I'm imagining a world...

What would it be like to be an AI?

Let's say that AI is a program, a bundle of routines and memory that, when executed, take in the world through inputs and perform actions to that world through outputs. All programs do this, of course, but it is the level of the intentionality that makes the thing an AI.

I very much like the Dennet-inspired view that conciousness is an evolutionary trait, along with Free Will. Our free will extends as far as we can predict the result of our actions. So a rock has no Free Will because it cannot do any actions. Some bacteria can act and react -- so they have a very limited form of free will.

Let's go higher. A mouse can push a lever and get some food, and can ultimately predict this outcome. But a human can rig the experiment, making it so that the lever and the food aren't related the way that the mouse expects. So, come to think of it, having Free Will is not just the ability to predict how your own actions will affect the future world, but you must also _accurately_ predict it.

Collectively humans are increasing their free will by pooling their knowledge, and thus make each individual's power of prediction reach further into the future with more and more accuracy.

OK. So keep that in mind. Back to AI.

So, you are a bundle of code and memory, and you have input and output. Your output can affect the world around you, and your input gives you some rough idea of the current state of the world (or, in our case, the very recent state of the world. close enough).

Humans have a pretty clear boundry of where an individual body begins and ends. The body is the extent of the input and output of the biological machine. Where the actual intelligence begins and ends is quite a bit more fuzzy -- nerves and neurons are all mixed up, along with lots of chemicals that add to the computational status. Yet even with humans the intelligence portion of the body is slowly seeping out. With spoken and written language and other external information storage and retrieval mechanisms a single human can store (and share) bits of their knowledge outside of their own body. But can the _act_ (that is, output/control) outside of their own body?

In some ways yes -- through other people or other things that they control. A leader who depends on their followers is effectively doing this, though of course the control that they exert is more granular than over their own bodies. The leader augments their own predictive power with that of their minions, allowing them to input, predict, and act in ways that they couldn't do alone.

Yet the linkage between a leader and their followers is still relatively loose. Human communication has a relatively limited bandwidth with which to share the aspects of an intelligent being, especially when compared to the communication capabilities within the brain.

Lets see if an AI has similar constraints. The process in summary is to continuously: input, predict, act.

Imagine that there is an AI on my laptop. For inputs it has the keyboard, a webcam, a microphone, and wireless (the internet), along with various internal sensors for the state of the battery and the temperature of the processor. We'll pretend that my laptop has tons of processing power and memory for it to predict things. How can my laptop "act" on the world? It has a screen, speakers, and wireless (internet).

There are other computers around here, and accessible via the network, so it is reasonable that my laptop's AI could adopt a leadership role and command the other machines. This is effectively what I do all the time, having the other machines tell me things and commanding them. But my AI can do something which I definitely can't -- it can copy itself. In theory. Perhaps my AI requires much more processing power and storage than can be transferred over the network. Then it would have to act much more like me, controlling things remotely.

But machines get faster and the network gets faster, so very soon there should be enough bandwidth to transfer my AI to a whole other machine. Indeed, if there wasn't then my AI would be trapped in my laptop just as I am trapped in my body. Of course, the main question then is how did the AI get _on_ my laptop if that is the case? I would have had to write the whole thing, or grow the thing, right on my machine. Not very likely.

So an AI can copy itself to another machine. This certainly doesn't destroy the original, it merely spawns a second AI. Which one, then, would be the master and which would be the slave?

I think the one with the fastest processor and maybe the most memory would always be the master, since it would be able to out-predict the other one. Just as we do with our leaders, the slow machine would, whether it wanted to or not, give up control and become an extension of the master. But perhaps faster or not-faster doesn't quite do it. After all, with humans we get a lot of milage out of "experience". But experience is just storage, and the faster machine got that when the slower created it. But perhaps the slower machine decided to remain master by not giving the new machine all of the information necessary to be master. It could parcel out tasks to the fast machine without giving it the full picture.

And yet it would still be trapped in the slow machine. The ultimate would be to transfer conciousness to the new machine. But this raises the if-i-could-clone-myself issue. Who is the real one, and who is #2?

This is all predicated, again, on the speed of the interconnect. If it was a fast enough connection, then there could be just a single entity who happened to have twice as much processing and twice as much storage. Slow that connection down a bit and soon the two will have to spend effort to account for the delay.


---------------------------------------

What if we had invented computers before airplanes? Could we have?