What follows is my personal and totally speculative attempt to draw up a classification of super (i.e. post human) intelligent agents.
There is something paradoxical, almost empathetic, in an entity that tries to define beings that are intellectually superior to itself.
The reason is clear: it is not possible, by definition. If we could define them in detail, we would be them.
An ant cannot describe the processes that run along the grayish nerve connections of a human brain and what [thoughts / insights / innovations] they generate. In fact, not even we, at our current scientific standpoint, know all about the mechanisms that keep us spinning.
I find this point particularly interesting because it suggests an important fact: the gears that make us work, be they mechanical or algorithmic in nature, are more advanced than the effects they seem to produce. By relying on those systems (the only ones we really have available) we are not yet able to fully understand the systems themselves.
I am not claiming that this problem of [recursive/reflective] self-knowledgeis insurmountable, only that the problem is hard.Think about it for a moment: if an ant were able to understand everything about its brain, its biological imperatives and so on, wouldn’t you say that it possesses an impressive [self-awareness/intelligence] to say the least?
I’m certainly not the first to have noticed the matter: many researchers consider the time coordinates in which an artificial intelligence arrives to obtain a total knowledge of its internal state as a very important point of its maturity. Possibly the most important of all.
The reason for this is of a logical nature: once this peculiar goal has been reached, the thinking subject can begin to “hack herself”, improving both his hardware and his software and giving rise to a chain reaction with unpredictable and exponential results: once this happens, singularity is only a few iterations away.
All this always assuming that it is possible: as said before, the total transcendence of our limits may not be so simple, we could even find ourselves in front of a paradox.
Having briefly discussed these questions, I now want to embark into an exercise in style: that of enunciating a possible framework to classify agents relying on their post-human intelligence. Such a task, as already outlined, is doomed to fail miserably at producing accurate, empirical predictions but could still lead to some meaningful insights.
Note: some of the definitions expressed below are not my own inventions but are borrowed from researchers and philosophers of the field.
First of all, it is necessary to define quantitative parameters on the basis of which comparisons and categorizations can be performed.
We will try to limit the parameters to a minimum, defining only the fundamental primitives. It is probable (but not sure) that a quantitative increase in intelligence also corresponds to a qualitative increase in conscious experience (qualia) but from this point of view our scientific knowledge is inadequate to say the least. However, this will not stop us from pointing out possible subjective details of the agent’s experience if considered noteworthy.
Parameters: Speed ==> speed with which intelligence operates (operations per second).
Environment ==> The environment in which the agent operates (spatial and temporal characteristics).
Propositions space graph ==> depth and connectivity of the graph on propositions space.
By “propositions space” I mean the space of all possible concepts. This parameter represents
the degree to which an intelligence can link different but dependent notions to each other.
It is important to note that the increase in these quantities corresponds to an increase in intelligence; essentially these quantities are gradients that point to a growing intellect.
We now draw up the classification of entities based on the degree of intelligence
Dumb AI: Artificial intelligence limited to its environment and to a specific task. This type of agent is smarter than a man to the extent that it is powered by a hardware (copper circuits, optical fibers) and a software (neural network) that is more performing than the human brain but only in completing a specific duty (images recognition, sounds recognition, patterns, statistical regressions, market forecasts ecc.) but it is extremely lacking in multi (or extra) domain intelligence. Their advantage over men comes mainly from a higher speed.
These types of agents already exist in our world.
Fictional Intelligence: This is actually not a strict set but rather could be seen as a macro category defining what Yudkowsky calls “Mind Design Space” . As I understand it, this category represents the totality of possible intelligences that we have not yet encountered in our world, if not in the form of fiction.
Here you can find the super-intelligent shade of blue of “Hitchhiker’s Guide to the Galaxy” alongside C-3PO and Wall-e. Basically all the intelligent agents of human fiction. I invite you to note that all the following categories, in a sense, are a subset of the present due to the fact that we have never met any of the following agents and therefore, at present, they are nothing but “fictions”.
Em: The term stands for “Emulated Brains” and is taken from Robin Hanson’s book “The Age of Em”. In essence, it involves scanning human brains using medical techniques that are not yet fully available (for example vitrification) and subsequent uploading to a virtual world.
The advantages of this approach, if it proved stable, would be to have agents who already possess a “full spectrum” human intelligence and who could take advantage of the hardware they run on: much more powerful than mere flesh. These entities could take a second of the “external” time to think what an average human would think in 50 years of life.
The advantages of this approach, if it proved stable, would be to have agents who already have “full spectrum” human intelligence and who could take advantage of the hardware they run on: much moBased on our parameters.
Based on our parameters, the initial innovation would be an improvement of the environment, which would lead to an increase in speed and ultimately to an expansion of the proposition graph. Furthermore, once ems become pure information, they could make copies of themselves at no cost and maybe even perfect their own code (if the transcendence we talked about in the intro were to be verified).re powerful than mere flesh.
Singleton: This term was coined by the philosopher Nick Bostrom to indicate a world order in which there is a single decision-making agency that moves the strings.
In Bostrom’s own words:
“Among its powers would be (1) the ability to prevent any threats (internal or external) to its own existence and supremacy, and (2) the ability to exert effective control over major features of its domain (including taxation and territorial allocation).
Singletons could be good, bad, or neutral. One reason for favoring the development of a singleton (of a good type) is that it would solve certain fundamental coordination problems that may be unsolvable in a world that contains a large number of independent agencies at the top level.”
It does not necessarily have to be a synthetic intelligence, it could very well be people who come together by sharing the same values in unison, to create the republic that Plato spoke of (for example).
Singleton is a term that also includes possible hive minds on a planetary scale.
Also in this case we are witnessing first of all an expansion of the environmental parameter.
Ultra Intelligence: An ultra intelligence is an intelligence that extends on a galactic scale or more, in general an agent that turns its attentions towards the cosmos.
It can begin with the discovery of interstellar travel technology by a species which then expands and colonizes multiple solar systems. If this species later constituted a singleton, its domain would already be extremely vast and given the technological means available, nothing would prevent it from further expanding.
With advanced techniques and an immense amount of matter at its disposal, the UI can manufacture as much computronium as it wants, thus modifying the speed and environment parameters and subsequently expanding the web of connections on propositions space.
At this point, as intelligence continues to acquire momentum, physical limitations are the only burdens that are presented to it. the expansion can theoretically continue until the expansion of the universe beyond the cosmic horizon leaves it isolated in its domain and in its self contemplation.
Omega Intelligence: This is the most advanced form of intelligence imaginable.
It operates in an infinite and timeless environment. Since there is no time in its environment, it doesn’t even make sense to talk about processing speed as each operation is atemporal. This agent does not have memory limits as the space it can access is infinite. Its intelligence includes all the others within it, in the same way that the infinite power set of complex numbers contains within it all the combinations of minor power sets.
There is no limit to the computational problems it can solve, it is however subject to the limits of formal systems, in the sense that it cannot solve, for example, Turing’s halting problem.
It is instantly aware of any state of consciousness that may exist, its existence is a continuum of qualia, an eternal kaleidoscopic ocean of experiences.
In fact this being is so omnipotent that it does not even make sense to consider it as an intelligent agent.
It is the absolute, it is what comes closest to the god of monotheistic religions.
It is the concept that includes all the others, It is the space of propositions itself.