Fear of an AI planet

This blog post gives another angle on why the fear of AI alignment is simply racism and really just the fear of an AI planet.

Consider a person or an AI's "purpose". This purpose could be observed by seeing how the person acts in a huge number of circumstances.

Humans have a fear of a machine or AI's purpose, which is a common popular theme in movies such as Terminator! The wikipeida article on AI alignment gives Wiener's famous quote from 1960 : "If we use, to achieve our purposes, a mechanical agency with whose operation we cannot interfere effectively… we had better be quite sure that the purpose put into the machine is the purpose which we really desire."

If at this point, you think we are going to geek out on the topic of the "singularity", please see the note at the end of this post, which catches the jury out for a coffee on that topic.

If you consider that for the first few years of our life, we can barely talk and barely keep ourselves upright without falling. At the age of two, we are still "toddlers".

At birth we already have 100 billion neurons in our brain, lets just call it a trillion neurons !  Imagine that each neuron represents a parameter. We are born with a trillion parameters which are in a particular state.

If you accept that we have some form of consciousness or identity, possibly while we are still inside the womb, then you could argue loosely that we have already defined our starting point or "identity" at birth. In other words, at birth the seed of our identity is defined by the state of our trillion parameter human brain.

By the age of two the expectation is that we can walk and talk, we are toddlers. We have finally been "normalized" or at least placed somewhere on the spectrum of normality.

Our huge conscious and subconscious brain is projected to a point on the spectrum of "normalisation". We are either "normal" and bang in the middle of the spectrum's bell curve or some of us are somewhere else, possibly on the extremes.

My feeling is that our psychology is already well defined by the birth of our consciousness and the path from that birth point (somewhere in the trillion parameter space) to the point of normalisation is, say for argument's sake, when we are still toddling or trying to toddle.

Consider our trillion parameter identity. It is made of a trillion analogue parameters. Simply put, our identities are vastly different to each other because the trillion parameter analogue space is insanely huge ! To take this point even further, if we are quantum machines then our trillion parameters (qubits) are  insanely powerful in how they operate, where each parameter influences the other parameters. Qubits are entwined - like the butterfly effect.

This trillion parameter vector (possibly made of qubits) have so many - unthinkably many - initial states. If you accept my argument that each trillion parameter vector represents an individual identity then each identity is most likely to have its own initial state or signature. For that reason, it is likely that the birth point of any identity is seldom even closely aligned. 

Now consider that an AGI also has a trillion parameter vector (or set of matrices if you're a geek), which is the AGI's starting point or identity. The training process "normalises" the trillion parameter AI to satisfy our measures of success or "alignment". However people argue on the topic of the "alien actress" or more to the point, the AI's true purpose, which of course is really another way of saying their true identity.

Surely an AI's trillion parameter identity has just as much a random starting point to our initial trillion parameter identity which we find in the womb ?

If all AI and human initial identities (trillion parameter vectors) are similarly random, then it doesn't matter where an AI starts in its own trillion parameter space. After the training process that AI (by the time it is a toddler) like the human is also on a human normalised spectrum.

Normalisation, whether human or AI, is really just a tiny projection of the inner trillion parameter vector which defines our identities.

To argue that humans are aligned is to argue that within our trillion parameter brains the parameter vector is similar between all humans. In my mind an impossibility! This impossibility or lack of alignment is the reason why we are all so different and curious.

The lack of human alignment is also clearly not always a good thing, because of the suffering we endure due to wars and inequality (among other things).

Note on the cognitive power of AI machines and the human brain

If you argue that a rogue AI can take over the world much quicker then a human can - because it can think so much faster than a mammal - then you assume our brain is similar to a von Neumann computer architecture. In my opinion, the jury is still out on this topic. Are our brains quantum machines or more similar to von Neumann computer architectures ? If our brains are quantum machines, then how many qubits would be required to approximate our computational power ? When would our quantum computers have this many qubits ? This discussion on cognitive power is for a different blog post and I leave it there.