abstractions. It might connect “seeing red” with “feeling liquid” and “tasting metallic”. It might define new abstractions around these connections. For example, it might call the red, metallic tasting liquid “blood”.
Like Level 2 and Level 3 machines, the Level 4 machine would also react, not just to raw information, or abstractions, but also connections. For example, it might be “afraid of blood”.
Level 5 — Time
A Level 5 machines not only records the physical world, but it also records “snapshots” of these recordings.
For example, it could store a copy of “yesterday’s world”.
The frequency of the snapshots might differ from machine to machine, and defines “how fast time passes” for these machines.
