The Title Question
Is there any paper, article or book that analyzes the feasibility of achieving AGI through brain-simulation?
Yes. There are various analyses that have been published. We have some early work like Some Philosophical Problems From the Standpoint of Artificial Intelligence, John McCarthy and Patrick J. Hayes, Stanford University, 1969. And there is more recent and optimistic work like Essentials of General Intelligence: The Direct Path to Artificial General Intelligence, Peter Voss, 2007. There are refutations of the concept of General Intelligence, artificial or not, in some posts here (listed below) and quite a bit of academic scrutiny of the assumptions behind the AGI concept and the offspring of it: The Singularity. Concern about The Singularity is a social phenomenon, not a scientific one. It began with the depiction of the emergence of a super-intelligence as an inevitability in science fiction and in sensational media.
But mathematically terse proofs of AGI feasibility are not found in the media or in the Terminator franchise started by the wonderful imagination of James Cameron. This is not to say that world domination cannot fall into the hands of a new species created by humans. In fact, Jaques Ellul, in his Techological Society points out that in many ways humans are already subservient to technology — that the creation of humankind is already autonomous and dominant. His 449 page heap of evidence is witty and quite convincing.
Academic publications either
(a) Assume feasibility and discuss approaches to its design or
(b) Illuminate caveats in the idea of a universally intelligent system
Where is the Mind?
Whether the entirety of the mind arises from a physical system is a question that dates back to René Descartes and Gottfried Leibniz. Current conceptions of intelligence tend to deny the interaction and interdependence of the human brain, the rest of the human, and the biosphere into which the human fits without extreme expenditure to produce a floating, underwater, or underground simulation of the biosphere. The mind may center in the brain, but any brain disconnected from the biosphere and the circulatory system for even a short time is completely useless. A think tank is not a bunch of smart people in a sensory deprivation tank. Quite the opposite. Early experimentation with think tanks revealed the importance of mental health of the participants through diversion and physical exercise so that the various forms of neurological stasis could be maintained.
It is quite questionable whether computers without bodies will ever learn as much of general use (that can be put to use in the biosphere) without moving about and interaction with the biosphere. It is highly likely that the mind is more like Carl Jung sees it, as in the brain and elsewhere. Ludwik Fleck, in a somewhat Jungian fashion, speaks of science as the product of a collective in his Genesis and Development of a Scientific Fact, a good read that throws considerable light on the folly of most people's naive conception of facts.
Quantum Computing Buzz
When Intel's founder, Robert Noyce, brought in Gordon Moore in 1975, it was because of various statistical observations he had made, including what is now known as Moore's Law, a curve that on semi-log paper in 1965 looked linear. Since then, the business assumption in Silicon Valley has been that compactness and speed would double frequently and consistently. In the last few years, speed has stopped doubling and compactness seems to have reached its limit around 7 nm. The transistors in this VLSI technology have only a handful of atoms to represent the semiconductor topologies that permit gating, the fundamental activity of digital circuitry.
In essence, to continue the Silicon Valley boon, it has to become Particle Valley, using carbon allotropes or other nano-tech to build computers where Moore's Law is not overturned by the limitations of the structure of the universe. Whether this can be done requires its own feasibility study. Already, SSD (solid state drive) technology relies heavily on probability. Errors are occurring in great proportions and error correcting techniques are used to keep them manageable. That's because, as we reach for designs closer to the quantum level, Brownian motion or the light electromagnetic interference can upset the digital operation intended.
Challenges Versus Show Stoppers
As with many trends in the history of human endeavors, determining what is a challenge and what is a show stopper is not clear until viewing in hindsight. That is why people are asking the various questions in this Stack Exchange listed in the question.
Stating that mentioning our partial knowledge of brain operation is off topic for this question might have been appropriate if we were trying to invent something that we had not already seen in operation in nature. However, that is not the case. We call it artificial intelligence because we think that we are intelligent and want to extend that facility we find in ourselves. The following fields, in addition to quantum physics and machine learning, are thus necessarily relevant to the feasibility of a more generally usable fabricated intelligent system.
- Neuro-chemistry in general
- Genetics as it applies to neurological network structuring and cell activity
- Correlation between DNA coding and human mental performance and stability
- Applied psychology
- Addiction science
- Cognitive science
- Mathematics in general and probability and statistics specifically
- Control theory
- Philosophy, especially in regard to whether intelligence actually exists and the ideas of culturally determined knowledge and belief
- Language and linguistics
- Symbology
- Conceptions of a thought collective
Not only do we not know enough in these areas, but we know very little. There is also considerable evidence that humanity tends to forget some of the things it once knew. Take away our petroleum and we may find out just how much we forgot.
Whether any of the things we have lost from our general knowledge are prerequisites to solving the challenges of universally intelligent system design is unknown. A simple example is that many computing science graduates do not know anything about Kurt Gödel's incompletness theorems and how they led to Alan Turing's completeness theorem. Many use back-propagation in their Python code but don't have any conception that it is a corrective feedback distribution scheme.
On the flip side, we have no strong argument to draw the conclusion that there is some theoretical limitation as solid as the second law of thermodynamics, one of the apparently insurmountable hard stops built into our physical reality on and off planet. We have no proof that intelligence does not exist and is rather an anthrocentric fantasy. We have no proof that whatever limitations humans have cannot be overcome by copying the intellectual life of human beings into human creations or using intellect to create greater intellect in machines.
What we know, as smart as we think we are, is greatly outweighed by what we don't know. Anyone in real research work will agree.
Refutations of AGI Philosophy
Some of the many refutations of AGI that appear throughout academia. Some articles include these.
These are some posts here that question various assumptions prevalent in popular AI media.
